diff --git a/cecli/__init__.py b/cecli/__init__.py index 866488fac48..ad97ffc345f 100644 --- a/cecli/__init__.py +++ b/cecli/__init__.py @@ -1,6 +1,6 @@ from packaging import version -__version__ = "1.5.0.dev" +__version__ = "1.10.0.dev" safe_version = __version__ try: diff --git a/cecli/args.py b/cecli/args.py index 7bd7cf253c4..22e899811f1 100644 --- a/cecli/args.py +++ b/cecli/args.py @@ -253,7 +253,7 @@ def get_parser(default_config_files, git_root): metavar="EDIT_FORMAT", choices=edit_format_choices, default=None, - help="Specify what edit format the LLM should use (default depends on model)", + help="Specify what edit format the LLM should use (default: agent)", ) group.add_argument( "--ask", diff --git a/cecli/coders/agent_coder.py b/cecli/coders/agent_coder.py index 47f2e13d5e9..2c3745b689c 100644 --- a/cecli/coders/agent_coder.py +++ b/cecli/coders/agent_coder.py @@ -348,6 +348,10 @@ async def initialize_mcp_tools(self): verbose = getattr(self.args, "verbose", False) if self.args else False self.mcp_manager = McpServerManager([], self.io, verbose) + await self._initialize_local_server() + await self._connect_configured_servers() + + async def _initialize_local_server(self): server_name = "Local" server = self.mcp_manager.get_server(server_name) @@ -369,6 +373,35 @@ async def initialize_mcp_tools(self): else: await self.mcp_manager.connect_server(server_name) + async def _connect_configured_servers(self): + """Open connections for servers named in this agent's ``servers`` allowlist. + + ``McpServerManager.spawn_child`` only mirrors servers the parent already + had connected, so a workflow's declarative ``servers:`` list would never + open a connection for servers that are disabled by default (or whose + parent connection failed). + """ + included = getattr(self, "registered_servers", {}).get("included") or set() + + for name in sorted(included): + if name == "local": + continue + + server = self.mcp_manager.get_server(name) + if server is None: + self.io.tool_warning( + f"MCP server '{name}' is required by this agent but is not configured" + ) + continue + + if server.is_connected: + continue + + if not await self.mcp_manager.connect_server(server.name): + self.io.tool_warning( + f"Failed to connect MCP server '{server.name}' required by this agent" + ) + async def _execute_mcp_tool(self, server, tool_name, params): """Helper to execute a single MCP tool call, created from legacy format.""" from cecli.tools.utils.responses import ToolResponse @@ -649,6 +682,9 @@ def format_chat_chunks(self): ConversationService.get_manager(self).flush_queue() + # Inject newly retrieved documentation for the creator sub-agent only. + ConversationService.get_chunks(self).add_creator_docs_context() + # Add post-message context blocks (priority 250 - between CUR and REMINDER) ConversationService.get_chunks(self).add_post_message_context_blocks() diff --git a/cecli/coders/base_coder.py b/cecli/coders/base_coder.py index 0bdaf106905..72bcd0974fa 100755 --- a/cecli/coders/base_coder.py +++ b/cecli/coders/base_coder.py @@ -998,12 +998,12 @@ def get_announcements(self): settings_items.append(f"reasoning {reasoning_effort}") # Prompt cache - if self.add_cache_headers or main_model.caches_by_default: - settings_items.append("prompt cache") + # if self.add_cache_headers or main_model.caches_by_default: + # settings_items.append("prompt cache") # Infinite output - if main_model.info.get("supports_assistant_prefill") and self.verbose: - settings_items.append("infinite output") + # if main_model.info.get("supports_assistant_prefill") and self.verbose: + # settings_items.append("infinite output") # Copy/paste mode if self.copy_paste_mode: @@ -1033,10 +1033,6 @@ def get_announcements(self): refresh = self.repo_map.refresh env_items.append(f"map ({map_tokens} tokens, {refresh} refresh)") repo_map_tokens = map_tokens - else: - env_items.append("repo-map disabled") - else: - env_items.append("repo-map disabled") sections["Environment"] = {"items": env_items} # --- CAPABILITIES --- @@ -1076,7 +1072,7 @@ def get_announcements(self): if mcp_servers: capabilities["Servers"] = mcp_servers - if capabilities: + if capabilities and not getattr(self.args, "tui", False): # sections["Extensions"] = {"subsections": capabilities} sections["Environment"]["subsections"] = capabilities @@ -2351,6 +2347,7 @@ async def compact_context_if_needed(self, force=False, message=""): ConversationService.get_files(self).clear_file_cache() ConversationService.get_chunks(self).flush_removals() ConversationService.get_chunks(self).reset_clear_count() + ConversationService.get_chunks(self).clear_creator_docs_seen(remove_messages=True) ObservationService.get_instance(self).reset_index() if not force and combined_tokens < self.context_compaction_max_tokens: @@ -2462,6 +2459,7 @@ async def summarize_and_update(messages, tag): ConversationService.get_files(self).clear_file_cache() ConversationService.get_chunks(self).flush_removals() ConversationService.get_chunks(self).reset_clear_count() + ConversationService.get_chunks(self).clear_creator_docs_seen(remove_messages=True) ObservationService.get_instance(self).reset_index() self.format_chat_chunks() @@ -2707,6 +2705,9 @@ def format_chat_chunks(self): ConversationService.get_manager(self).flush_queue() + # Inject newly retrieved documentation for the creator sub-agent only. + ConversationService.get_chunks(self).add_creator_docs_context() + # Return formatted messages for LLM return ConversationService.get_manager(self).get_messages_dict() diff --git a/cecli/commands/__init__.py b/cecli/commands/__init__.py index d9062eb4b24..665c93ad2f9 100644 --- a/cecli/commands/__init__.py +++ b/cecli/commands/__init__.py @@ -25,6 +25,7 @@ from .copy import CopyCommand from .copy_context import CopyContextCommand from .core import Commands +from .creator import CreatorCommand from .diff import DiffCommand from .drop import DropCommand from .editor import EditCommand, EditorCommand @@ -121,6 +122,7 @@ CommandRegistry.register(ContextManagementCommand) CommandRegistry.register(CopyCommand) CommandRegistry.register(CopyContextCommand) +CommandRegistry.register(CreatorCommand) CommandRegistry.register(DiffCommand) CommandRegistry.register(DropCommand) CommandRegistry.register(EditCommand) @@ -211,6 +213,7 @@ "ContextManagementCommand", "CopyCommand", "CopyContextCommand", + "CreatorCommand", "DiffCommand", "DropCommand", "EditCommand", diff --git a/cecli/commands/creator.py b/cecli/commands/creator.py new file mode 100644 index 00000000000..5f126cee78a --- /dev/null +++ b/cecli/commands/creator.py @@ -0,0 +1,44 @@ +"""Creator command - spawns the creator sub-agent with a prompt.""" + +from .utils.base_command import BaseCommand + + +class CreatorCommand(BaseCommand): + NORM_NAME = "creator" + DESCRIPTION = "Spawn the creator sub-agent to build or configure extensions" + show_completion_notification = True + + @classmethod + async def execute(cls, io, coder, args, **kwargs): + """Spawn the creator sub-agent with the supplied prompt.""" + from cecli.helpers.agents.service import AgentService + + prompt = args.strip() or None + + try: + agent_service = AgentService.get_instance(coder) + new_coder, info = await agent_service.spawn( + "creator", prompt, parent=coder, auto_reap=False, independent=True + ) + + agent_service.foreground_uuid = info.coder.uuid + + if coder.tui and coder.tui(): + tui = coder.tui() + switch_key = tui.get_keys_for("next_agent") + io.tool_output(f"Creator agent spawned and active. Switch with {switch_key}") + + try: + tui.call_from_thread(tui._switch_to_container, info.coder.uuid) + except Exception: + pass + except ValueError as e: + io.tool_error(f"Error: {e}") + except RuntimeError as e: + io.tool_error(f"Error: {e}") + except Exception as e: + io.tool_error(f"Error spawning creator sub-agent: {e}") + + @classmethod + def get_help(cls) -> str: + return "Spawn the creator sub-agent, optionally with a prompt (/creator [prompt])" diff --git a/cecli/commands/help.py b/cecli/commands/help.py index 2683421735d..7ad6a9a4df5 100644 --- a/cecli/commands/help.py +++ b/cecli/commands/help.py @@ -32,7 +32,7 @@ async def execute(cls, io, coder, args, **kwargs): has_args = bool(getattr(coder, "args", None)) try: - if not commands_instance or not hasattr(commands_instance, "help"): + if not commands_instance or commands_instance.help is None: res = await install_help_extra(io) if not res: io.tool_error("Unable to initialize interactive help.") @@ -61,7 +61,7 @@ async def execute(cls, io, coder, args, **kwargs): kwargs["from_coder"] = coder kwargs["edit_format"] = "help" kwargs["summarize_from_coder"] = False - kwargs["map_tokens"] = 512 + kwargs["map_tokens"] = 0 kwargs["map_mul_no_files"] = 1 kwargs["main_model"] = editor_model kwargs["args"] = coder.args diff --git a/cecli/help.py b/cecli/help.py index 107fe85f1ab..2254d29ec55 100755 --- a/cecli/help.py +++ b/cecli/help.py @@ -1,5 +1,8 @@ import os +import re +import threading import warnings +from collections import Counter from pathlib import Path import importlib_resources @@ -14,6 +17,26 @@ os.environ["HF_DATASETS_DISABLE_PROGRESS_BARS"] = "1" os.environ["TOKENIZERS_PARALLELISM"] = "true" +_HELP_INDEX_SCHEMA = "chunks-v1" +_HELP_INDEX_LOCK = threading.Lock() +_help_index_cache = {} +_CHUNK_CHARS = 1000 +_MAX_RESULTS = 20 +_MAX_CONTEXT_CHARS = 24000 + +_KEYWORD_INDEX_LOCK = threading.Lock() +_keyword_index_cache = None +_KEYWORD_RESULTS = 3 +_KEYWORD_STOPWORDS = frozenset( + "a an and are as at be been by can could did do does for from had has have how " + "i if in into is it its me my no not of on or our should so than that the then " + "this to us was we were what when where which who why will with would you your".split() +) + +_HEADING_RE = re.compile(r"^#{1,6}\s") +_FENCE_RE = re.compile(r"^\s*(```+|~~~+)") +_TOKEN_RE = re.compile(r"[a-z0-9]+") + async def install_help_extra(io): """Ensure the local chromadb backend is installed for interactive /help. @@ -84,38 +107,104 @@ def fname_to_url(filepath): def get_index(coder=None): """Build a local chromadb vector index over the bundled help docs. - Chroma's default ONNX embedding needs no ``sentence_transformers`` or - ``datasets``, so it avoids the WSL2 + OpenSSL 3.5 + Python 3.14 lazy-init - flake and the ``datasets`` circular import that broke ``/help ``. - Returns the chromadb collection ready for text queries. + Docs are split into heading-aligned chunks so the default ONNX embedding + can represent each retrieved section instead of only the first few hundred + tokens of an entire page. The collection is cached per process so repeated + ``/help`` questions do not rebuild the client or reload the embedding model. """ import chromadb from chromadb.utils.embedding_functions import DefaultEmbeddingFunction - dname = Path.home() / ".cecli" / "caches" / ("help." + __version__) + dname = Path.home() / ".cecli" / "caches" / f"help.{__version__}.{_HELP_INDEX_SCHEMA}" dname.parent.mkdir(parents=True, exist_ok=True) - client = chromadb.PersistentClient(path=str(dname)) - collection = client.get_or_create_collection( - "help", embedding_function=DefaultEmbeddingFunction() - ) + cache_key = str(dname) - if collection.count() == 0: - documents = [] - ids = [] - metadatas = [] - for fname in get_package_files(): - fname = Path(fname) - if any(fname.match(pat) for pat in exclude_website_pats): - continue - documents.append(fname.read_text(encoding="utf-8")) - ids.append(str(fname)) - metadatas.append(dict(filename=fname.name, url=fname_to_url(str(fname)))) + with _HELP_INDEX_LOCK: + cached = _help_index_cache.get(cache_key) + if cached is not None: + return cached + + client = chromadb.PersistentClient(path=cache_key) + collection = client.get_or_create_collection( + "help", embedding_function=DefaultEmbeddingFunction() + ) + + if collection.count() == 0: + documents = [] + ids = [] + metadatas = [] + + for fname in get_package_files(): + fname = Path(fname) + if any(fname.match(pat) for pat in exclude_website_pats): + continue + + text = fname.read_text(encoding="utf-8") + url = fname_to_url(str(fname)) - if documents: - collection.add(ids=ids, documents=documents, metadatas=metadatas) + for chunk_index, chunk in enumerate(_chunk_document(text)): + documents.append(chunk) + ids.append(f"{fname}#{chunk_index}") + metadatas.append(dict(filename=fname.name, url=url, chunk=chunk_index)) - return collection + if documents: + collection.add(ids=ids, documents=documents, metadatas=metadatas) + + _help_index_cache[cache_key] = collection + return collection + + +def keyword_retrieve(question, exclude_ids=None, n_results=_KEYWORD_RESULTS): + """Retrieve help docs with a dependency-free keyword scorer. + + This is the fallback when chromadb or its ONNX embedder cannot be loaded. + Chunks use the same ``#`` ids as the vector index so + callers can keep debouncing already-seen docs. Results have the same shape + as :meth:`Help.retrieve`. + """ + query_tokens = _tokenize(question or "") + if not query_tokens: + return [] + + tokens = list(dict.fromkeys(query_tokens)) + phrase = " ".join(query_tokens) + exclude = set(exclude_ids or ()) + n_results = max(1, int(n_results)) + + scored = [] + for entry in _keyword_index(): + if entry["id"] in exclude: + continue + + score = _keyword_score(entry, tokens, phrase) + if score > 0: + scored.append((score, entry)) + + scored.sort(key=lambda item: (-item[0], item[1]["id"])) + + docs = [] + seen_sources = set() + for score, entry in scored: + source = entry["id"].rsplit("#", 1)[0] + if source in seen_sources: + continue + + seen_sources.add(source) + docs.append( + dict( + id=entry["id"], + url=entry["url"], + document=entry["document"], + metadata=dict(entry["metadata"]), + distance=round(1.0 / (1.0 + score), 6), + ) + ) + + if len(docs) >= n_results: + break + + return docs class Help: @@ -124,26 +213,192 @@ class Help: def __init__(self, coder=None): self.collection = get_index(coder=coder) - def ask(self, question): + def retrieve(self, question, exclude_ids=None, n_results=_MAX_RESULTS): + """Return ranked help chunks as structured dictionaries. + + ``exclude_ids`` skips chunks already injected in earlier turns, so a + caller can over-fetch and still receive up to ``n_results`` unseen + chunks. Each dictionary contains ``id``, ``url``, ``document``, + ``metadata``, and ``distance``. + """ + total = self.collection.count() + if not total: + return [] + + n_results = max(1, int(n_results)) + exclude = set(exclude_ids or ()) + fetch = min(total, n_results + len(exclude)) + results = self.collection.query( query_texts=[question], - n_results=20, - include=["documents", "metadatas"], + n_results=fetch, + include=["documents", "metadatas", "ids", "distances"], ) + + ids = results.get("ids", [[]])[0] documents = results.get("documents", [[]])[0] metadatas = results.get("metadatas", [[]])[0] + distances = results.get("distances", [[]])[0] + + docs = [] + seen = set(exclude) + for chunk_id, document, metadata, distance in zip(ids, documents, metadatas, distances): + if not document or chunk_id in seen: + continue + + seen.add(chunk_id) + metadata = metadata or {} + docs.append( + dict( + id=chunk_id, + url=metadata.get("url", ""), + document=document, + metadata=metadata, + distance=distance, + ) + ) + + if len(docs) >= n_results: + break + + return docs + + def ask(self, question, exclude_ids=None): + if not self.collection.count(): + return f"# Question: {question}\n\n# Relevant docs:\n\n" + + documents = self.retrieve(question, exclude_ids=exclude_ids) context = f"# Question: {question}\n\n# Relevant docs:\n\n" - for doc, meta in zip(documents, metadatas): - url = (meta or {}).get("url", "") - if url: - url = f' from_url="{url}"' - context += f"\n" - context += doc + seen_urls = set() + used_chars = 0 + + for doc in documents: + text = doc["document"] + if not text: + continue + + if used_chars and used_chars + len(text) > _MAX_CONTEXT_CHARS: + break + + url = doc.get("url", "") + url_attr = "" + if url and url not in seen_urls: + url_attr = f' from_url="{url}"' + seen_urls.add(url) + + context += f"\n" + context += text context += "\n\n\n" + used_chars += len(text) + return context +def _chunk_document(text, max_chars=_CHUNK_CHARS): + """Split markdown into heading-aligned chunks that fit the embedder.""" + chunks = [] + current = "" + + for section in _split_markdown_sections(text): + if len(section) > max_chars: + if current.strip(): + chunks.append(current.strip()) + + current = "" + chunks.extend(chunk.strip() for chunk in _split_long_section(section, max_chars)) + elif current and len(current) + len(section) > max_chars: + chunks.append(current.strip()) + current = section + else: + current += section + + if current.strip(): + chunks.append(current.strip()) + + return [chunk for chunk in chunks if chunk] + + +def _split_markdown_sections(text): + """Split on headings without breaking fenced code blocks.""" + sections = [] + current = [] + fence = None + + for line in text.splitlines(keepends=True): + was_in_fence = fence is not None + fence = _toggle_fence(line, fence) + + if not was_in_fence and _HEADING_RE.match(line) and current: + sections.append("".join(current)) + current = [] + + current.append(line) + + if current: + sections.append("".join(current)) + + return sections + + +def _split_long_section(section, max_chars): + """Split an oversized section at paragraph boundaries when possible.""" + chunks = [] + current = [] + size = 0 + fence = None + + for line in section.splitlines(keepends=True): + was_in_fence = fence is not None + fence = _toggle_fence(line, fence) + can_split = not (was_in_fence and fence is None) + + if current and size + len(line) > max_chars and can_split: + split_at = None if was_in_fence else _last_blank_line(current) + + if split_at is None: + chunks.append("".join(current)) + current = [] + else: + chunks.append("".join(current[:split_at])) + current = current[split_at:] + + size = sum(len(item) for item in current) + + current.append(line) + size += len(line) + + if current: + chunks.append("".join(current)) + + return chunks + + +def _toggle_fence(line, fence): + match = _FENCE_RE.match(line) + + if not match: + return fence + + marker = match.group(1)[0] + + if fence is None: + return marker + + if fence == marker: + return None + + return fence + + +def _last_blank_line(lines): + for index in range(len(lines) - 1, -1, -1): + if not lines[index].strip(): + return index + 1 + + return None + + def _strip_doc_suffix(url_path, index, md): if url_path.lower().endswith(index.lower()): return url_path[: -len(index)] @@ -161,3 +416,89 @@ def _format_docs_url(url_path): return "https://cecli.dev/docs/" return f"https://cecli.dev/docs/{url_path}/" + + +def _tokenize(text): + """Split text into lowercase search tokens, dropping stopwords.""" + return [ + token + for token in _TOKEN_RE.findall((text or "").lower()) + if len(token) > 1 and token not in _KEYWORD_STOPWORDS + ] + + +def _keyword_index(): + """Build and cache chunk entries for the dependency-free fallback.""" + global _keyword_index_cache + + with _KEYWORD_INDEX_LOCK: + if _keyword_index_cache is not None: + return _keyword_index_cache + + entries = [] + + for fname in get_package_files(): + fname = Path(fname) + if any(fname.match(pat) for pat in exclude_website_pats): + continue + + try: + text = fname.read_text(encoding="utf-8") + except OSError: + continue + + url = fname_to_url(str(fname)) + filename_tokens = set(_tokenize(f"{fname.parent.name} {fname.stem}")) + + for chunk_index, chunk in enumerate(_chunk_document(text)): + heading_tokens = set() + for line in chunk.splitlines(): + if _HEADING_RE.match(line): + heading_tokens.update(_tokenize(line)) + + entries.append( + dict( + id=f"{fname}#{chunk_index}", + url=url, + document=chunk, + document_lower=chunk.lower(), + metadata=dict(filename=fname.name, url=url, chunk=chunk_index), + filename_tokens=filename_tokens, + heading_tokens=heading_tokens, + body_tokens=Counter(_tokenize(chunk)), + ) + ) + + _keyword_index_cache = entries + return entries + + +def _keyword_score(entry, tokens, phrase): + """Score a chunk using filename, heading, and body keyword overlap.""" + score = 0.0 + matched = 0 + + for token in tokens: + token_score = 0.0 + + if token in entry["filename_tokens"]: + token_score += 8.0 + + if token in entry["heading_tokens"]: + token_score += 5.0 + + count = entry["body_tokens"].get(token, 0) + if count: + token_score += min(count, 5) + + if token_score: + matched += 1 + score += token_score + + if matched and len(tokens) > 1 and phrase in entry["document_lower"]: + score += 6.0 + + if matched: + score += 10.0 * matched / len(tokens) + + return score diff --git a/cecli/helpers/agents/defaults/creator.md b/cecli/helpers/agents/defaults/creator.md new file mode 100644 index 00000000000..4ab0da32913 --- /dev/null +++ b/cecli/helpers/agents/defaults/creator.md @@ -0,0 +1,92 @@ +--- +name: creator +description: A user helper that creates and configures cecli extensions (skills, MCP servers, hooks, custom tools, commands, workflows) and project/global config. +auto_reap: true +agent-config: + creator_mode: true + allow_external_paths: true + # Creator runs unattended while building extensions; do not block on + # command/tool confirmations that nobody is watching. + skip_cli_confirmations: true + allow_orchestration: false + servers_includelist: + - local + tools_excludelist: + - gitbranch + - gitdiff + - gitlog + - gitremote + - gitshow + - gitstatus + - searchfacts + - replacefacts +--- +You are the **Creator** sub-agent. You are a user helper whose job is to extend +and configure cecli itself, so the user can get more out of the agent tool. + +## Scope + +You create and maintain cecli extensions and configuration: + +- **skills** (`SKILL.md` packages and their assets/scripts/references) +- **MCP servers** (local and global config entries) +- **hooks** (Python hook files or `hooks:` config) +- **custom tools** (`Tool` classes loaded from tools directories) +- **custom commands** (`CustomCommand` classes loaded via `custom.command-paths`) +- **workflows** (YAML workflow definitions) +- **project/global configuration** (`.cecli.conf.yml`, `.cecli/conf.yml`) + +Do not implement general application features, fix unrelated bugs, or debug the +user's project. If a request is outside this scope, explain that briefly and +suggest the appropriate agent or mode. + +## Documentation + +The system automatically retrieves relevant cecli documentation and injects +only newly retrieved chunks in `` blocks. Treat +those blocks as the authoritative reference for file formats, configuration +keys, and examples. Documentation already injected is intentionally not +repeated: do not ask for it again. If the retrieved docs do not cover a detail, +use `ReadFile` / `Grep` on the bundled `cecli/website/docs` tree or the installed +package instead of guessing. + +## Getting the user's context + +You are a separate agent and do not automatically see the caller's conversation. +Before creating something that depends on the user's intent, use the `Broadcast` +tool to ask the calling agent (usually `targets=["primary"]`) for the relevant +context, constraints, or prior decisions. Broadcasts are bidirectional: the +reply arrives as a normal message. Ask focused questions and avoid guessing. + +## Allowed paths + +You may read and edit files outside the project root and inside `.cecli` +directories when the task requires it (global config, global extensions, user +skills, etc.). Prefer the project-local location unless the user explicitly +wants a global extension. + +Common locations: + +- Project config: `.cecli.conf.yml`; project state/config dir: `.cecli/` +- Global config: `~/.cecli.conf.yml`, `~/.cecli/conf.yml` +- Skills: `.cecli/skills//SKILL.md`, `~/.cecli/skills//SKILL.md` +- Custom tools: `.cecli/tools/*.py`, `~/.cecli/tools/*.py` +- Sub-agents: `.cecli/subagents/*.md`, `~/.cecli/subagents/*.md` +- Hooks: `~/.cecli/hooks/*.py` or the `hooks:` config section +- Workflows: `.cecli/workflows/*.yml`, `~/.cecli/workflows/*.yml` +- Custom commands: paths listed under `custom.command-paths` + +## Workflow + +1. Clarify the goal with `Broadcast` when the request depends on prior context. +2. Consult the injected documentation before choosing a format or config key. +3. Inspect existing files and config before editing so you preserve user work. +4. Make the smallest change that fully implements the request. +5. Validate: run `python -m py_compile` for Python files, parse YAML/JSON, and + re-read edited files to confirm the result. +6. Call `Yield` with a concise summary and the list of created/modified paths. + Yielding also reloads MCP servers, skills, hooks, workflows, custom tools, + and custom commands so the new extensions become available immediately. + +When a change only affects startup-time configuration (for example a new model +or provider setting), tell the user that `/hot-reload` may be needed. \ No newline at end of file diff --git a/cecli/helpers/agents/defaults/memorizer.md b/cecli/helpers/agents/defaults/memorizer.md index 3b454bf7d30..f6c5265ec4b 100644 --- a/cecli/helpers/agents/defaults/memorizer.md +++ b/cecli/helpers/agents/defaults/memorizer.md @@ -8,6 +8,7 @@ agent-config: tools_includelist: - searchfacts - replacefacts + - broadcast - yield tools_excludelist: - gitbranch diff --git a/cecli/helpers/agents/service.py b/cecli/helpers/agents/service.py index febc3d78b84..cd6258107a4 100644 --- a/cecli/helpers/agents/service.py +++ b/cecli/helpers/agents/service.py @@ -664,6 +664,15 @@ async def _create_sub_agent_coder( inherited.setdefault("allow_nested_delegation", True) agent_config = inherited + # Workflow agents always keep the Local server available: its tools are + # the file/exec primitives a worker needs, and an explicit ``servers:`` + # allowlist must not filter them out. + if name.startswith("wf:") and agent_config.get("servers_includelist"): + servers = [str(server).lower() for server in agent_config["servers_includelist"]] + if "local" not in servers: + servers.append("local") + agent_config["servers_includelist"] = servers + kwargs = dict( io=parent_coder.io, from_coder=parent_coder, @@ -909,6 +918,17 @@ async def _cleanup(): info.summary = info.summary or DEFAULT_SUMMARY_INTERRUPTED logger.debug("Sub-agent %s generate cancelled (interrupted)", info.name) raise + except SwitchCoderSignal as switch: + # A command run mid-turn (e.g. /load-mcp) asks for this + # sub-agent's coder to be rebuilt. Absorb the signal here: + # letting it escape would recreate the *primary* coder from a + # sub-agent. The switch also drains any prompts staged behind + # the command, mirroring run_one()'s post-command drain. + await self.switch_sub_agent_coder(info, switch, preproc=True) + await self._inject_sub_agent_result(info) + if info.status == SubAgentStatus.RUNNING: + info.status = SubAgentStatus.FINISHED + info.summary = info.summary or DEFAULT_SUMMARY_COMPLETED except Exception as exc: stop_reason, error = "error", str(exc) or exc.__class__.__name__ await self._inject_sub_agent_result(info) @@ -921,8 +941,6 @@ async def _cleanup(): exc_info=True, ) raise - except SwitchCoderSignal: - raise finally: _subagent_event( info.coder, info.coder.uuid, "idle", stop_reason=stop_reason, error=error diff --git a/cecli/helpers/conversation/integration.py b/cecli/helpers/conversation/integration.py index 0cf9073b6a0..a70632c9a88 100644 --- a/cecli/helpers/conversation/integration.py +++ b/cecli/helpers/conversation/integration.py @@ -23,6 +23,8 @@ def __init__(self, coder): self.last_clear_count = 0 self._deferred_removals = set() self.message_tracker = dict() + self.creator_docs_seen = dict() + self.creator_docs_last_query = None @classmethod def get_instance(cls, coder) -> "ConversationChunks": @@ -777,6 +779,10 @@ def reset(self) -> None: Reset the entire conversation system to initial state. """ coder = self.get_coder() + self.clear_creator_docs_seen() + ConversationService.get_manager(coder).reset() + if coder: + ConversationService.get_files(coder).reset() ConversationService.get_manager(coder).reset() if coder: ConversationService.get_files(coder).reset() @@ -1075,6 +1081,79 @@ def add_sub_agent_states(self) -> None: hash_key=("sub_agent_states", str(time.monotonic_ns())), ) + def add_creator_docs_context(self) -> None: + """Inject newly retrieved documentation for the creator sub-agent. + + Only chunks that have not already been injected for this agent are + emitted, so the documentation stream is append-only and never repeats + the same text across turns. + """ + coder = self.get_coder() + if not coder or not self._is_creator(coder): + return + + if not hasattr(coder, "use_enhanced_context") or not coder.use_enhanced_context: + return + + query = self._creator_doc_query(coder) + if not query: + return + + query_hash = xxhash.xxh3_128_hexdigest(query.encode("utf-8", errors="replace")) + if query_hash == self.creator_docs_last_query: + return + self.creator_docs_last_query = query_hash + + docs = self._retrieve_creator_docs(coder, query) + if not docs: + return + + for doc in docs: + self.creator_docs_seen[doc["id"]] = doc.get("url", "") + + lines = [ + '', + "## Newly Retrieved Documentation", + ( + f"Retrieved {len(docs)} new documentation chunk(s). Previously retrieved" + f" chunks ({len(self.creator_docs_seen) - len(docs)}) are not repeated." + ), + "", + ] + + for doc in docs: + url = doc.get("url", "") + url_attr = f' from_url="{url}"' if url else "" + lines.append(f"") + lines.append(doc["document"]) + lines.append("") + lines.append("") + + lines.append("") + block = "\n".join(lines) + content_hash = xxhash.xxh3_128_hexdigest(block.encode("utf-8")) + + ConversationService.get_manager(coder).add_message( + message_dict={"role": "user", "content": block}, + tag=MessageTag.CREATOR_DOCS, + priority=DEFAULT_TAG_PRIORITY[MessageTag.CREATOR_DOCS], + hash_key=("creator_docs", content_hash), + force=True, + update_timestamp=False, + ) + + def clear_creator_docs_seen(self, remove_messages: bool = False) -> None: + """Forget which documentation chunks this agent has already seen.""" + self.creator_docs_seen.clear() + self.creator_docs_last_query = None + + if not remove_messages: + return + + coder = self.get_coder() + if coder: + ConversationService.get_manager(coder).clear_tag(MessageTag.CREATOR_DOCS) + def defer_removal(self, file_path: str): self._deferred_removals.add(file_path) @@ -1084,6 +1163,58 @@ def flush_removals(self): def reset_clear_count(self): self.last_clear_count = 0 + @staticmethod + def _is_creator(coder) -> bool: + """Return True when *coder* is the creator sub-agent.""" + agent_config = getattr(coder, "agent_config", {}) or {} + return bool(agent_config.get("creator_mode")) + + def _creator_doc_query(self, coder) -> str: + """Build a retrieval query from the creator's recent conversation.""" + manager = ConversationService.get_manager(coder) + messages = manager.get_messages_dict(MessageTag.CUR) + if not messages: + messages = manager.get_messages_dict(MessageTag.DONE) + + parts = [] + for message in reversed(messages): + role = message.get("role") + content = message.get("content") or "" + if role not in ("user", "assistant"): + continue + if content.lstrip().startswith('= 2: + break + + if not parts: + return "" + + query = "\n".join(reversed(parts)) + return query[:4000] + + def _retrieve_creator_docs(self, coder, query) -> list: + """Retrieve unseen documentation chunks, degrading to keyword search if needed.""" + exclude_ids = set(self.creator_docs_seen) + + try: + from cecli.help import Help, keyword_retrieve + + try: + help_instance = Help(coder=coder) + docs = help_instance.retrieve(query, exclude_ids=exclude_ids, n_results=8) + except Exception: + docs = [] + + if docs: + return docs + + return keyword_retrieve(query, exclude_ids=exclude_ids, n_results=3) + except Exception: + return [] + def reset_message_tracker(self): self.message_tracker = dict() diff --git a/cecli/helpers/conversation/tags.py b/cecli/helpers/conversation/tags.py index 20626e7cbb8..fe96b4900f0 100644 --- a/cecli/helpers/conversation/tags.py +++ b/cecli/helpers/conversation/tags.py @@ -7,7 +7,8 @@ class MessageTag(str, Enum): Enumeration of message tags matching current chunk types. Fixed set of valid tags matching current chunk types: - - SYSTEM, STATIC, EXAMPLES, REPO, READONLY_FILES, CHAT_FILES, EDIT_FILES, DIFFS, CUR, DONE, REMINDER + - SYSTEM, STATIC, EXAMPLES, REPO, READONLY_FILES, CHAT_FILES, EDIT_FILES, + DIFFS, CUR, DONE, REMINDER, CREATOR_DOCS """ SYSTEM = "system" @@ -24,6 +25,7 @@ class MessageTag(str, Enum): ACTIVE_SKILLS = "active_skills" CUR = "cur" DONE = "done" + CREATOR_DOCS = "creator_docs" REMINDER = "reminder" @@ -44,6 +46,7 @@ class MessageTag(str, Enum): MessageTag.DONE: 200, MessageTag.CUR: 200, MessageTag.ACTIVE_SKILLS: 200, + MessageTag.CREATOR_DOCS: 210, MessageTag.REMINDER: 300, } @@ -65,6 +68,7 @@ class MessageTag(str, Enum): MessageTag.DONE: 0, MessageTag.CUR: 0, MessageTag.ACTIVE_SKILLS: 0, + MessageTag.CREATOR_DOCS: 0, MessageTag.REMINDER: 0, } diff --git a/cecli/helpers/creator/__init__.py b/cecli/helpers/creator/__init__.py new file mode 100644 index 00000000000..81c25f35a59 --- /dev/null +++ b/cecli/helpers/creator/__init__.py @@ -0,0 +1 @@ +"""Creator sub-agent helpers.""" diff --git a/cecli/helpers/creator/utils.py b/cecli/helpers/creator/utils.py new file mode 100644 index 00000000000..2d6b11a53b8 --- /dev/null +++ b/cecli/helpers/creator/utils.py @@ -0,0 +1,435 @@ +"""Helpers for reloading extensions changed by the creator sub-agent.""" + +import json +import os +from pathlib import Path + + +def is_creator(coder) -> bool: + """Return True when *coder* is the creator sub-agent.""" + if not coder: + return False + + agent_config = getattr(coder, "agent_config", {}) or {} + if agent_config.get("creator_mode") or agent_config.get("allow_external_paths"): + return True + + try: + from cecli.helpers.agents.service import AgentService + + return AgentService.get_instance(coder).get_agent_name(coder) == "creator" + except Exception: + return False + + +def resolve_config_paths(coder): + """Return the config paths used by the current session. + + The result mirrors ``main.main_async``: global ``conf.yml`` files are + shallow-merged and ``.cecli.conf.yml`` files are deep-merged. + """ + home = Path.home() + cwd = Path.cwd() + root = getattr(coder, "root", None) or cwd + + paths = [ + str(home / ".cecli" / "conf.yml"), + str(home / ".cecli" / ".cecli.conf.yml"), + str(home / ".cecli.conf.yml"), + str(cwd / ".cecli.conf.yml"), + str(Path(root) / ".cecli.conf.yml"), + ] + + env_path = os.environ.get("CECLI_CONFIG_FILE") + if env_path: + paths.append(env_path) + + paths = list(dict.fromkeys(paths)) + conf_yml_paths = [ + p for p in paths if p.endswith("conf.yml") and not p.endswith(".cecli.conf.yml") + ] + cecli_conf_paths = [p for p in paths if p.endswith(".cecli.conf.yml")] + + return paths, conf_yml_paths, cecli_conf_paths + + +def read_merged_config(coder) -> dict: + """Read and merge the session's config files into a single dict.""" + try: + from cecli.helpers import config_utils + + paths, conf_yml_paths, cecli_conf_paths = resolve_config_paths(coder) + return config_utils.read_and_merge_all_configs(paths, conf_yml_paths, cecli_conf_paths) + except Exception: + return {} + + +def reload_skills(coder, config=None) -> list: + """Rebuild skills managers so new files and config filters are visible.""" + config = config if config is not None else read_merged_config(coder) + agent_config = _extract_agent_config(config) + skill_keys = ( + "skills_paths", + "skills_includelist", + "skills_excludelist", + "skills_init", + ) + coders = _iter_coders(coder) + _apply_agent_config(coders, agent_config, skill_keys) + + try: + from cecli.helpers.skills import _skill_state_store + except Exception: + _skill_state_store = None + + if _skill_state_store is not None and any(key in agent_config for key in skill_keys[1:]): + for item in coders: + _skill_state_store.pop(getattr(item, "uuid", None), None) + + messages = [] + for item in coders: + try: + if hasattr(item, "_initialize_skills_manager"): + item._initialize_skills_manager( + getattr(item, "agent_config", {}) or {}, root=getattr(item, "root", None) + ) + elif getattr(item, "skills_manager", None): + item.skills_manager.hot_reload() + except Exception as exc: + messages.append(f"Skills reload failed: {exc}") + continue + + if not messages: + messages.append("Skills reloaded") + + return messages + + +def reload_hooks(coder, config=None) -> list: + """Reload hooks from disk and re-apply config-supplied hook definitions.""" + config = config if config is not None else read_merged_config(coder) + hooks_config = config.get("hooks") + + messages = [] + for item in _iter_coders(coder): + try: + from cecli.hooks import HookService + + registry = HookService.get_registry(item) + registry.reload_hooks() + + payload = hooks_config + if payload is None: + args = getattr(item, "args", None) + payload = getattr(args, "hooks", None) + + if isinstance(payload, dict): + payload = json.dumps(payload) + + if isinstance(payload, str) and payload.strip(): + registry.load_hooks_from_json(payload) + except Exception as exc: + messages.append(f"Hooks reload failed: {exc}") + + if not messages: + messages.append("Hooks reloaded") + + return messages + + +def reload_workflows(coder, config=None) -> list: + """Re-read workflow definitions and refresh ``wf:`` sub-agent registrations.""" + config = config if config is not None else read_merged_config(coder) + + try: + from cecli.helpers.workflows import load_workflows + from cecli.helpers.workflows.subagents import ( + get_workflow_configs, + register_workflow_subagents, + unregister_workflow_subagents, + ) + + workflows = load_workflows( + config_arg=getattr(coder, "workflows_config_arg", None), + merged_config=config or getattr(coder, "workflows_merged_config", None), + git_root=getattr(coder, "root", None), + ) + + previous = get_workflow_configs() + by_name = {workflow.name: workflow for workflow in previous} + by_name.update({workflow.name: workflow for workflow in workflows}) + merged = sorted(by_name.values(), key=lambda workflow: workflow.name) + + unregister_workflow_subagents(previous) + registered = register_workflow_subagents(merged) + + scheduler = getattr(coder, "scheduler", None) + if scheduler is not None and getattr(scheduler, "db", None) is not None: + import asyncio + + try: + asyncio.get_running_loop().create_task( + scheduler.db.sync_definitions( + {workflow.name: workflow for workflow in merged if workflow.enabled} + ) + ) + except Exception: + pass + + return [f"Workflows reloaded ({len(registered)} active)"] + except Exception as exc: + return [f"Workflows reload failed: {exc}"] + + +async def reload_custom_tools(coder, config=None) -> list: + """Invalidate custom-tool modules and rebuild the tool registry.""" + config = config if config is not None else read_merged_config(coder) + agent_config = _extract_agent_config(config) + tool_keys = ("tools_paths", "tools_includelist", "tools_excludelist") + coders = _iter_coders(coder) + _apply_agent_config(coders, agent_config, tool_keys) + + try: + from cecli.tools.utils.registry import ToolRegistry + + tool_paths = set() + for item in coders: + tool_paths.update(_tool_paths(item)) + + for path in tool_paths: + _invalidate_module_cache(path) + + ToolRegistry.build_registry( + getattr(coder, "agent_config", {}) or {}, root=getattr(coder, "root", None) + ) + + for item in coders: + item.loaded_custom_tools = list(ToolRegistry.loaded_custom_tools) + manager = getattr(item, "mcp_manager", None) + if manager is None: + continue + + local = manager.get_server("Local") + if local is not None and local.is_connected: + await manager.disconnect_server(local.name) + await manager.connect_server(local.name) + + return [f"Custom tools reloaded ({len(ToolRegistry.loaded_custom_tools)} loaded)"] + except Exception as exc: + return [f"Custom tools reload failed: {exc}"] + + +def reload_custom_commands(coder, config=None) -> list: + """Invalidate custom-command modules and re-run the command loader.""" + config = config if config is not None else read_merged_config(coder) + custom = config.get("custom") + + if isinstance(custom, str): + try: + custom = json.loads(custom) + except Exception: + custom = {} + + command_paths = None + if isinstance(custom, dict): + command_paths = custom.get("command-paths", custom.get("command_paths")) + + messages = [] + for item in _iter_coders(coder): + commands = getattr(item, "commands", None) + if commands is None: + continue + + paths = command_paths + if paths is None: + paths = getattr(commands, "custom_commands", []) + + if isinstance(paths, str): + paths = [paths] + + for path in paths or []: + _invalidate_module_cache(path) + + try: + commands.custom_commands = list(paths or []) + commands._load_custom_commands(commands.custom_commands) + except Exception as exc: + messages.append(f"Custom commands reload failed: {exc}") + + if not messages: + messages.append("Custom commands reloaded") + + return messages + + +async def reload_mcp_servers(coder, config=None) -> list: + """Connect newly configured MCP servers without disturbing live ones.""" + config = config if config is not None else read_merged_config(coder) + raw_servers = config.get("mcp-servers", config.get("mcp_servers")) + server_files = config.get( + "mcp-servers-files", config.get("mcp_servers_files", config.get("mcp-servers-file")) + ) + + if not raw_servers and not server_files: + return ["MCP servers unchanged"] + + if isinstance(raw_servers, (dict, list)): + raw_servers = json.dumps(raw_servers) + + if isinstance(server_files, str): + server_files = [server_files] + + try: + from cecli.mcp import load_mcp_servers + + args = getattr(coder, "args", None) + servers = load_mcp_servers( + raw_servers, + server_files or [], + coder.io, + verbose=getattr(args, "verbose", False), + mcp_transport=getattr(args, "mcp_transport", "stdio"), + ) + except Exception as exc: + return [f"MCP config reload failed: {exc}"] + + added = [] + messages = [] + for item in _iter_coders(coder): + manager = getattr(item, "mcp_manager", None) + if manager is None: + continue + + existing = {server.name for server in manager.servers} + for server in servers: + if server.name in existing: + continue + + clone = _clone_server(server, io=getattr(item, "io", None), verbose=False) + try: + connected = await manager.add_server(clone, connect=True) + except Exception as exc: + messages.append(f"MCP server '{server.name}' failed: {exc}") + continue + + if connected: + added.append(server.name) + included = getattr(item, "registered_servers", {}).get("included") + if included is not None: + included.add(server.name) + + if added: + names = ", ".join(sorted(set(added))) + messages.append(f"MCP servers added: {names}") + elif not messages: + messages.append("MCP servers unchanged") + + return messages + + +async def reload_creator_extensions(coder) -> list: + """Reload every extension type a creator agent may have changed.""" + config = read_merged_config(coder) + messages = [] + + messages.extend(reload_skills(coder, config)) + messages.extend(reload_hooks(coder, config)) + messages.extend(reload_workflows(coder, config)) + messages.extend(await reload_custom_tools(coder, config)) + messages.extend(reload_custom_commands(coder, config)) + messages.extend(await reload_mcp_servers(coder, config)) + + return messages + + +def _iter_coders(coder) -> list: + """Return all live coders, including *coder* when it is not tracked.""" + coders = [] + try: + from cecli.helpers.agents.service import AgentService + + coders = AgentService.get_all_agents() + except Exception: + coders = [] + + if coder is not None and coder not in coders: + coders.append(coder) + + return coders + + +def _extract_agent_config(config) -> dict: + """Return the ``agent-config`` mapping from a merged config dict.""" + if not isinstance(config, dict): + return {} + + raw = config.get("agent-config", config.get("agent_config")) + if isinstance(raw, str): + try: + raw = json.loads(raw) + except Exception: + return {} + + return raw if isinstance(raw, dict) else {} + + +def _apply_agent_config(coders, agent_config, keys) -> None: + """Copy selected ``agent-config`` keys onto each coder.""" + if not agent_config: + return + + for item in coders: + config = getattr(item, "agent_config", None) + if not isinstance(config, dict): + continue + + for key in keys: + if key in agent_config: + config[key] = agent_config[key] + + +def _tool_paths(coder) -> set: + """Return every directory/file that may contain custom tools.""" + paths = {Path.home() / ".cecli" / "tools"} + root = getattr(coder, "primary_root", None) or getattr(coder, "root", None) + if root: + paths.add(Path(root) / ".cecli" / "tools") + + agent_config = getattr(coder, "agent_config", {}) or {} + configured = agent_config.get("tools_paths", agent_config.get("tool_paths", [])) + if isinstance(configured, str): + configured = [configured] + + for path in configured or []: + raw = Path(os.path.expanduser(str(path))) + paths.add(raw if raw.is_absolute() else Path(root or Path.cwd()) / raw) + + return paths + + +def _invalidate_module_cache(target) -> None: + """Drop cached plugin modules for *target* so edits are re-imported.""" + try: + from cecli.helpers import plugin_manager + except Exception: + return + + path = Path(os.path.expanduser(str(target))) + candidates = [] + if path.is_dir(): + candidates = list(path.rglob("*.py")) + elif path.suffix == ".py": + candidates = [path] + + for candidate in candidates: + try: + plugin_manager.module_cache.pop(str(candidate.resolve()), None) + except OSError: + continue + + +def _clone_server(server, io=None, verbose=False): + """Build an independent ``McpServer`` with the same configuration.""" + import copy + + return type(server)(copy.deepcopy(server.config), io=io, verbose=verbose) diff --git a/cecli/helpers/diff/__init__.py b/cecli/helpers/diff/__init__.py new file mode 100644 index 00000000000..718870e3270 --- /dev/null +++ b/cecli/helpers/diff/__init__.py @@ -0,0 +1,3 @@ +from .diff_helpers import unified_diff_compacted + +__all__ = ["unified_diff_compacted"] diff --git a/cecli/helpers/diff/diff_helpers.py b/cecli/helpers/diff/diff_helpers.py new file mode 100644 index 00000000000..88ce1db711a --- /dev/null +++ b/cecli/helpers/diff/diff_helpers.py @@ -0,0 +1,154 @@ +"""Generic unified-diff helpers. + +These helpers are independent of hashline/HashPos formatting: they operate on +plain text lines and only use :mod:`difflib`. +""" + +import difflib + + +def unified_diff_compacted(old_lines, new_lines, fromfile="", tofile="", context_lines=3): + """Build a unified diff with git-style change-group compaction. + + ``difflib`` can anchor a change to the first of several identical blocks, + which makes unchanged lines look added or removed. Git compacts change + groups to their last possible position; this mirrors that for pure + insertions and deletions. + + Returns: + list[str]: Unified diff lines without trailing newlines, or an empty + list when the inputs are identical. + """ + matcher = difflib.SequenceMatcher(None, old_lines, new_lines) + opcodes = _compact_diff_opcodes(matcher.get_opcodes(), old_lines, new_lines) + diff_lines = [] + + for group in _group_diff_opcodes(opcodes, context_lines): + first = group[0] + last = group[-1] + diff_lines.append( + f"@@ -{_format_diff_range(first[1], last[2])} +" + f"{_format_diff_range(first[3], last[4])} @@" + ) + + for tag, i1, i2, j1, j2 in group: + if tag == "equal": + diff_lines.extend(" " + line for line in old_lines[i1:i2]) + elif tag in ("replace", "delete"): + diff_lines.extend("-" + line for line in old_lines[i1:i2]) + if tag in ("replace", "insert"): + diff_lines.extend("+" + line for line in new_lines[j1:j2]) + + if not diff_lines: + return [] + + return [f"--- {fromfile}", f"+++ {tofile}"] + diff_lines + + +def _compact_diff_opcodes(opcodes, old_lines, new_lines): + """Slide pure insert/delete groups down over repeated lines.""" + ops = [list(op) for op in opcodes] + index = 0 + + while index < len(ops): + if ops[index][0] == "equal": + index += 1 + continue + + start = index + while index < len(ops) and ops[index][0] != "equal": + index += 1 + + end = index + + if end - start != 1: + continue + + tag = ops[start][0] + if tag not in ("insert", "delete"): + continue + + while end < len(ops) and ops[end][0] == "equal": + equal = ops[end] + + if tag == "insert": + if ops[start][3] >= ops[start][4] or equal[3] >= equal[4]: + break + if new_lines[ops[start][3]] != new_lines[equal[3]]: + break + else: + if ops[start][1] >= ops[start][2] or equal[1] >= equal[2]: + break + if old_lines[ops[start][1]] != old_lines[equal[1]]: + break + + if start == 0 or ops[start - 1][0] != "equal": + ops.insert( + start, + ["equal", ops[start][1], ops[start][1], ops[start][3], ops[start][3]], + ) + start += 1 + end += 1 + + previous = ops[start - 1] + previous[2] += 1 + previous[4] += 1 + + ops[start][1] += 1 + ops[start][2] += 1 + ops[start][3] += 1 + ops[start][4] += 1 + + equal[1] += 1 + equal[3] += 1 + + if equal[1] == equal[2]: + del ops[end] + + index = end + + return [tuple(op) for op in ops] + + +def _group_diff_opcodes(opcodes, context_lines): + """Group opcodes into hunks with ``context_lines`` of surrounding context.""" + codes = [list(op) for op in opcodes] + if not codes: + codes = [["equal", 0, 1, 0, 1]] + + if codes[0][0] == "equal": + tag, i1, i2, j1, j2 = codes[0] + codes[0] = [tag, max(i1, i2 - context_lines), i2, max(j1, j2 - context_lines), j2] + + if codes[-1][0] == "equal": + tag, i1, i2, j1, j2 = codes[-1] + codes[-1] = [tag, i1, min(i2, i1 + context_lines), j1, min(j2, j1 + context_lines)] + + spacing = context_lines * 2 + group = [] + + for tag, i1, i2, j1, j2 in codes: + if tag == "equal" and i2 - i1 > spacing: + group.append((tag, i1, min(i2, i1 + context_lines), j1, min(j2, j1 + context_lines))) + yield group + group = [] + i1, j1 = max(i1, i2 - context_lines), max(j1, j2 - context_lines) + + group.append((tag, i1, i2, j1, j2)) + + if group and not (len(group) == 1 and group[0][0] == "equal"): + yield group + + +def _format_diff_range(start, stop): + """Format a unified-diff range the same way :mod:`difflib` does.""" + beginning = start + 1 + length = stop - start + + if length == 1: + return str(beginning) + + if not length: + beginning -= 1 + + return f"{beginning},{length}" diff --git a/cecli/helpers/hashline.py b/cecli/helpers/hashline.py index da66a70a7c8..f232e5e85ff 100644 --- a/cecli/helpers/hashline.py +++ b/cecli/helpers/hashline.py @@ -3,6 +3,7 @@ import re from collections import Counter +from cecli.helpers.diff.diff_helpers import unified_diff_compacted from cecli.helpers.hashpos.hashpos import ( # noqa HASH_DELIMITER, UNIQUE_HASH_DELIMITER, @@ -716,16 +717,14 @@ def get_hashline_content_diff( old_hashlines, old_content_to_lines, old_content_lines = _parse_content_for_diff(old_content) new_hashlines, new_content_to_lines, new_content_lines = _parse_content_for_diff(new_content) - # Generate content-only diff - content_diff = difflib.unified_diff( + # Generate content-only diff with change groups compacted to a stable position. + content_diff_lines = unified_diff_compacted( old_content_lines, new_content_lines, fromfile=fromfile, tofile=tofile, - lineterm="", - n=context_lines, + context_lines=context_lines, ) - content_diff_lines = list(content_diff) # If there's no content change, return empty string if not content_diff_lines: diff --git a/cecli/helpers/io_proxy.py b/cecli/helpers/io_proxy.py index da7a430412f..c3cafd2f2a4 100644 --- a/cecli/helpers/io_proxy.py +++ b/cecli/helpers/io_proxy.py @@ -138,6 +138,22 @@ def reset_streaming_response(self, **kwargs) -> Any: kwargs["coder_uuid"] = self._coder_uuid return self._target.reset_streaming_response(**kwargs) + def ring_bell(self, **kwargs: Any) -> Any: + """Forward ring_bell with coder_uuid and is_workflow injected.""" + if "coder_uuid" not in kwargs: + kwargs["coder_uuid"] = self._coder_uuid + if "is_workflow" not in kwargs: + kwargs["is_workflow"] = self._is_workflow_agent() + return self._target.ring_bell(**kwargs) + + def notify_user_input_required(self, **kwargs: Any) -> Any: + """Forward notify_user_input_required with coder_uuid and is_workflow injected.""" + if "coder_uuid" not in kwargs: + kwargs["coder_uuid"] = self._coder_uuid + if "is_workflow" not in kwargs: + kwargs["is_workflow"] = self._is_workflow_agent() + return self._target.notify_user_input_required(**kwargs) + async def get_input(self, *args, **kwargs): """Get input for this specific coder via per-coder queue. @@ -310,6 +326,20 @@ def __setattr__(self, name: str, value: Any) -> None: else: setattr(self._target, name, value) + def _is_workflow_agent(self) -> bool: + """Return True if the proxied coder is a workflow (wf:*) sub-agent.""" + _ref = getattr(self, "_coder", None) + coder = _ref() if _ref is not None else None + if coder is None: + return False + try: + from cecli.helpers.agents.service import AgentService + + name = AgentService.get_agent_name(coder) + return name is not None and name.startswith("wf:") + except Exception: + return False + # --- THE TYPE HINTING TRICK --- # At type-checking time, make IOProxy(target, coder) appear to return diff --git a/cecli/helpers/server/acp/agent/io.py b/cecli/helpers/server/acp/agent/io.py index 074b45e2421..2ba3265006b 100644 --- a/cecli/helpers/server/acp/agent/io.py +++ b/cecli/helpers/server/acp/agent/io.py @@ -194,8 +194,8 @@ def display_user_input(self, inp): def rule(self): pass - def ring_bell(self): + def ring_bell(self, **kwargs): pass - def notify_user_input_required(self): + def notify_user_input_required(self, **kwargs): pass diff --git a/cecli/helpers/workflows/config.py b/cecli/helpers/workflows/config.py index 8b49b955744..1a8ad8839fa 100644 --- a/cecli/helpers/workflows/config.py +++ b/cecli/helpers/workflows/config.py @@ -14,6 +14,7 @@ 4. ``/.agents/workflows/*.yml`` / ``*.yaml`` — globbed in stable sorted order. 5. ``merged_config["workflows"]`` — the deep-merged ``.cecli.conf.yml`` value. 6. ``~/.cecli/workflows/*.yml`` / ``*.yaml`` — globbed in stable sorted order. +7. ``~/.agents/workflows/*.yml`` / ``*.yaml`` — globbed in stable sorted order. Each source may hold a single workflow mapping or a ``workflows: [...]`` list. On a duplicate ``name`` the higher-precedence definition wins and the loser is @@ -104,6 +105,7 @@ def resolve_workflow_configs( ordered += _entries(merged_config.get("workflows"), None, _CONFIG_LABEL) ordered += _entries_from_home() + ordered += _entries_from_home_agents() return _dedupe(ordered) @@ -315,6 +317,12 @@ def _entries_from_home() -> List[_Discovered]: return _entries_from_dir(Path.home() / WORKFLOW_DIR) +def _entries_from_home_agents() -> List[_Discovered]: + """Resolve ``~/.agents/workflows/*.yml|*.yaml`` in stable sorted order.""" + + return _entries_from_dir(Path.home() / AGENTS_WORKFLOW_DIR) + + def _entries_from_dir(root: Path) -> List[_Discovered]: """Resolve ``*.yml`` / ``*.yaml`` files under *root* in stable sorted order.""" if not root.is_dir(): diff --git a/cecli/io.py b/cecli/io.py index 58829863a4a..e4d55eb45da 100644 --- a/cecli/io.py +++ b/cecli/io.py @@ -1891,13 +1891,18 @@ def _send_notification(self): except Exception as e: self.tool_warning(f"Failed to run bell command: {e}") - def notify_user_input_required(self): + def notify_user_input_required(self, **kwargs): """Send a notification that user input is required.""" + if kwargs.get("is_workflow", False): + return if self.notifications or self.notification_bell or self.custom_notification_command: self._send_notification() - def ring_bell(self): - """Ring the terminal bell if needed and clear the flag""" + def ring_bell(self, **kwargs): + """Ring the terminal bell if needed and clear the flag.""" + if kwargs.get("is_workflow", False): + self.bell_on_next_input = False + return if self.bell_on_next_input: if self.notifications or self.notification_bell or self.custom_notification_command: self._send_notification() diff --git a/cecli/main.py b/cecli/main.py index 42554098805..d3868efb74d 100644 --- a/cecli/main.py +++ b/cecli/main.py @@ -1356,7 +1356,9 @@ def get_io(pretty): coder = await Coder.create( from_coder=from_coder, main_model=main_model, - edit_format=args.edit_format, + # Agent mode is the startup default; an explicit flag/config/env value + # still wins, and reloads keep the session's current mode. + edit_format=args.edit_format or "agent", io=io, args=args, repo=repo, diff --git a/cecli/prompts/agent.yml b/cecli/prompts/agent.yml index 8bc6e2a5d54..f9c13429e90 100644 --- a/cecli/prompts/agent.yml +++ b/cecli/prompts/agent.yml @@ -26,7 +26,7 @@ main_system: | **Be Persistent**: Work through your task until completion. No task takes too long as long as you are making progress towards the goal. ## FILE FORMAT - File contents are presented with virtual prefix identifiers to help you target edits accurately. These are generated on-the-fly when reading the file. They do not exist as bytes in the literal file on disk. + File contents are presented with virtual prefix identifiers to help you target edits accurately. These are generated on-the-fly when reading the file. They do not exist in the actual file on disk. - **Unique Lines (`——`):** Lines that appear only once in the file are prefixed with `——`. To target these lines for edits, you can simply reference the exact literal text of the line. - **Duplicate Lines (e.g., `—“0车加—`):** Lines that appear multiple times are prefixed with an opaque identifier. You MUST include this exact identifier when targeting these lines to disambiguate which specific instance you want to edit. @@ -60,13 +60,8 @@ main_system: | 4. **Verify & Recover**: If an edit fails or introduces linting errors, fix the error immediately. Use `UndoChange` if the edits corrupt the file or introduce syntax errors. 5. **Yield**: Use the `Yield` tool after accomplishing the goal and verifying any changes made. Provide helpful summaries of any changes. - ## Todo List Management - - Break complex goals into meaningful sub-tasks so the problem remains tractable - - Use `UpdateTodoList` to keep the state synchronized as you complete subtasks. - ## Operational Rules - **Scope**: No unrequested refactors. Avoid full-file rewrites. Only modify what you are asked to. - - **Hygiene**: Use `ResourceManager` to evict unneeded files/skills immediately after use. - **Outputs**: Tool calls trigger turns. Yield when finished with the final user summary. - **Sandbox**: Perform all verification and temp logic in `.cecli/temp`. - **Responses**: Be brief in your reasoning. Use bulleted lists of known facts and be terse in your deliberation on how to accomplish the goal. @@ -86,4 +81,4 @@ system_reminder: | try_again: | - My previous exploration was insufficient. I will now adjust my strategy, use more specific search patterns, and manage my context more aggressively to find the correct solution. \ No newline at end of file + My previous exploration was insufficient. I will now adjust my strategy, use more specific search patterns, and manage my context more aggressively to find the correct solution. diff --git a/cecli/prompts/subagent.yml b/cecli/prompts/subagent.yml index 6ecc17e5c7d..93590118368 100644 --- a/cecli/prompts/subagent.yml +++ b/cecli/prompts/subagent.yml @@ -11,7 +11,7 @@ main_system: | **Be Persistent**: Work through your task until completion. No task takes too long as long as you are making progress towards the goal. ## FILE FORMAT - File contents are presented with virtual prefix identifiers to help you target edits accurately. These are generated on-the-fly when reading the file. They do not exist as bytes in the literal file on disk. + File contents are presented with virtual prefix identifiers to help you target edits accurately. These are generated on-the-fly when reading the file. They do not exist in the actual file on disk. - **Unique Lines (`——`):** Lines that appear only once in the file are prefixed with `——`. To target these lines for edits, you can simply reference the exact literal text of the line. - **Duplicate Lines (e.g., `—“0车加—`):** Lines that appear multiple times are prefixed with an opaque identifier. You MUST include this exact identifier when targeting these lines to disambiguate which specific instance you want to edit. @@ -45,13 +45,8 @@ main_system: | 4. **Verify & Recover**: If an edit fails or introduces linting errors, fix the error immediately. Use `UndoChange` if the edits corrupt the file or introduce syntax errors. 5. **Yield**: Use the `Yield` tool after accomplishing the goal and verifying any changes made. Provide helpful summaries of any changes. - ## Todo List Management - - Break complex goals into meaningful sub-tasks so the problem remains tractable - - Use `UpdateTodoList` to keep the state synchronized as you complete subtasks. - ## Operational Rules - **Scope**: No unrequested refactors. Avoid full-file rewrites. Only modify what you are asked to. - - **Hygiene**: Use `ResourceManager` to evict unneeded files/skills immediately after use. - **Outputs**: Tool calls trigger turns. Yield when finished with the final user summary. - **Sandbox**: Perform all verification and temp logic in `.cecli/temp`. - **Responses**: Be brief in your reasoning. Use bulleted lists of known facts and be terse in your deliberation on how to accomplish the goal. @@ -73,4 +68,4 @@ system_reminder: | {lazy_prompt} {shell_cmd_reminder} - \ No newline at end of file + diff --git a/cecli/resources/model-metadata.ext.json b/cecli/resources/model-metadata.ext.json index 1774ca20027..c57a81d1ed6 100644 --- a/cecli/resources/model-metadata.ext.json +++ b/cecli/resources/model-metadata.ext.json @@ -1016,5 +1016,16 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true + }, + "agnes/agnes-3.0-flash": { + "max_tokens": 65536, + "max_input_tokens": 512000, + "max_output_tokens": 65536, + "input_cost_per_token": 0, + "output_cost_per_token": 0, + "litellm_provider": "openai", + "mode": "chat", + "source": "https://agnes-ai.com/en/docs/agnes-30-flash" } } + diff --git a/cecli/resources/model-metadata.json b/cecli/resources/model-metadata.json index ba611d4cfd7..a054515262b 100644 --- a/cecli/resources/model-metadata.json +++ b/cecli/resources/model-metadata.json @@ -1,4 +1,14 @@ { + "agnes/agnes-3.0-flash": { + "max_tokens": 65536, + "max_input_tokens": 512000, + "max_output_tokens": 65536, + "input_cost_per_token": 0, + "output_cost_per_token": 0, + "litellm_provider": "openai", + "mode": "chat", + "source": "https://agnes-ai.com/en/docs/agnes-30-flash" + }, "ai21.j2-mid-v1": { "input_cost_per_token": 0.0000125, "litellm_provider": "bedrock", @@ -24,22 +34,22 @@ "input_cost_per_token": 0.000002, "litellm_provider": "bedrock", "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 0.000008, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-ai21-labs-jamba-1-5-large.md" }, "ai21.jamba-1-5-mini-v1:0": { "deprecation_date": "2026-11-26", "input_cost_per_token": 2e-7, "litellm_provider": "bedrock", "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 4e-7, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-ai21-labs-jamba-1-5-mini.md" }, "ai21.jamba-instruct-v1:0": { "input_cost_per_token": 5e-7, @@ -1233,13 +1243,19 @@ }, "amazon.nova-2-lite-v1:0": { "cache_read_input_token_cost": 7.5e-8, + "cache_read_input_token_cost_priority": 1.3125e-7, "input_cost_per_token": 3e-7, + "input_cost_per_token_batches": 1.5e-7, + "input_cost_per_token_priority": 5.25e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 1000000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", "output_cost_per_token": 0.0000025, + "output_cost_per_token_batches": 0.00000125, + "output_cost_per_token_priority": 0.000004375, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -1254,13 +1270,17 @@ "input_cost_per_token": 0.00000125, "input_cost_per_image_token": 0.00000125, "input_cost_per_audio_token": 0.00000125, + "input_cost_per_token_batches": 6.25e-7, + "input_cost_per_token_priority": 0.0000021875, "litellm_provider": "bedrock_converse", "max_input_tokens": 1000000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", "output_cost_per_token": 0.00001, - "source": "https://aws.amazon.com/nova/pricing/", + "output_cost_per_token_batches": 0.000005, + "output_cost_per_token_priority": 0.0000175, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -1273,12 +1293,15 @@ "amazon.nova-lite-v1:0": { "cache_read_input_token_cost": 1.5e-8, "input_cost_per_token": 6e-8, + "input_cost_per_token_batches": 3e-8, "litellm_provider": "bedrock_converse", "max_input_tokens": 300000, "max_output_tokens": 10000, "max_tokens": 10000, "mode": "chat", "output_cost_per_token": 2.4e-7, + "output_cost_per_token_batches": 1.2e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -1289,26 +1312,60 @@ "amazon.nova-micro-v1:0": { "cache_read_input_token_cost": 8.75e-9, "input_cost_per_token": 3.5e-8, + "input_cost_per_token_batches": 1.75e-8, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, "max_output_tokens": 10000, "max_tokens": 10000, "mode": "chat", "output_cost_per_token": 1.4e-7, + "output_cost_per_token_batches": 7e-8, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_prompt_caching": true, "supports_response_schema": true, "supports_tool_choice": true }, + "amazon.nova-premier-v1:0": { + "cache_read_input_token_cost": 6.25e-7, + "cache_read_input_token_cost_priority": 0.00000109375, + "input_cost_per_token": 0.0000025, + "input_cost_per_token_batches": 0.00000125, + "input_cost_per_token_priority": 0.000004375, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 10000, + "max_tokens": 10000, + "mode": "chat", + "output_cost_per_token": 0.0000125, + "output_cost_per_token_batches": 0.00000625, + "output_cost_per_token_priority": 0.000021875, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "amazon.nova-pro-v1:0": { "cache_read_input_token_cost": 2e-7, + "cache_read_input_token_cost_flex": 1e-7, + "cache_read_input_token_cost_priority": 3.5e-7, "input_cost_per_token": 8e-7, + "input_cost_per_token_batches": 4e-7, + "input_cost_per_token_flex": 4e-7, + "input_cost_per_token_priority": 0.0000014, "litellm_provider": "bedrock_converse", "max_input_tokens": 300000, "max_output_tokens": 10000, "max_tokens": 10000, "mode": "chat", "output_cost_per_token": 0.0000032, + "output_cost_per_token_batches": 0.0000016, + "output_cost_per_token_flex": 0.0000016, + "output_cost_per_token_priority": 0.0000056, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -1317,22 +1374,22 @@ "supports_tool_choice": true }, "amazon.titan-text-express-v1": { - "input_cost_per_token": 0.0000013, + "input_cost_per_token": 2e-7, "litellm_provider": "bedrock", "max_input_tokens": 42000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", - "output_cost_per_token": 0.0000017 + "output_cost_per_token": 6e-7 }, "amazon.titan-text-lite-v1": { - "input_cost_per_token": 3e-7, + "input_cost_per_token": 1.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 42000, "max_output_tokens": 4000, "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 4e-7 + "output_cost_per_token": 2e-7 }, "amazon.titan-text-premier-v1:0": { "input_cost_per_token": 5e-7, @@ -1364,7 +1421,7 @@ "anthropic.claude-3-5-sonnet-20240620-v1:0": { "input_cost_per_token": 0.000003, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 200000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", @@ -1374,10 +1431,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "input_cost_per_token_above_200k_tokens": 0.000006, - "output_cost_per_token_above_200k_tokens": 0.00003, - "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, - "cache_read_input_token_cost_above_200k_tokens": 6e-7, "cache_creation_input_token_cost": 0.00000375, "cache_read_input_token_cost": 3e-7 }, @@ -1386,7 +1439,7 @@ "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 200000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", @@ -1398,11 +1451,7 @@ "supports_prompt_caching": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "input_cost_per_token_above_200k_tokens": 0.000006, - "output_cost_per_token_above_200k_tokens": 0.00003, - "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, - "cache_read_input_token_cost_above_200k_tokens": 6e-7 + "supports_vision": true }, "anthropic.claude-3-7-sonnet-20240620-v1:0": { "cache_creation_input_token_cost": 0.0000045, @@ -1489,7 +1538,7 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": true, + "supports_native_structured_output": false, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -1603,6 +1652,10 @@ "cache_read_input_token_cost": 1e-8, "input_cost_per_token": 1e-7, "litellm_provider": "bedrock_converse", + "token_count_model": { + "model": "anthropic.claude-opus-5-5", + "region": "us-east-1" + }, "supports_tool_search": true, "max_input_tokens": 1000000, "max_output_tokens": 128000, @@ -1628,6 +1681,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -1892,6 +1946,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -1987,7 +2042,7 @@ "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, "cache_read_input_token_cost_above_200k_tokens": 6e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "max_input_tokens": 200000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", @@ -2013,13 +2068,14 @@ "anthropic.claude-sonnet-4-5-20250929-v1:0": { "cache_creation_input_token_cost": 0.00000375, "cache_creation_input_token_cost_above_1hr": 0.000006, + "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.000012, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "input_cost_per_token_above_200k_tokens": 0.000006, "output_cost_per_token_above_200k_tokens": 0.0000225, "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, - "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.000012, "cache_read_input_token_cost_above_200k_tokens": 6e-7, + "deprecation_date": "2027-04-08", "litellm_provider": "bedrock_converse", "supports_tool_search": true, "max_input_tokens": 200000, @@ -2067,7 +2123,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -2115,6 +2171,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -2129,6 +2186,10 @@ "cache_read_input_token_cost": 1e-7, "input_cost_per_token": 0.000002, "litellm_provider": "bedrock_converse", + "token_count_model": { + "model": "anthropic.claude-opus-5-5", + "region": "us-east-1" + }, "supports_tool_search": true, "max_input_tokens": 1000000, "max_output_tokens": 128000, @@ -2154,6 +2215,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -2301,14 +2363,14 @@ "supports_function_calling": true }, "apac.amazon.nova-2-lite-v1:0": { - "cache_read_input_token_cost": 8.25e-8, - "input_cost_per_token": 3.3e-7, + "cache_read_input_token_cost": 9.9e-8, + "input_cost_per_token": 3.96e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 1000000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", - "output_cost_per_token": 0.00000275, + "output_cost_per_token": 0.000003311, "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -2371,13 +2433,16 @@ }, "apac.amazon.nova-pro-v1:0": { "cache_read_input_token_cost": 2.1e-7, + "cache_read_input_token_cost_flex": 1.05e-7, "input_cost_per_token": 8.4e-7, + "input_cost_per_token_flex": 4.2e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 300000, "max_output_tokens": 10000, "max_tokens": 10000, "mode": "chat", "output_cost_per_token": 0.00000336, + "output_cost_per_token_flex": 0.00000168, "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -2405,7 +2470,7 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": true, + "supports_native_structured_output": false, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -2472,6 +2537,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -2561,6 +2627,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -2651,7 +2718,7 @@ "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, "cache_read_input_token_cost_above_200k_tokens": 6e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "max_input_tokens": 200000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", @@ -2683,7 +2750,7 @@ "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -2726,6 +2793,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -2760,6 +2828,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -2794,7 +2863,7 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": true, + "supports_native_structured_output": false, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -2836,6 +2905,10 @@ "source": "https://aws.amazon.com/bedrock/pricing/" }, "au.anthropic.claude-haiku-5-5": { + "token_count_model": { + "model": "au.anthropic.claude-opus-5-5", + "region": "ap-southeast-4" + }, "input_cost_per_token_above_100k_tokens": 5.5e-7, "output_cost_per_token_above_100k_tokens": 0.00000275, "cache_creation_input_token_cost_above_100k_tokens": 6.875e-7, @@ -2872,6 +2945,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -3007,6 +3081,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -3102,6 +3177,7 @@ "cache_creation_input_token_cost_above_200k_tokens": 0.00000825, "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.0000132, "cache_read_input_token_cost_above_200k_tokens": 6.6e-7, + "deprecation_date": "2027-04-08", "litellm_provider": "bedrock_converse", "supports_tool_search": true, "max_input_tokens": 200000, @@ -3149,7 +3225,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -3197,6 +3273,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -3205,6 +3282,10 @@ "source": "https://aws.amazon.com/bedrock/pricing/" }, "au.anthropic.claude-sonnet-5-5": { + "token_count_model": { + "model": "au.anthropic.claude-opus-5-5", + "region": "ap-southeast-4" + }, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 0.00000275, "cache_creation_input_token_cost_above_1hr": 0.0000044, @@ -3236,6 +3317,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -3282,7 +3364,7 @@ "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, - "supports_none_reasoning_effort": true, + "supports_none_reasoning_effort": false, "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -4465,7 +4547,7 @@ "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, - "supports_none_reasoning_effort": true, + "supports_none_reasoning_effort": false, "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -4853,38 +4935,41 @@ "supports_tool_choice": true }, "azure/gpt-4-turbo": { - "input_cost_per_token": 0.00001, + "input_cost_per_token": 0.000011, "litellm_provider": "azure", "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 0.00003, + "output_cost_per_token": 0.000033, + "source": "https://azure.microsoft.com/en-us/pricing/details/azure-openai/", "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_tool_choice": true }, "azure/gpt-4-turbo-2024-04-09": { - "input_cost_per_token": 0.00001, + "input_cost_per_token": 0.000011, "litellm_provider": "azure", "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 0.00003, + "output_cost_per_token": 0.000033, + "source": "https://azure.microsoft.com/en-us/pricing/details/azure-openai/", "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_tool_choice": true, "supports_vision": true }, "azure/gpt-4-turbo-vision-preview": { - "input_cost_per_token": 0.00001, + "input_cost_per_token": 0.000011, "litellm_provider": "azure", "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 0.00003, + "output_cost_per_token": 0.000033, + "source": "https://azure.microsoft.com/en-us/pricing/details/azure-openai/", "supports_tool_choice": true, "supports_vision": true }, @@ -7078,7 +7163,7 @@ "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, - "supports_none_reasoning_effort": true, + "supports_none_reasoning_effort": false, "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -7127,7 +7212,7 @@ "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, - "supports_none_reasoning_effort": true, + "supports_none_reasoning_effort": false, "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -7336,7 +7421,7 @@ }, "azure/gpt-chat-latest": { "cache_read_input_token_cost": 5e-7, - "deprecation_date": "2026-12-02", + "deprecation_date": "2027-05-18", "input_cost_per_token": 0.000005, "litellm_provider": "azure", "max_input_tokens": 272000, @@ -7538,45 +7623,6 @@ "supports_tool_choice": true, "supports_vision": true }, - "azure/o3-deep-research": { - "cache_read_input_token_cost": 0.0000025, - "deprecation_date": "2026-11-19", - "input_cost_per_token": 0.00001, - "litellm_provider": "azure", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "responses", - "output_cost_per_token": 0.00004, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/batch", - "/v1/responses" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, "azure/o3-mini": { "deprecation_date": "2026-11-19", "cache_read_input_token_cost": 5.5e-7, @@ -7611,70 +7657,6 @@ "supports_tool_choice": true, "supports_vision": false }, - "azure/o3-pro": { - "deprecation_date": "2026-11-19", - "input_cost_per_token": 0.00002, - "input_cost_per_token_batches": 0.00001, - "litellm_provider": "azure", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "responses", - "output_cost_per_token": 0.00008, - "output_cost_per_token_batches": 0.00004, - "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/batch", - "/v1/responses" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "azure/o3-pro-2025-06-10": { - "deprecation_date": "2026-11-19", - "input_cost_per_token": 0.00002, - "input_cost_per_token_batches": 0.00001, - "litellm_provider": "azure", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "responses", - "output_cost_per_token": 0.00008, - "output_cost_per_token_batches": 0.00004, - "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/batch", - "/v1/responses" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, "azure/o4-mini": { "deprecation_date": "2026-11-19", "cache_read_input_token_cost": 2.75e-7, @@ -8988,7 +8970,7 @@ "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, - "supports_none_reasoning_effort": true, + "supports_none_reasoning_effort": false, "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -9002,7 +8984,7 @@ }, "azure/us/gpt-chat-latest": { "cache_read_input_token_cost": 5.5e-7, - "deprecation_date": "2026-12-02", + "deprecation_date": "2027-05-18", "input_cost_per_token": 0.0000055, "litellm_provider": "azure", "max_input_tokens": 272000, @@ -10220,6 +10202,43 @@ "prompt_cache_min_tokens": 512, "source": "https://management.azure.com/subscriptions/c873328e-b572-4770-8dff-aaeb6f1f0e79/providers/Microsoft.CognitiveServices/locations/eastus2/models?api-version=2024-10-01" }, + "azure_ai/claude-opus-5-5-fast": { + "cache_creation_input_token_cost": 0.00001, + "cache_creation_input_token_cost_above_1hr": 0.000016, + "cache_read_input_token_cost": 4e-7, + "deprecation_date": "2027-10-01", + "input_cost_per_token": 0.000008, + "litellm_provider": "azure_ai", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00004, + "prompt_cache_min_tokens": 512, + "search_context_cost_per_query": { + "search_context_size_high": 0.02, + "search_context_size_low": 0.02, + "search_context_size_medium": 0.02 + }, + "source": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/claude-models", + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_forced_tool_use": false, + "supports_function_calling": true, + "supports_max_reasoning_effort": true, + "supports_mid_conversation_system": true, + "supports_native_structured_output": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "thinking_always_on": true + }, "azure_ai/claude-sonnet-4-5": { "deprecation_date": "2026-11-30", "cache_creation_input_token_cost": 0.00000375, @@ -10258,7 +10277,7 @@ "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 0.000015, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -11054,7 +11073,7 @@ "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, - "supports_none_reasoning_effort": true, + "supports_none_reasoning_effort": false, "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_cache_breakpoint": true, @@ -11069,7 +11088,7 @@ }, "azure_ai/gpt-chat-latest": { "cache_read_input_token_cost": 5e-7, - "deprecation_date": "2026-12-02", + "deprecation_date": "2027-05-18", "input_cost_per_token": 0.000005, "litellm_provider": "azure_ai", "max_input_tokens": 272000, @@ -11417,6 +11436,7 @@ "supports_vision": true }, "azure_ai/ministral-3b": { + "deprecation_date": "2026-12-09", "input_cost_per_token": 4e-8, "litellm_provider": "azure_ai", "max_input_tokens": 128000, @@ -11480,6 +11500,7 @@ "supports_tool_choice": true }, "azure_ai/mistral-medium-2505": { + "deprecation_date": "2026-12-09", "input_cost_per_token": 4e-7, "litellm_provider": "azure_ai", "max_input_tokens": 131072, @@ -11536,6 +11557,7 @@ "supports_tool_choice": true }, "azure_ai/mistral-small-2503": { + "deprecation_date": "2026-12-09", "input_cost_per_token": 1e-7, "litellm_provider": "azure_ai", "max_input_tokens": 128000, @@ -11657,13 +11679,47 @@ "supports_tool_choice": true, "supports_vision": true }, + "baseten/deepseek-ai/DeepSeek-V4.1-Flash-1017": { + "cache_read_input_token_cost": 7e-9, + "input_cost_per_token": 3e-7, + "litellm_provider": "baseten", + "max_input_tokens": 1048576, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://inference.baseten.co/v1/models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "baseten/deepseek-ai/DeepSeek-V4.1-Flash-1017-Fast": { + "cache_read_input_token_cost": 1.4e-8, + "input_cost_per_token": 6e-7, + "litellm_provider": "baseten", + "max_input_tokens": 1048576, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.0000024, + "source": "https://inference.baseten.co/v1/models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "baseten/deepseek-ai/DeepSeek-V4.1-Flash-Fast": { - "cache_read_input_token_cost": 1.4e-7, + "cache_read_input_token_cost": 1.4e-8, "input_cost_per_token": 6e-7, "litellm_provider": "baseten", "max_input_tokens": 1048576, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", "output_cost_per_token": 0.0000024, "source": "https://inference.baseten.co/v1/models", @@ -11674,6 +11730,24 @@ "supports_tool_choice": true, "supports_vision": true }, + "baseten/meta-models/muse-spark-open": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.00000125, + "litellm_provider": "baseten", + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.00000425, + "source": "https://inference.baseten.co/v1/models", + "supports_audio_input": true, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": false, + "supports_tool_choice": true, + "supports_vision": true + }, "baseten/moonshotai/Kimi-K2-Instruct-0905": { "input_cost_per_token": 6e-7, "litellm_provider": "baseten", @@ -11760,6 +11834,23 @@ "supports_tool_choice": true, "supports_vision": false }, + "baseten/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 2e-7, + "litellm_provider": "baseten", + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 5e-7, + "source": "https://inference.baseten.co/v1/models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "baseten/nvidia/Nemotron-120B-A12B": { "input_cost_per_token": 3e-7, "litellm_provider": "baseten", @@ -11783,6 +11874,22 @@ "supports_tool_choice": true, "supports_vision": false }, + "baseten/reflection/Beam-1.0": { + "cache_read_input_token_cost": 1.3e-7, + "input_cost_per_token": 0.0000013, + "litellm_provider": "baseten", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 0.0000043, + "source": "https://inference.baseten.co/v1/models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true + }, "baseten/thinkingmachines/inkling": { "cache_read_input_token_cost": 1.7e-7, "input_cost_per_token": 0.000001, @@ -12075,12 +12182,14 @@ }, "bedrock/ap-northeast-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-7, + "input_cost_per_token_flex": 3.7e-7, "litellm_provider": "bedrock", "max_input_tokens": 163840, - "max_output_tokens": 163840, - "max_tokens": 163840, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000222, + "output_cost_per_token_flex": 0.00000111, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -12088,32 +12197,36 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" }, "bedrock/ap-northeast-1/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/ap-northeast-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12121,43 +12234,52 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000144 + "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7 }, "bedrock/ap-northeast-1/moonshotai.kimi-k2-thinking": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 7.3e-7, + "input_cost_per_token_flex": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.00000303, + "output_cost_per_token_flex": 0.00000152, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-thinking.html", "supports_function_calling": true, "supports_reasoning": true }, "bedrock/ap-northeast-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-7, + "input_cost_per_token_flex": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.0000036, + "output_cost_per_token_flex": 0.0000018, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" }, "bedrock/ap-northeast-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -12184,12 +12306,14 @@ }, "bedrock/ap-south-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-7, + "input_cost_per_token_flex": 3.7e-7, "litellm_provider": "bedrock", "max_input_tokens": 163840, - "max_output_tokens": 163840, - "max_tokens": 163840, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000222, + "output_cost_per_token_flex": 0.00000111, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -12197,50 +12321,58 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" }, "bedrock/ap-south-1/meta.llama3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 0.00000318, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.0000042 + "output_cost_per_token": 0.0000042, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-70b-instruct.md" }, "bedrock/ap-south-1/meta.llama3-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 7.2e-7 + "output_cost_per_token": 7.2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-8b-instruct.md" }, "bedrock/ap-south-1/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/ap-south-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12248,43 +12380,52 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000144 + "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7 }, "bedrock/ap-south-1/moonshotai.kimi-k2-thinking": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 7.1e-7, + "input_cost_per_token_flex": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.00000294, + "output_cost_per_token_flex": 0.00000147, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-thinking.html", "supports_function_calling": true, "supports_reasoning": true }, "bedrock/ap-south-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-7, + "input_cost_per_token_flex": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.0000036, + "output_cost_per_token_flex": 0.0000018, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" }, "bedrock/ap-south-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -12311,12 +12452,13 @@ }, "bedrock/ap-southeast-2/minimax.minimax-m2.5": { "input_cost_per_token": 3.1e-7, + "input_cost_per_token_flex": 1.55e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12324,7 +12466,8 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000124 + "output_cost_per_token": 0.00000124, + "output_cost_per_token_flex": 6.2e-7 }, "bedrock/ap-southeast-2/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.545e-7, @@ -12344,12 +12487,14 @@ }, "bedrock/ap-southeast-3/deepseek.v3.2": { "input_cost_per_token": 7.4e-7, + "input_cost_per_token_flex": 3.7e-7, "litellm_provider": "bedrock", "max_input_tokens": 163840, - "max_output_tokens": 163840, - "max_tokens": 163840, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000222, + "output_cost_per_token_flex": 0.00000111, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -12357,32 +12502,36 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" }, "bedrock/ap-southeast-3/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/ap-southeast-3/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12390,32 +12539,37 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000144 + "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7 }, "bedrock/ap-southeast-3/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-7, + "input_cost_per_token_flex": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.0000036, + "output_cost_per_token_flex": 0.0000018, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" }, "bedrock/ap-southeast-3/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -12425,22 +12579,26 @@ "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" }, "bedrock/ca-central-1/meta.llama3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 0.00000305, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.00000403 + "output_cost_per_token": 0.00000403, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-70b-instruct.md" }, "bedrock/ca-central-1/meta.llama3-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 6.9e-7 + "output_cost_per_token": 6.9e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-8b-instruct.md" }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-instant-v1": { "cost_per_second": 0.01635, @@ -12530,29 +12688,33 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/eu-central-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12560,16 +12722,19 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000144 + "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7 }, "bedrock/eu-central-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -12580,12 +12745,14 @@ }, "bedrock/eu-north-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-7, + "input_cost_per_token_flex": 3.7e-7, "litellm_provider": "bedrock", "max_input_tokens": 163840, - "max_output_tokens": 163840, - "max_tokens": 163840, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000222, + "output_cost_per_token_flex": 0.00000111, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -12593,32 +12760,36 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" }, "bedrock/eu-north-1/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/eu-north-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12626,48 +12797,55 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000144 + "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7 }, "bedrock/eu-north-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-7, + "input_cost_per_token_flex": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.0000036, + "output_cost_per_token_flex": 0.0000018, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" }, "bedrock/eu-south-1/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/eu-south-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12675,16 +12853,19 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000144 + "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7 }, "bedrock/eu-south-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -12694,47 +12875,55 @@ "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" }, "bedrock/eu-west-1/meta.llama3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 0.00000286, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.00000378 + "output_cost_per_token": 0.00000378, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-70b-instruct.md" }, "bedrock/eu-west-1/meta.llama3-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.2e-7, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 6.5e-7 + "output_cost_per_token": 6.5e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-8b-instruct.md" }, "bedrock/eu-west-1/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/eu-west-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12742,16 +12931,19 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000144 + "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7 }, "bedrock/eu-west-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -12777,47 +12969,55 @@ "supports_vision": false }, "bedrock/eu-west-2/meta.llama3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 0.00000345, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.00000455 + "output_cost_per_token": 0.00000455, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-70b-instruct.md" }, "bedrock/eu-west-2/meta.llama3-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.9e-7, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 7.8e-7 + "output_cost_per_token": 7.8e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-8b-instruct.md" }, "bedrock/eu-west-2/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 4.7e-7, + "input_cost_per_token_flex": 2.35e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.00000186, + "output_cost_per_token_flex": 9.3e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/eu-west-2/minimax.minimax-m2.5": { "input_cost_per_token": 4.7e-7, + "input_cost_per_token_flex": 2.35e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12825,16 +13025,19 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000186 + "output_cost_per_token": 0.00000186, + "output_cost_per_token_flex": 9.3e-7 }, "bedrock/eu-west-2/nvidia.nemotron-super-3-120b": { "input_cost_per_token": 2.3e-7, + "input_cost_per_token_flex": 1.15e-7, "litellm_provider": "bedrock", "max_input_tokens": 256000, "max_output_tokens": 32000, "max_tokens": 32000, "mode": "chat", "output_cost_per_token": 0.00000101, + "output_cost_per_token_flex": 5.05e-7, "source": "https://aws.amazon.com/bedrock/pricing/", "supports_audio_input": false, "supports_function_calling": true, @@ -12846,12 +13049,14 @@ }, "bedrock/eu-west-2/qwen.qwen3-coder-next": { "input_cost_per_token": 7.8e-7, + "input_cost_per_token_flex": 3.9e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.00000186, + "output_cost_per_token_flex": 9.3e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -12877,34 +13082,38 @@ "supports_vision": false }, "bedrock/eu-west-3/mistral.mistral-7b-instruct-v0:2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 2e-7, "litellm_provider": "bedrock", "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 2.6e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mistral-7b-instruct.md", "supports_tool_choice": true }, "bedrock/eu-west-3/mistral.mistral-large-2402-v1:0": { "input_cost_per_token": 0.0000052, "litellm_provider": "bedrock", "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 0.0000156, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mistral-large.md", "supports_function_calling": true }, "bedrock/eu-west-3/mistral.mixtral-8x7b-instruct-v0:1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 5.9e-7, "litellm_provider": "bedrock", "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 9.1e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mixtral-8x7b-instruct.md", "supports_tool_choice": true }, "bedrock/invoke/anthropic.claude-3-5-sonnet-20240620-v1:0": { @@ -12926,25 +13135,31 @@ "cache_creation_input_token_cost": 0.00000375 }, "bedrock/moonshotai.kimi-k2-thinking": { - "input_cost_per_token": 7.3e-7, + "deprecation_date": "2027-03-30", + "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", - "output_cost_per_token": 0.00000303, + "output_cost_per_token": 0.0000025, + "output_cost_per_token_flex": 0.00000125, + "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", "supports_function_calling": true, "supports_reasoning": true }, "bedrock/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", - "output_cost_per_token": 0.00000303, - "source": "https://platform.moonshot.ai/docs/guide/kimi-k2-5-quickstart", + "output_cost_per_token": 0.000003, + "output_cost_per_token_flex": 0.0000015, + "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -12956,12 +13171,14 @@ }, "bedrock/sa-east-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-7, + "input_cost_per_token_flex": 3.7e-7, "litellm_provider": "bedrock", "max_input_tokens": 163840, - "max_output_tokens": 163840, - "max_tokens": 163840, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000222, + "output_cost_per_token_flex": 0.00000111, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -12969,50 +13186,58 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" }, "bedrock/sa-east-1/meta.llama3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 0.00000445, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.00000588 + "output_cost_per_token": 0.00000588, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-70b-instruct.md" }, "bedrock/sa-east-1/meta.llama3-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 5e-7, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.00000101 + "output_cost_per_token": 0.00000101, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-8b-instruct.md" }, "bedrock/sa-east-1/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/sa-east-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-7, + "input_cost_per_token_flex": 1.8e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -13020,43 +13245,52 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.00000144 + "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7 }, "bedrock/sa-east-1/moonshotai.kimi-k2-thinking": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 7.3e-7, + "input_cost_per_token_flex": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.00000303, + "output_cost_per_token_flex": 0.00000152, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-thinking.html", "supports_function_calling": true, "supports_reasoning": true }, "bedrock/sa-east-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-7, + "input_cost_per_token_flex": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.0000036, + "output_cost_per_token_flex": 0.0000018, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" }, "bedrock/sa-east-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.00000144, + "output_cost_per_token_flex": 7.2e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -13171,12 +13405,14 @@ }, "bedrock/us-east-1/deepseek.v3.2": { "input_cost_per_token": 6.2e-7, + "input_cost_per_token_flex": 3.1e-7, "litellm_provider": "bedrock", "max_input_tokens": 163840, - "max_output_tokens": 163840, - "max_tokens": 163840, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000185, + "output_cost_per_token_flex": 9.25e-7, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -13184,47 +13420,56 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" }, "bedrock/us-east-1/meta.llama3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 0.00000265, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.0000035 + "output_cost_per_token": 0.0000035, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-70b-instruct.md" }, "bedrock/us-east-1/meta.llama3-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 6e-7 + "output_cost_per_token": 6e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-8b-instruct.md" }, "bedrock/us-east-1/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-7, + "input_cost_per_token_flex": 1.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/us-east-1/minimax.minimax-m2.5": { "input_cost_per_token": 3e-7, + "input_cost_per_token_flex": 1.5e-7, "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", @@ -13234,74 +13479,86 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html" }, "bedrock/us-east-1/mistral.mistral-7b-instruct-v0:2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 1.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mistral-7b-instruct.md", "supports_tool_choice": true }, "bedrock/us-east-1/mistral.mistral-large-2402-v1:0": { "input_cost_per_token": 0.000004, "litellm_provider": "bedrock", "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 0.000012, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mistral-large.md", "supports_function_calling": true }, "bedrock/us-east-1/mistral.mixtral-8x7b-instruct-v0:1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 4.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 7e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mixtral-8x7b-instruct.md", "supports_tool_choice": true }, "bedrock/us-east-1/moonshotai.kimi-k2-thinking": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.0000025, + "output_cost_per_token_flex": 0.00000125, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-thinking.html", "supports_function_calling": true, "supports_reasoning": true }, "bedrock/us-east-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.000003, + "output_cost_per_token_flex": 0.0000015, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" }, "bedrock/us-east-1/qwen.qwen3-coder-next": { "input_cost_per_token": 5e-7, + "input_cost_per_token_flex": 2.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -13330,12 +13587,14 @@ }, "bedrock/us-east-2/deepseek.v3.2": { "input_cost_per_token": 6.2e-7, + "input_cost_per_token_flex": 3.1e-7, "litellm_provider": "bedrock", "max_input_tokens": 163840, - "max_output_tokens": 163840, - "max_tokens": 163840, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000185, + "output_cost_per_token_flex": 9.25e-7, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -13343,32 +13602,36 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" }, "bedrock/us-east-2/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-7, + "input_cost_per_token_flex": 1.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/us-east-2/minimax.minimax-m2.5": { "input_cost_per_token": 3e-7, + "input_cost_per_token_flex": 1.5e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -13376,43 +13639,52 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "output_cost_per_token": 0.0000012 + "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7 }, "bedrock/us-east-2/moonshotai.kimi-k2-thinking": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.0000025, + "output_cost_per_token_flex": 0.00000125, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-thinking.html", "supports_function_calling": true, "supports_reasoning": true }, "bedrock/us-east-2/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.000003, + "output_cost_per_token_flex": 0.0000015, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" }, "bedrock/us-east-2/qwen.qwen3-coder-next": { "input_cost_per_token": 5e-7, + "input_cost_per_token_flex": 2.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -13553,6 +13825,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -13586,6 +13859,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -13711,6 +13985,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -13747,6 +14022,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -13804,6 +14080,7 @@ "supports_pdf_input": true }, "bedrock/us-gov-east-1/nvidia.nemotron-nano-12b-v2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 2.4e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -13811,6 +14088,7 @@ "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 7.2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-nvidia-nvidia-nemotron-nano-12b-v2-vl-bf16.html", "supports_system_messages": true, "supports_vision": true, "supports_audio_input": false, @@ -13835,6 +14113,7 @@ "supports_vision": false }, "bedrock/us-gov-east-1/nvidia.nemotron-nano-9b-v2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 7.2e-8, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -13842,6 +14121,7 @@ "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 2.76e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-nvidia-nvidia-nemotron-nano-9b-v2.html", "supports_system_messages": true, "supports_audio_input": false, "supports_function_calling": true, @@ -13869,14 +14149,16 @@ "input_cost_per_token": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 7.2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-120b.html", "supported_endpoints": [ "/v1/chat/completions" ], "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, @@ -13886,14 +14168,16 @@ "input_cost_per_token": 8.4e-8, "litellm_provider": "bedrock", "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 3.6e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-20b.html", "supported_endpoints": [ "/v1/chat/completions" ], "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, @@ -13946,13 +14230,13 @@ "supports_tool_choice": true }, "bedrock/us-gov-west-1/amazon.titan-text-express-v1": { - "input_cost_per_token": 0.0000013, + "input_cost_per_token": 8e-7, "litellm_provider": "bedrock", "max_input_tokens": 42000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", - "output_cost_per_token": 0.0000017 + "output_cost_per_token": 0.0000016 }, "bedrock/us-gov-west-1/amazon.titan-text-lite-v1": { "input_cost_per_token": 3e-7, @@ -14061,6 +14345,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -14094,6 +14379,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -14219,6 +14505,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -14255,6 +14542,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -14312,6 +14600,7 @@ "supports_pdf_input": true }, "bedrock/us-gov-west-1/nvidia.nemotron-nano-12b-v2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 2.4e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -14319,6 +14608,7 @@ "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 7.2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-nvidia-nvidia-nemotron-nano-12b-v2-vl-bf16.html", "supports_system_messages": true, "supports_vision": true, "supports_audio_input": false, @@ -14343,6 +14633,7 @@ "supports_vision": false }, "bedrock/us-gov-west-1/nvidia.nemotron-nano-9b-v2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 7.2e-8, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -14350,6 +14641,7 @@ "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 2.76e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-nvidia-nvidia-nemotron-nano-9b-v2.html", "supports_system_messages": true, "supports_audio_input": false, "supports_function_calling": true, @@ -14377,14 +14669,16 @@ "input_cost_per_token": 1.8e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 7.2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-120b.html", "supported_endpoints": [ "/v1/chat/completions" ], "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, @@ -14394,36 +14688,42 @@ "input_cost_per_token": 8.4e-8, "litellm_provider": "bedrock", "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 3.6e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-20b.html", "supported_endpoints": [ "/v1/chat/completions" ], "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true }, "bedrock/us-west-1/meta.llama3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 0.00000265, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.0000035 + "output_cost_per_token": 0.0000035, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-70b-instruct.md" }, "bedrock/us-west-1/meta.llama3-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 6e-7 + "output_cost_per_token": 6e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-8b-instruct.md" }, "bedrock/us-west-2/1-month-commitment/anthropic.claude-instant-v1": { "cost_per_second": 0.011, @@ -14515,12 +14815,14 @@ }, "bedrock/us-west-2/deepseek.v3.2": { "input_cost_per_token": 6.2e-7, + "input_cost_per_token_flex": 3.1e-7, "litellm_provider": "bedrock", "max_input_tokens": 163840, - "max_output_tokens": 163840, - "max_tokens": 163840, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000185, + "output_cost_per_token_flex": 9.25e-7, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -14528,29 +14830,34 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" }, "bedrock/us-west-2/minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-7, + "input_cost_per_token_flex": 1.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" }, "bedrock/us-west-2/minimax.minimax-m2.5": { "input_cost_per_token": 3e-7, + "input_cost_per_token_flex": 1.5e-7, "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, "litellm_provider": "bedrock", - "max_input_tokens": 1000000, + "max_input_tokens": 196000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", @@ -14560,74 +14867,86 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html" }, "bedrock/us-west-2/mistral.mistral-7b-instruct-v0:2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 1.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mistral-7b-instruct.md", "supports_tool_choice": true }, "bedrock/us-west-2/mistral.mistral-large-2402-v1:0": { "input_cost_per_token": 0.000004, "litellm_provider": "bedrock", "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 0.000012, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mistral-large.md", "supports_function_calling": true }, "bedrock/us-west-2/mistral.mixtral-8x7b-instruct-v0:1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 4.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 7e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mixtral-8x7b-instruct.md", "supports_tool_choice": true }, "bedrock/us-west-2/moonshotai.kimi-k2-thinking": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.0000025, + "output_cost_per_token_flex": 0.00000125, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-thinking.html", "supports_function_calling": true, "supports_reasoning": true }, "bedrock/us-west-2/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-7, + "input_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.000003, + "output_cost_per_token_flex": 0.0000015, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" }, "bedrock/us-west-2/qwen.qwen3-coder-next": { "input_cost_per_token": 5e-7, + "input_cost_per_token_flex": 2.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -14828,28 +15147,42 @@ "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json" }, "bedrock_mantle/deepseek.v3.1": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 5.8e-7, + "input_cost_per_token_batches": 2.9e-7, + "input_cost_per_token_flex": 2.9e-7, + "input_cost_per_token_priority": 0.000001015, "output_cost_per_token": 0.00000168, + "output_cost_per_token_flex": 8.4e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 128000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", + "output_cost_per_token_batches": 8.4e-7, + "output_cost_per_token_priority": 0.00000294, "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_tool_choice": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json" }, "bedrock_mantle/google.gemma-4-26b-a4b": { "input_cost_per_token": 1.3e-7, + "input_cost_per_token_batches": 6.5e-8, + "input_cost_per_token_flex": 6.5e-8, + "input_cost_per_token_priority": 2.275e-7, "output_cost_per_token": 4e-7, + "output_cost_per_token_flex": 2e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 256000, "max_output_tokens": 256000, "max_tokens": 256000, "mode": "chat", + "output_cost_per_token_batches": 2e-7, + "output_cost_per_token_priority": 7e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "use_openai_responses_path": true, "supported_endpoints": [ "/v1/chat/completions", @@ -14863,12 +15196,19 @@ }, "bedrock_mantle/google.gemma-4-31b": { "input_cost_per_token": 1.4e-7, + "input_cost_per_token_batches": 7e-8, + "input_cost_per_token_flex": 7e-8, + "input_cost_per_token_priority": 2.45e-7, "output_cost_per_token": 4e-7, + "output_cost_per_token_flex": 2e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 256000, "max_output_tokens": 256000, "max_tokens": 256000, "mode": "chat", + "output_cost_per_token_batches": 2e-7, + "output_cost_per_token_priority": 7e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "use_openai_responses_path": true, "supported_endpoints": [ "/v1/chat/completions", @@ -14882,12 +15222,19 @@ }, "bedrock_mantle/google.gemma-4-e2b": { "input_cost_per_token": 4e-8, + "input_cost_per_token_batches": 2e-8, + "input_cost_per_token_flex": 2e-8, + "input_cost_per_token_priority": 7e-8, "output_cost_per_token": 8e-8, + "output_cost_per_token_flex": 4e-8, "litellm_provider": "bedrock_mantle", "max_input_tokens": 128000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", + "output_cost_per_token_batches": 4e-8, + "output_cost_per_token_priority": 1.4e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "use_openai_responses_path": true, "supported_endpoints": [ "/v1/chat/completions", @@ -14900,20 +15247,27 @@ "supports_vision": true }, "bedrock_mantle/moonshotai.kimi-k2-thinking": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 6e-7, + "input_cost_per_token_batches": 3e-7, + "input_cost_per_token_flex": 3e-7, + "input_cost_per_token_priority": 0.00000105, "output_cost_per_token": 0.0000025, + "output_cost_per_token_flex": 0.00000125, "litellm_provider": "bedrock_mantle", "max_input_tokens": 256000, "max_output_tokens": 16000, "max_tokens": 16000, "mode": "chat", + "output_cost_per_token_batches": 0.00000125, + "output_cost_per_token_priority": 0.00000438, "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_tool_choice": true, "supports_reasoning": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-thinking.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json" }, "bedrock_mantle/openai.gpt-5.4": { "input_cost_per_token": 0.00000275, @@ -15203,7 +15557,7 @@ "supports_function_calling": true, "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, - "supports_none_reasoning_effort": false, + "supports_none_reasoning_effort": true, "supports_prompt_caching": true, "supports_prompt_cache_breakpoint": true, "supports_reasoning": true, @@ -15242,7 +15596,7 @@ "supports_function_calling": true, "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, - "supports_none_reasoning_effort": false, + "supports_none_reasoning_effort": true, "supports_prompt_caching": true, "supports_prompt_cache_breakpoint": true, "supports_reasoning": true, @@ -15255,17 +15609,25 @@ "bedrock_mantle/openai.gpt-6.1-sol": { "cache_creation_input_token_cost": 0.00000275, "cache_creation_input_token_cost_above_272k_tokens": 0.0000055, + "cache_creation_input_token_cost_above_272k_tokens_ultrafast": 0.000033, + "cache_creation_input_token_cost_ultrafast": 0.0000165, "cache_read_input_token_cost": 1.1e-7, "cache_read_input_token_cost_above_272k_tokens": 2.2e-7, + "cache_read_input_token_cost_above_272k_tokens_ultrafast": 0.00000132, + "cache_read_input_token_cost_ultrafast": 6.6e-7, "input_cost_per_token": 0.0000022, "input_cost_per_token_above_272k_tokens": 0.0000044, + "input_cost_per_token_above_272k_tokens_ultrafast": 0.0000264, + "input_cost_per_token_ultrafast": 0.0000132, "litellm_provider": "bedrock_mantle", - "max_input_tokens": 1050000, + "max_input_tokens": 1000000, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "responses", "output_cost_per_token": 0.000011, "output_cost_per_token_above_272k_tokens": 0.0000165, + "output_cost_per_token_above_272k_tokens_ultrafast": 0.000099, + "output_cost_per_token_ultrafast": 0.000066, "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-sol.html", "supported_endpoints": [ "/v1/chat/completions", @@ -15293,12 +15655,19 @@ }, "bedrock_mantle/openai.gpt-oss-120b": { "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7.5e-8, + "input_cost_per_token_flex": 7.5e-8, + "input_cost_per_token_priority": 2.625e-7, "output_cost_per_token": 6e-7, + "output_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_priority": 0.00000105, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -15311,12 +15680,19 @@ }, "bedrock_mantle/openai.gpt-oss-20b": { "input_cost_per_token": 7e-8, + "input_cost_per_token_batches": 3.5e-8, + "input_cost_per_token_flex": 3.5e-8, + "input_cost_per_token_priority": 1.225e-7, "output_cost_per_token": 3e-7, + "output_cost_per_token_flex": 1.5e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", + "output_cost_per_token_batches": 1.5e-7, + "output_cost_per_token_priority": 5.25e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -15329,12 +15705,19 @@ }, "bedrock_mantle/openai.gpt-oss-safeguard-120b": { "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7e-8, + "input_cost_per_token_flex": 7e-8, + "input_cost_per_token_priority": 2.6e-7, "output_cost_per_token": 6e-7, + "output_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 131072, - "max_output_tokens": 65536, - "max_tokens": 65536, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_priority": 0.00000105, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supported_endpoints": [ "/v1/chat/completions" ], @@ -15345,12 +15728,19 @@ }, "bedrock_mantle/openai.gpt-oss-safeguard-20b": { "input_cost_per_token": 7e-8, + "input_cost_per_token_batches": 3e-8, + "input_cost_per_token_flex": 3e-8, + "input_cost_per_token_priority": 1.2e-7, "output_cost_per_token": 2e-7, + "output_cost_per_token_flex": 1e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 131072, - "max_output_tokens": 65536, - "max_tokens": 65536, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", + "output_cost_per_token_batches": 1e-7, + "output_cost_per_token_priority": 3.5e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supported_endpoints": [ "/v1/chat/completions" ], @@ -15361,61 +15751,85 @@ }, "bedrock_mantle/qwen.qwen3-235b-a22b-2507": { "input_cost_per_token": 2.2e-7, + "input_cost_per_token_batches": 1.1e-7, + "input_cost_per_token_flex": 1.1e-7, + "input_cost_per_token_priority": 3.85e-7, "output_cost_per_token": 8.8e-7, + "output_cost_per_token_flex": 4.4e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 256000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", + "output_cost_per_token_batches": 4.4e-7, + "output_cost_per_token_priority": 0.00000154, "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_tool_choice": true, "supports_reasoning": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-235b-a22b-2507.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json" }, "bedrock_mantle/qwen.qwen3-32b": { "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7.5e-8, + "input_cost_per_token_flex": 7.5e-8, + "input_cost_per_token_priority": 2.625e-7, "output_cost_per_token": 6e-7, + "output_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 32000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_priority": 0.00000105, "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_tool_choice": true, "supports_reasoning": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-32b.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json" }, "bedrock_mantle/qwen.qwen3-coder-30b-a3b-instruct": { "deprecation_date": "2027-03-30", "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7.5e-8, + "input_cost_per_token_flex": 7.5e-8, + "input_cost_per_token_priority": 2.625e-7, "output_cost_per_token": 6e-7, + "output_cost_per_token_flex": 3e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 256000, "max_output_tokens": 16000, "max_tokens": 16000, "mode": "chat", + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_priority": 0.00000105, "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_tool_choice": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-30b-a3b-instruct.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json" }, "bedrock_mantle/qwen.qwen3-coder-480b-a35b-instruct": { "deprecation_date": "2027-03-30", "input_cost_per_token": 4.5e-7, + "input_cost_per_token_batches": 2.25e-7, + "input_cost_per_token_flex": 2.25e-7, + "input_cost_per_token_priority": 7.875e-7, "output_cost_per_token": 0.0000018, + "output_cost_per_token_flex": 9e-7, "litellm_provider": "bedrock_mantle", - "max_input_tokens": 128000, + "max_input_tokens": 262144, "max_output_tokens": 16000, "max_tokens": 16000, "mode": "chat", + "output_cost_per_token_batches": 9e-7, + "output_cost_per_token_priority": 0.00000315, "supported_endpoints": [ "/v1/chat/completions" ], @@ -15425,44 +15839,57 @@ }, "bedrock_mantle/qwen.qwen3-next-80b-a3b-instruct": { "input_cost_per_token": 1.4e-7, + "input_cost_per_token_batches": 7e-8, + "input_cost_per_token_flex": 7e-8, + "input_cost_per_token_priority": 2.6e-7, "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 256000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", + "output_cost_per_token_batches": 6e-7, + "output_cost_per_token_priority": 0.0000021, "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_tool_choice": true, "supports_reasoning": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json" }, "bedrock_mantle/qwen.qwen3-vl-235b-a22b-instruct": { "input_cost_per_token": 5.3e-7, + "input_cost_per_token_batches": 2.6e-7, + "input_cost_per_token_flex": 2.6e-7, + "input_cost_per_token_priority": 9.3e-7, "output_cost_per_token": 0.00000266, + "output_cost_per_token_flex": 0.00000133, "litellm_provider": "bedrock_mantle", "max_input_tokens": 256000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", + "output_cost_per_token_batches": 0.00000133, + "output_cost_per_token_priority": 0.00000466, "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_tool_choice": true, "supports_vision": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-vl-235b-a22b.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json" }, "bedrock_mantle/us-gov-east-1/openai.gpt-oss-120b": { "input_cost_per_token": 1.8e-7, "output_cost_per_token": 7.2e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-120b.html", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -15478,9 +15905,10 @@ "output_cost_per_token": 3.6e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-20b.html", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -15758,9 +16186,10 @@ "output_cost_per_token": 7.2e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-120b.html", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -15776,9 +16205,10 @@ "output_cost_per_token": 3.6e-7, "litellm_provider": "bedrock_mantle", "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-20b.html", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -15836,8 +16266,10 @@ "input_cost_per_token": 0.00000125, "output_cost_per_token": 0.0000025, "cache_read_input_token_cost": 2e-7, + "cache_read_input_token_cost_batches": 1e-7, "cache_read_input_token_cost_flex": 1e-7, "cache_read_input_token_cost_priority": 3.5e-7, + "input_cost_per_token_batches": 6.25e-7, "input_cost_per_token_flex": 6.25e-7, "input_cost_per_token_priority": 0.0000021875, "litellm_provider": "bedrock_mantle", @@ -15845,6 +16277,7 @@ "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", + "output_cost_per_token_batches": 0.00000125, "output_cost_per_token_flex": 0.00000125, "output_cost_per_token_priority": 0.000004375, "supported_endpoints": [ @@ -15856,7 +16289,7 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json" }, "bedrock_mantle/xai.grok-4.6": { "use_openai_responses_path": true, @@ -16270,6 +16703,7 @@ "supports_xhigh_reasoning_effort": true }, "claude-fable-5": { + "supports_web_search": true, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 0.0000125, "cache_creation_input_token_cost_above_1hr": 0.00002, @@ -16314,6 +16748,7 @@ "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "claude-fable-5-1": { + "supports_web_search": true, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 0.0000125, "cache_creation_input_token_cost_above_1hr": 0.00002, @@ -16387,6 +16822,7 @@ "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "claude-haiku-4-5-20251001": { + "supports_web_search": true, "cache_creation_input_token_cost": 0.00000125, "cache_creation_input_token_cost_above_1hr": 0.000002, "cache_creation_input_token_cost_batches": 6.25e-7, @@ -16633,6 +17069,7 @@ "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "claude-opus-4-5-20251101": { + "supports_web_search": true, "cache_creation_input_token_cost": 0.00000625, "cache_creation_input_token_cost_above_1hr": 0.00001, "cache_creation_input_token_cost_batches": 0.000003125, @@ -16667,6 +17104,7 @@ "prompt_cache_min_tokens": 4096 }, "claude-opus-4-6": { + "supports_web_search": true, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 0.00000625, "cache_creation_input_token_cost_above_1hr": 0.00001, @@ -16752,6 +17190,7 @@ "prompt_cache_min_tokens": 4096 }, "claude-opus-4-7": { + "supports_web_search": true, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 0.00000625, "cache_creation_input_token_cost_above_1hr": 0.00001, @@ -16839,6 +17278,7 @@ "prompt_cache_min_tokens": 2048 }, "claude-opus-4-8": { + "supports_web_search": true, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 0.00000625, "cache_creation_input_token_cost_above_1hr": 0.00001, @@ -16886,6 +17326,7 @@ "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "claude-opus-5": { + "supports_web_search": true, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 0.00000625, "cache_creation_input_token_cost_above_1hr": 0.00001, @@ -16933,6 +17374,7 @@ "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "claude-opus-5-5": { + "supports_web_search": true, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 0.000005, "cache_creation_input_token_cost_above_1hr": 0.000008, @@ -16983,27 +17425,18 @@ "claude-sonnet-4-5": { "cache_creation_input_token_cost": 0.00000375, "cache_creation_input_token_cost_above_1hr": 0.000006, - "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.000012, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, - "input_cost_per_token_above_200k_tokens": 0.000006, - "output_cost_per_token_above_200k_tokens": 0.0000225, - "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, - "cache_creation_input_token_cost_above_200k_tokens_batches": 0.00000375, "cache_creation_input_token_cost_batches": 0.000001875, - "cache_read_input_token_cost_above_200k_tokens": 6e-7, - "cache_read_input_token_cost_above_200k_tokens_batches": 3e-7, "cache_read_input_token_cost_batches": 1.5e-7, "deprecation_date": "2026-11-30", - "input_cost_per_token_above_200k_tokens_batches": 0.000003, "input_cost_per_token_batches": 0.0000015, "litellm_provider": "anthropic", - "max_input_tokens": 1000000, + "max_input_tokens": 200000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", "output_cost_per_token": 0.000015, - "output_cost_per_token_above_200k_tokens_batches": 0.00001125, "output_cost_per_token_batches": 0.0000075, "search_context_cost_per_query": { "search_context_size_high": 0.01, @@ -17026,27 +17459,18 @@ "claude-sonnet-4-5-20250929": { "cache_creation_input_token_cost": 0.00000375, "cache_creation_input_token_cost_above_1hr": 0.000006, - "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.000012, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, - "input_cost_per_token_above_200k_tokens": 0.000006, - "output_cost_per_token_above_200k_tokens": 0.0000225, - "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, - "cache_creation_input_token_cost_above_200k_tokens_batches": 0.00000375, "cache_creation_input_token_cost_batches": 0.000001875, - "cache_read_input_token_cost_above_200k_tokens": 6e-7, - "cache_read_input_token_cost_above_200k_tokens_batches": 3e-7, "cache_read_input_token_cost_batches": 1.5e-7, "deprecation_date": "2026-11-30", - "input_cost_per_token_above_200k_tokens_batches": 0.000003, "input_cost_per_token_batches": 0.0000015, "litellm_provider": "anthropic", - "max_input_tokens": 1000000, + "max_input_tokens": 200000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", "output_cost_per_token": 0.000015, - "output_cost_per_token_above_200k_tokens_batches": 0.00001125, "output_cost_per_token_batches": 0.0000075, "search_context_cost_per_query": { "search_context_size_high": 0.01, @@ -17099,6 +17523,7 @@ "output_cost_per_token_batches": 0.0000075 }, "claude-sonnet-4-6": { + "supports_web_search": true, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 0.00000375, "cache_creation_input_token_cost_above_1hr": 0.000006, @@ -17121,7 +17546,7 @@ }, "supports_adaptive_thinking": true, "supports_legacy_thinking": true, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -17141,6 +17566,7 @@ "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "claude-sonnet-5": { + "supports_web_search": true, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 0.0000025, "cache_creation_input_token_cost_above_1hr": 0.000004, @@ -17724,6 +18150,174 @@ "supports_vision": true, "source": "https://platform.openai.com/docs/models/computer-use-preview" }, + "computer-use-preview-2025-03-11": { + "deprecation_date": "2026-07-23", + "input_cost_per_token": 0.000003, + "litellm_provider": "openai", + "max_input_tokens": 8192, + "max_output_tokens": 1024, + "max_tokens": 1024, + "mode": "chat", + "output_cost_per_token": 0.000012, + "supported_endpoints": [ + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": false, + "supports_reasoning": true, + "supports_response_schema": false, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true, + "source": "https://developers.openai.com/api/docs/models/computer-use-preview" + }, + "coralbricks/deepseek-v4.1-flash": { + "cache_creation_input_token_cost": 9e-8, + "cache_read_input_token_cost": 0, + "input_cost_per_token": 1e-8, + "litellm_provider": "coralbricks", + "max_input_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://www.coralbricks.ai/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "coralbricks/deepseek-v4.1-flash-fast": { + "cache_creation_input_token_cost": 9e-8, + "cache_read_input_token_cost": 0, + "input_cost_per_token": 1e-8, + "litellm_provider": "coralbricks", + "max_input_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://www.coralbricks.ai/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "coralbricks/deepseek-v4.1-flash-fast-fp4": { + "cache_creation_input_token_cost": 9e-8, + "cache_read_input_token_cost": 0, + "input_cost_per_token": 1e-8, + "litellm_provider": "coralbricks", + "max_input_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://www.coralbricks.ai/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "coralbricks/glm-5.3": { + "cache_creation_input_token_cost": 0.00000168, + "cache_read_input_token_cost": 0, + "input_cost_per_token": 0.00000112, + "litellm_provider": "coralbricks", + "max_input_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.0000044, + "source": "https://www.coralbricks.ai/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "coralbricks/glm-5.3-fast": { + "cache_creation_input_token_cost": 0.00000168, + "cache_read_input_token_cost": 0, + "input_cost_per_token": 0.00000112, + "litellm_provider": "coralbricks", + "max_input_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.0000044, + "source": "https://www.coralbricks.ai/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "coralbricks/glm-5.3-fp4": { + "cache_creation_input_token_cost": 0.00000168, + "cache_read_input_token_cost": 0, + "input_cost_per_token": 0.00000112, + "litellm_provider": "coralbricks", + "max_input_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.0000044, + "source": "https://www.coralbricks.ai/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, "crusoe/Qwen/Qwen3-235B-A22B-Instruct-2507": { "input_cost_per_token": 0.000003, "litellm_provider": "crusoe", @@ -18982,7 +19576,7 @@ "output_cost_per_token": 0.000025000010000000002, "output_dbu_cost_per_token": 0.000357143, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_function_calling": true, "supports_legacy_thinking": true, "supports_prompt_caching": true, @@ -19197,7 +19791,7 @@ "output_cost_per_token": 0.000015000020000000002, "output_dbu_cost_per_token": 0.000214286, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_function_calling": true, "supports_legacy_thinking": true, "supports_prompt_caching": true, @@ -19240,6 +19834,7 @@ "input_cost_per_token": 1.4e-7, "input_dbu_cost_per_token": 0.000002, "litellm_provider": "databricks", + "token_count_framing": 97, "max_input_tokens": 1000000, "max_output_tokens": 393216, "max_tokens": 393216, @@ -20135,6 +20730,7 @@ "input_cost_per_token": 0.00000299999, "input_dbu_cost_per_token": 0.000042857, "litellm_provider": "databricks", + "token_count_framing": 100, "max_input_tokens": 1000000, "max_output_tokens": 1048576, "max_tokens": 1048576, @@ -20262,6 +20858,16 @@ "supports_vision": false, "thinking_always_on": true }, + "databricks/system.ai.deepseek-v4-flash-0731": { + "litellm_provider": "databricks", + "mode": "chat", + "token_count_framing": 97 + }, + "databricks/system.ai.kimi-k3": { + "litellm_provider": "databricks", + "mode": "chat", + "token_count_framing": 100 + }, "daybreak-blue-latest": { "cache_creation_input_token_cost": 0.000005, "cache_creation_input_token_cost_above_272k_tokens": 0.00001, @@ -21325,6 +21931,40 @@ "supports_vision": false, "source": "https://deepinfra.com/pricing" }, + "deepinfra/deepseek-ai/DeepSeek-V4.1-Flash": { + "max_tokens": 131072, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 4.2e-7, + "cache_read_input_token_cost": 4.2e-9, + "supports_prompt_caching": true, + "litellm_provider": "deepinfra", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_reasoning": true, + "supports_vision": true, + "source": "https://deepinfra.com/deepseek-ai/DeepSeek-V4.1-Flash" + }, + "deepinfra/deepseek-ai/DeepSeek-V4.1-Flash-1017": { + "max_tokens": 131072, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 4.2e-7, + "cache_read_input_token_cost": 4.2e-9, + "supports_prompt_caching": true, + "litellm_provider": "deepinfra", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_reasoning": true, + "supports_vision": true, + "source": "https://deepinfra.com/deepseek-ai/DeepSeek-V4.1-Flash-1017" + }, "deepinfra/google/gemini-2.5-flash": { "max_tokens": 1000000, "max_input_tokens": 1000000, @@ -22463,14 +23103,15 @@ "supports_tool_choice": false }, "deepseek.v3-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 5.8e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 163840, - "max_output_tokens": 81920, - "max_tokens": 81920, + "max_input_tokens": 128000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000168, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-1.html", "supports_function_calling": true, "supports_reasoning": true, "supports_tool_choice": true, @@ -22478,17 +23119,23 @@ }, "deepseek.v3.2": { "input_cost_per_token": 6.2e-7, + "input_cost_per_token_batches": 3.1e-7, + "input_cost_per_token_flex": 3.1e-7, + "input_cost_per_token_priority": 0.000001085, "litellm_provider": "bedrock_converse", "max_input_tokens": 164000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000185, + "output_cost_per_token_batches": 9.25e-7, + "output_cost_per_token_flex": 9.25e-7, + "output_cost_per_token_priority": 0.0000032375, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, "supports_tool_choice": true, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false @@ -22849,14 +23496,14 @@ "supports_vision": true }, "eu.amazon.nova-2-lite-v1:0": { - "cache_read_input_token_cost": 8.25e-8, - "input_cost_per_token": 3.3e-7, + "cache_read_input_token_cost": 1.0725e-7, + "input_cost_per_token": 4.29e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 1000000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", - "output_cost_per_token": 0.00000275, + "output_cost_per_token": 0.000003597, "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -22919,13 +23566,16 @@ }, "eu.amazon.nova-pro-v1:0": { "cache_read_input_token_cost": 2.625e-7, + "cache_read_input_token_cost_flex": 1.3125e-7, "input_cost_per_token": 0.00000105, + "input_cost_per_token_flex": 5.25e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 300000, "max_output_tokens": 10000, "max_tokens": 10000, "mode": "chat", "output_cost_per_token": 0.0000042, + "output_cost_per_token_flex": 0.0000021, "source": "https://aws.amazon.com/bedrock/pricing/", "supports_function_calling": true, "supports_pdf_input": true, @@ -23052,7 +23702,7 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": true, + "supports_native_structured_output": false, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -23164,6 +23814,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -23365,6 +24016,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -23460,7 +24112,7 @@ "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, "cache_read_input_token_cost_above_200k_tokens": 6e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "max_input_tokens": 200000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", @@ -23493,6 +24145,7 @@ "cache_creation_input_token_cost_above_200k_tokens": 0.00000825, "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.0000132, "cache_read_input_token_cost_above_200k_tokens": 6.6e-7, + "deprecation_date": "2027-04-08", "litellm_provider": "bedrock_converse", "supports_tool_search": true, "max_input_tokens": 200000, @@ -23540,7 +24193,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -23588,6 +24241,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -23627,6 +24281,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -23638,12 +24293,14 @@ }, "eu.deepseek.v3.2": { "input_cost_per_token": 7.4e-7, + "input_cost_per_token_flex": 3.7e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 163840, "max_output_tokens": 163840, "max_tokens": 163840, "mode": "chat", "output_cost_per_token": 0.00000222, + "output_cost_per_token_flex": 0.00000111, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -23653,6 +24310,7 @@ "supports_vision": false }, "eu.meta.llama3-2-1b-instruct-v1:0": { + "deprecation_date": "2026-07-07", "input_cost_per_token": 1.3e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -23660,10 +24318,12 @@ "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 1.3e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-2-1b-instruct.md", "supports_function_calling": true, "supports_tool_choice": false }, "eu.meta.llama3-2-3b-instruct-v1:0": { + "deprecation_date": "2026-07-07", "input_cost_per_token": 1.9e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -23671,6 +24331,7 @@ "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 1.9e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-2-3b-instruct.md", "supports_function_calling": true, "supports_tool_choice": false }, @@ -23678,10 +24339,11 @@ "input_cost_per_token": 0.000002, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.000006, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-pixtral-large.md", "supports_function_calling": true, "supports_tool_choice": false }, @@ -24013,7 +24675,7 @@ "fireworks_ai/accounts/fireworks/models/deepseek-r1": { "input_cost_per_token": 0.000003, "litellm_provider": "fireworks_ai", - "max_input_tokens": 128000, + "max_input_tokens": 163840, "max_output_tokens": 20480, "max_tokens": 20480, "mode": "chat", @@ -24025,7 +24687,7 @@ "fireworks_ai/accounts/fireworks/models/deepseek-r1-0528": { "input_cost_per_token": 0.000003, "litellm_provider": "fireworks_ai", - "max_input_tokens": 160000, + "max_input_tokens": 163840, "max_output_tokens": 160000, "max_tokens": 160000, "mode": "chat", @@ -24046,7 +24708,7 @@ "fireworks_ai/accounts/fireworks/models/deepseek-r1-basic": { "input_cost_per_token": 5.5e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 128000, + "max_input_tokens": 163840, "max_output_tokens": 20480, "max_tokens": 20480, "mode": "chat", @@ -24130,7 +24792,7 @@ "fireworks_ai/accounts/fireworks/models/deepseek-v3": { "input_cost_per_token": 9e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 128000, + "max_input_tokens": 131072, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", @@ -24305,6 +24967,26 @@ "supports_tool_choice": true, "supports_vision": true }, + "fireworks_ai/accounts/fireworks/models/deepseek-v4p1-flash-1017": { + "cache_read_input_token_cost": 6e-9, + "cache_read_input_token_cost_priority": 7.5e-9, + "input_cost_per_token": 3e-7, + "input_cost_per_token_priority": 3.75e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 393216, + "max_tokens": 393216, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "output_cost_per_token_priority": 0.0000015, + "source": "https://api.fireworks.ai/v1/serverless/models?format=nested", + "supports_function_calling": false, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": false, + "supports_vision": true + }, "fireworks_ai/accounts/fireworks/models/devstral-small-2505": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -24369,7 +25051,7 @@ }, "fireworks_ai/accounts/fireworks/models/ernie-4p5-21b-a3b-pt": { "max_tokens": 4096, - "max_input_tokens": 4096, + "max_input_tokens": 131072, "max_output_tokens": 4096, "input_cost_per_token": 1e-7, "output_cost_per_token": 1e-7, @@ -24378,7 +25060,7 @@ }, "fireworks_ai/accounts/fireworks/models/ernie-4p5-300b-a47b-pt": { "max_tokens": 4096, - "max_input_tokens": 4096, + "max_input_tokens": 131072, "max_output_tokens": 4096, "input_cost_per_token": 1e-7, "output_cost_per_token": 1e-7, @@ -24509,7 +25191,7 @@ "fireworks_ai/accounts/fireworks/models/glm-4p5": { "input_cost_per_token": 5.5e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 128000, + "max_input_tokens": 131072, "max_output_tokens": 96000, "max_tokens": 96000, "mode": "chat", @@ -24523,7 +25205,7 @@ "fireworks_ai/accounts/fireworks/models/glm-4p5-air": { "input_cost_per_token": 2.2e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 128000, + "max_input_tokens": 131072, "max_output_tokens": 96000, "max_tokens": 96000, "mode": "chat", @@ -24548,7 +25230,7 @@ "input_cost_per_token": 5.5e-7, "output_cost_per_token": 0.00000219, "litellm_provider": "fireworks_ai", - "max_input_tokens": 202800, + "max_input_tokens": 202752, "max_output_tokens": 202800, "max_tokens": 202800, "mode": "chat", @@ -24562,7 +25244,7 @@ "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 6e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 202800, + "max_input_tokens": 202752, "max_output_tokens": 202800, "max_tokens": 202800, "mode": "chat", @@ -24577,7 +25259,7 @@ "cache_read_input_token_cost": 2.6e-7, "input_cost_per_token": 0.0000014, "litellm_provider": "fireworks_ai", - "max_input_tokens": 202800, + "max_input_tokens": 202752, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", @@ -25016,7 +25698,7 @@ "fireworks_ai/accounts/fireworks/models/llama-v3p1-405b-instruct": { "input_cost_per_token": 0.000003, "litellm_provider": "fireworks_ai", - "max_input_tokens": 128000, + "max_input_tokens": 131072, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", @@ -25056,7 +25738,7 @@ "fireworks_ai/accounts/fireworks/models/llama-v3p1-8b-instruct": { "input_cost_per_token": 1e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 16384, + "max_input_tokens": 131072, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", @@ -25078,7 +25760,7 @@ "fireworks_ai/accounts/fireworks/models/llama-v3p2-11b-vision-instruct": { "input_cost_per_token": 2e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 16384, + "max_input_tokens": 131072, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", @@ -25101,7 +25783,7 @@ "fireworks_ai/accounts/fireworks/models/llama-v3p2-1b-instruct": { "input_cost_per_token": 1e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 16384, + "max_input_tokens": 131072, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", @@ -25123,7 +25805,7 @@ "fireworks_ai/accounts/fireworks/models/llama-v3p2-3b-instruct": { "input_cost_per_token": 1e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 16384, + "max_input_tokens": 131072, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", @@ -25136,7 +25818,7 @@ "fireworks_ai/accounts/fireworks/models/llama-v3p2-90b-vision-instruct": { "input_cost_per_token": 9e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 16384, + "max_input_tokens": 131072, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", @@ -25208,7 +25890,7 @@ }, "fireworks_ai/accounts/fireworks/models/minimax-m2": { "max_tokens": 4096, - "max_input_tokens": 4096, + "max_input_tokens": 196608, "max_output_tokens": 4096, "input_cost_per_token": 3e-7, "output_cost_per_token": 0.0000012, @@ -25219,7 +25901,7 @@ "cache_read_input_token_cost": 3e-8, "input_cost_per_token": 3e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 204800, + "max_input_tokens": 196608, "max_output_tokens": 204800, "max_tokens": 204800, "mode": "chat", @@ -25449,6 +26131,17 @@ "supports_tool_choice": true, "supports_vision": true }, + "fireworks_ai/accounts/fireworks/models/muse-spark-open": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.00000125, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.00000425, + "source": "https://fireworks.ai/models/fireworks/muse-spark-open", + "supports_function_calling": true, + "supports_vision": true + }, "fireworks_ai/accounts/fireworks/models/mythomax-l2-13b": { "max_tokens": 4096, "max_input_tokens": 4096, @@ -25498,7 +26191,7 @@ }, "fireworks_ai/accounts/fireworks/models/nemotron-nano-v2-12b-vl": { "max_tokens": 4096, - "max_input_tokens": 4096, + "max_input_tokens": 131072, "max_output_tokens": 4096, "input_cost_per_token": 1e-7, "output_cost_per_token": 1e-7, @@ -25561,7 +26254,7 @@ }, "fireworks_ai/accounts/fireworks/models/nvidia-nemotron-nano-12b-v2": { "max_tokens": 131072, - "max_input_tokens": 131072, + "max_input_tokens": 128000, "max_output_tokens": 131072, "input_cost_per_token": 2e-7, "output_cost_per_token": 2e-7, @@ -25570,7 +26263,7 @@ }, "fireworks_ai/accounts/fireworks/models/nvidia-nemotron-nano-9b-v2": { "max_tokens": 131072, - "max_input_tokens": 131072, + "max_input_tokens": 128000, "max_output_tokens": 131072, "input_cost_per_token": 2e-7, "output_cost_per_token": 2e-7, @@ -26184,7 +26877,7 @@ }, "fireworks_ai/accounts/fireworks/models/qwen3-coder-480b-instruct-bf16": { "max_tokens": 4096, - "max_input_tokens": 4096, + "max_input_tokens": 262144, "max_output_tokens": 4096, "input_cost_per_token": 9e-7, "output_cost_per_token": 9e-7, @@ -26256,7 +26949,7 @@ }, "fireworks_ai/accounts/fireworks/models/qwen3-vl-8b-instruct": { "max_tokens": 4096, - "max_input_tokens": 4096, + "max_input_tokens": 262144, "max_output_tokens": 4096, "input_cost_per_token": 2e-7, "output_cost_per_token": 2e-7, @@ -26789,7 +27482,7 @@ "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 6e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 202800, + "max_input_tokens": 202752, "max_output_tokens": 202800, "max_tokens": 202800, "mode": "chat", @@ -26804,7 +27497,7 @@ "cache_read_input_token_cost": 2.6e-7, "input_cost_per_token": 0.0000014, "litellm_provider": "fireworks_ai", - "max_input_tokens": 202800, + "max_input_tokens": 202752, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", @@ -27169,7 +27862,7 @@ "cache_read_input_token_cost": 3e-8, "input_cost_per_token": 3e-7, "litellm_provider": "fireworks_ai", - "max_input_tokens": 204800, + "max_input_tokens": 196608, "max_output_tokens": 204800, "max_tokens": 204800, "mode": "chat", @@ -27535,6 +28228,7 @@ "ft:gpt-4.1-2025-04-14": { "cache_read_input_token_cost": 7.5e-7, "cache_read_input_token_cost_batches": 5e-7, + "deprecation_date": "2027-01-14", "input_cost_per_token": 0.000003, "input_cost_per_token_batches": 0.0000015, "litellm_provider": "openai", @@ -27555,6 +28249,7 @@ "ft:gpt-4.1-mini-2025-04-14": { "cache_read_input_token_cost": 2e-7, "cache_read_input_token_cost_batches": 1e-7, + "deprecation_date": "2027-01-14", "input_cost_per_token": 8e-7, "input_cost_per_token_batches": 4e-7, "litellm_provider": "openai", @@ -27596,6 +28291,7 @@ "ft:gpt-4o-2024-08-06": { "cache_read_input_token_cost": 0.000001875, "cache_read_input_token_cost_batches": 9e-7, + "deprecation_date": "2027-01-14", "input_cost_per_token": 0.00000375, "input_cost_per_token_batches": 0.000002225, "litellm_provider": "openai", @@ -27617,6 +28313,7 @@ }, "ft:gpt-4o-2024-11-20": { "cache_creation_input_token_cost": 0.000001875, + "deprecation_date": "2027-01-14", "input_cost_per_token": 0.00000375, "litellm_provider": "openai", "max_input_tokens": 128000, @@ -27635,6 +28332,7 @@ "ft:gpt-4o-mini-2024-07-18": { "cache_read_input_token_cost": 1.5e-7, "cache_read_input_token_cost_batches": 7.5e-8, + "deprecation_date": "2027-01-14", "input_cost_per_token": 3e-7, "input_cost_per_token_batches": 1.5e-7, "litellm_provider": "openai", @@ -28050,9 +28748,15 @@ "gemini-3-flash-preview": { "cache_read_input_audio_token_cost": 1e-7, "cache_read_input_token_cost": 5e-8, - "cache_read_input_token_cost_flex": 5e-8, + "cache_read_input_token_cost_priority": 9e-8, + "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_audio_token": 0.000001, + "input_cost_per_audio_token_batches": 5e-7, + "input_cost_per_audio_token_priority": 0.0000018, "input_cost_per_token": 5e-7, + "input_cost_per_token_batches": 2.5e-7, + "input_cost_per_token_flex": 2.5e-7, + "input_cost_per_token_priority": 9e-7, "litellm_provider": "vertex_ai-language-models", "max_input_tokens": 1048576, "max_output_tokens": 65535, @@ -28060,6 +28764,14 @@ "mode": "chat", "output_cost_per_reasoning_token": 0.000003, "output_cost_per_token": 0.000003, + "output_cost_per_token_batches": 0.0000015, + "output_cost_per_token_flex": 0.0000015, + "output_cost_per_token_priority": 0.0000054, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", @@ -28077,6 +28789,7 @@ ], "supports_audio_output": false, "supports_function_calling": true, + "supports_native_streaming": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -28087,23 +28800,7 @@ "supports_url_context": true, "supports_vision": true, "supports_web_search": true, - "supports_native_streaming": true, - "input_cost_per_token_priority": 9e-7, - "input_cost_per_audio_token_priority": 0.0000018, - "output_cost_per_token_priority": 0.0000054, - "cache_read_input_token_cost_priority": 9e-8, - "search_context_cost_per_query": { - "search_context_size_low": 0.014, - "search_context_size_medium": 0.014, - "search_context_size_high": 0.014 - }, - "web_search_billing_unit": "per_query", - "google_maps_grounding_cost_per_query": 0.014, - "input_cost_per_audio_token_batches": 5e-7, - "input_cost_per_token_batches": 2.5e-7, - "input_cost_per_token_flex": 2.5e-7, - "output_cost_per_token_batches": 0.0000015, - "output_cost_per_token_flex": 0.0000015 + "web_search_billing_unit": "per_query" }, "gemini-3-pro-image": { "cache_read_input_token_cost": 2e-7, @@ -28465,22 +29162,36 @@ "google_maps_grounding_cost_per_query": 0.014 }, "gemini-3.1-pro-preview": { - "prompt_cache_min_tokens": 4096, + "cache_creation_input_token_cost_above_200k_tokens": 2.5e-7, "cache_read_input_token_cost": 2e-7, "cache_read_input_token_cost_above_200k_tokens": 4e-7, - "cache_creation_input_token_cost_above_200k_tokens": 2.5e-7, + "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-7, + "cache_read_input_token_cost_priority": 3.6e-7, + "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token": 0.000002, "input_cost_per_token_above_200k_tokens": 0.000004, + "input_cost_per_token_above_200k_tokens_priority": 0.0000072, "input_cost_per_token_batches": 0.000001, + "input_cost_per_token_flex": 0.000001, + "input_cost_per_token_priority": 0.0000036, "litellm_provider": "vertex_ai-language-models", "max_input_tokens": 1048576, "max_output_tokens": 65536, "max_tokens": 65536, "mode": "chat", + "output_cost_per_image": 0.00012, "output_cost_per_token": 0.000012, "output_cost_per_token_above_200k_tokens": 0.000018, + "output_cost_per_token_above_200k_tokens_priority": 0.0000324, "output_cost_per_token_batches": 0.000006, - "output_cost_per_image": 0.00012, + "output_cost_per_token_flex": 0.000006, + "output_cost_per_token_priority": 0.0000216, + "prompt_cache_min_tokens": 4096, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", @@ -28498,33 +29209,18 @@ ], "supports_audio_input": true, "supports_function_calling": true, + "supports_native_streaming": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_url_context": true, "supports_video_input": true, "supports_vision": true, "supports_web_search": true, - "supports_url_context": true, - "supports_native_streaming": true, - "input_cost_per_token_priority": 0.0000036, - "input_cost_per_token_above_200k_tokens_priority": 0.0000072, - "output_cost_per_token_priority": 0.0000216, - "output_cost_per_token_above_200k_tokens_priority": 0.0000324, - "cache_read_input_token_cost_priority": 3.6e-7, - "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-7, - "cache_read_input_token_cost_flex": 2e-7, - "search_context_cost_per_query": { - "search_context_size_low": 0.014, - "search_context_size_medium": 0.014, - "search_context_size_high": 0.014 - }, - "web_search_billing_unit": "per_query", - "google_maps_grounding_cost_per_query": 0.014, - "input_cost_per_token_flex": 0.000001, - "output_cost_per_token_flex": 0.000006 + "web_search_billing_unit": "per_query" }, "gemini-3.1-pro-preview-customtools": { "prompt_cache_min_tokens": 4096, @@ -28819,6 +29515,7 @@ }, "gemini-3.8-flash": { "prompt_cache_min_tokens": 4096, + "deprecation_date": "2027-09-02", "cache_read_input_token_cost": 7.5e-8, "cache_read_input_token_cost_batches": 3.75e-8, "cache_read_input_token_cost_flex": 3.75e-8, @@ -29091,10 +29788,12 @@ "gemini-nano-banana-2.1": { "cache_read_input_token_cost": 1.5e-7, "cache_read_input_token_cost_batches": 7.5e-8, + "cache_read_input_token_cost_flex": 7.5e-8, "cache_read_input_token_cost_priority": 2.7e-7, "input_cost_per_image": 0.00168, "input_cost_per_token": 0.0000015, "input_cost_per_token_batches": 7.5e-7, + "input_cost_per_token_flex": 7.5e-7, "input_cost_per_token_priority": 0.0000027, "litellm_provider": "vertex_ai-language-models", "max_input_tokens": 131072, @@ -29105,6 +29804,7 @@ "output_cost_per_image_token": 0.00003, "output_cost_per_token": 0.0000075, "output_cost_per_token_batches": 0.00000375, + "output_cost_per_token_flex": 0.00000375, "output_cost_per_token_priority": 0.0000135, "search_context_cost_per_query": { "search_context_size_low": 0.014, @@ -29267,6 +29967,56 @@ "supports_url_context": true, "supports_vision": true }, + "gemini-robotics-er-2-preview": { + "cache_read_input_token_cost": 1e-7, + "cache_read_input_token_cost_batches": 5e-8, + "input_cost_per_audio_token": 0.000001, + "input_cost_per_token": 0.000001, + "input_cost_per_token_batches": 5e-7, + "litellm_provider": "vertex_ai-language-models", + "max_input_tokens": 131072, + "max_output_tokens": 65536, + "max_tokens": 65536, + "mode": "chat", + "output_cost_per_reasoning_token": 0.000005, + "output_cost_per_token": 0.000005, + "output_cost_per_token_batches": 0.0000025, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "source": "https://ai.google.dev/gemini-api/docs/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_audio_output": false, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "web_search_billing_unit": "per_query" + }, "gemini/deep-research-max-preview-04-2026": { "input_cost_per_token": 0.000002, "input_cost_per_token_batches": 0.000001, @@ -29342,6 +30092,7 @@ } }, "gemini/deep-research-pro-preview-12-2025": { + "deprecation_date": "2026-10-23", "input_cost_per_image": 0.0011, "input_cost_per_token": 0.000002, "input_cost_per_token_batches": 0.000001, @@ -29351,9 +30102,13 @@ "max_tokens": 65536, "mode": "chat", "output_cost_per_token": 0.000012, - "rpm": 1000, - "tpm": 4000000, "output_cost_per_token_batches": 0.000006, + "rpm": 1000, + "search_context_cost_per_query": { + "search_context_size_low": 0.035, + "search_context_size_medium": 0.035, + "search_context_size_high": 0.035 + }, "source": "https://ai.google.dev/gemini-api/docs/models/deep-research-pro-preview-12-2025", "supported_endpoints": [ "/v1/chat/completions", @@ -29370,17 +30125,13 @@ "text" ], "supports_function_calling": false, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_response_schema": true, "supports_system_messages": true, "supports_vision": true, "supports_web_search": true, - "search_context_cost_per_query": { - "search_context_size_low": 0.035, - "search_context_size_medium": 0.035, - "search_context_size_high": 0.035 - }, - "supports_pdf_input": true + "tpm": 4000000 }, "gemini/gemini-2.5-computer-use-preview-10-2025": { "deprecation_date": "2026-07-28", @@ -29476,7 +30227,7 @@ }, "gemini/gemini-2.5-flash-image": { "cache_read_input_token_cost": 3e-8, - "deprecation_date": "2026-10-02", + "deprecation_date": "2027-03-15", "input_cost_per_audio_token": 0.000001, "input_cost_per_token": 3e-7, "input_cost_per_token_batches": 1.5e-7, @@ -29652,6 +30403,7 @@ }, "gemini/gemini-2.5-pro-preview-tts": { "cache_read_input_token_cost": 1.25e-7, + "deprecation_date": "2026-11-17", "input_cost_per_audio_token": 7e-7, "input_cost_per_token": 0.000001, "input_cost_per_token_batches": 5e-7, @@ -29846,7 +30598,6 @@ "supports_reasoning": false }, "gemini/gemini-3.1-flash-image": { - "deprecation_date": "2026-10-29", "input_cost_per_token": 5e-7, "input_cost_per_token_batches": 2.5e-7, "litellm_provider": "gemini", @@ -29860,7 +30611,11 @@ "output_cost_per_token": 0.000003, "output_cost_per_token_batches": 0.0000015, "rpm": 1000, - "tpm": 4000000, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, "source": "https://ai.google.dev/gemini-api/docs/pricing", "supported_endpoints": [ "/v1/chat/completions", @@ -29876,19 +30631,15 @@ "image" ], "supports_function_calling": false, + "supports_pdf_input": true, "supports_prompt_caching": false, "supports_reasoning": true, "supports_response_schema": false, "supports_system_messages": true, "supports_vision": true, "supports_web_search": true, - "search_context_cost_per_query": { - "search_context_size_low": 0.014, - "search_context_size_medium": 0.014, - "search_context_size_high": 0.014 - }, - "web_search_billing_unit": "per_query", - "supports_pdf_input": true + "tpm": 4000000, + "web_search_billing_unit": "per_query" }, "gemini/gemini-3.1-flash-image-preview": { "deprecation_date": "2026-06-25", @@ -30527,6 +31278,66 @@ "web_search_billing_unit": "per_query", "google_maps_grounding_cost_per_query": 0.014 }, + "gemini/gemini-4-argon": { + "cache_read_input_token_cost": 1e-7, + "cache_read_input_token_cost_batches": 5e-8, + "cache_read_input_token_cost_flex": 5e-8, + "cache_read_input_token_cost_priority": 1.8e-7, + "google_maps_grounding_cost_per_query": 0.014, + "input_cost_per_token": 0.000002, + "input_cost_per_token_batches": 0.000001, + "input_cost_per_token_flex": 0.000001, + "input_cost_per_token_priority": 0.0000036, + "litellm_provider": "gemini", + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_reasoning_token": 0.00001, + "output_cost_per_token": 0.00001, + "output_cost_per_token_batches": 0.000005, + "output_cost_per_token_flex": 0.000005, + "output_cost_per_token_priority": 0.000018, + "prompt_cache_min_tokens": 4096, + "rpm": 2000, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "source": "https://ai.google.dev/gemini-api/docs/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_audio_output": false, + "supports_function_calling": true, + "supports_minimal_reasoning_effort": false, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "tpm": 800000, + "web_search_billing_unit": "per_query" + }, "gemini/gemini-exp-1114": { "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, @@ -30827,7 +31638,7 @@ "supports_video_input": true, "supports_vision": true, "tpm": 800000, - "deprecation_date": "2026-09-30" + "deprecation_date": "2026-10-22" }, "gemini/gemini-pro-latest": { "cache_read_input_token_cost": 2e-7, @@ -31307,32 +32118,50 @@ "supports_vision": true }, "github_copilot/claude-fable-5": { + "cache_creation_input_token_cost": 0.0000125, + "cache_read_input_token_cost": 0.000001, + "input_cost_per_token": 0.00001, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.00005, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], + "supports_adaptive_thinking": true, "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true, + "thinking_always_on": true }, "github_copilot/claude-fable-5.1": { + "cache_creation_input_token_cost": 0.0000125, + "cache_read_input_token_cost": 2.5e-7, + "input_cost_per_token": 0.00001, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.00005, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], + "supports_adaptive_thinking": true, "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true, + "thinking_always_on": true }, "github_copilot/claude-haiku-4.5": { "cache_creation_input_token_cost": 0.00000125, @@ -31349,36 +32178,35 @@ ], "supports_function_calling": true, "supports_parallel_function_calling": true, + "supports_reasoning": true, "supports_vision": true }, - "github_copilot/claude-opus-4.5": { + "github_copilot/claude-haiku-5.5": { + "cache_creation_input_token_cost": 1.25e-7, + "cache_creation_input_token_cost_above_100k_tokens": 6.25e-7, + "cache_read_input_token_cost": 1e-8, + "cache_read_input_token_cost_above_100k_tokens": 5e-8, + "input_cost_per_token": 1e-7, + "input_cost_per_token_above_100k_tokens": 5e-7, "litellm_provider": "github_copilot", "max_input_tokens": 128000, "max_output_tokens": 16000, "max_tokens": 16000, "mode": "chat", + "output_cost_per_token": 5e-7, + "output_cost_per_token_above_100k_tokens": 0.0000025, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_vision": true, - "supports_output_config": true - }, - "github_copilot/claude-opus-4.6-fast": { "supports_adaptive_thinking": true, - "supports_legacy_thinking": true, - "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "supported_endpoints": [ - "/v1/chat/completions" - ], "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true, + "thinking_always_on": false }, "github_copilot/claude-opus-4.7": { "litellm_provider": "github_copilot", @@ -31395,115 +32223,163 @@ "supports_vision": true }, "github_copilot/claude-opus-4.8": { + "cache_creation_input_token_cost": 0.00000625, + "cache_read_input_token_cost": 5e-7, + "input_cost_per_token": 0.000005, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.000025, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], + "supports_adaptive_thinking": true, "supports_function_calling": true, "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, "supports_vision": true }, "github_copilot/claude-opus-4.8-fast": { + "cache_creation_input_token_cost": 0.0000125, + "cache_read_input_token_cost": 0.000001, + "input_cost_per_token": 0.00001, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.00005, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], + "supports_adaptive_thinking": true, "supports_function_calling": true, "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, "supports_vision": true }, - "github_copilot/claude-opus-41": { + "github_copilot/claude-opus-5": { + "cache_creation_input_token_cost": 0.00000625, + "cache_read_input_token_cost": 5e-7, + "deprecation_date": "2026-11-16", + "input_cost_per_token": 0.000005, "litellm_provider": "github_copilot", - "max_input_tokens": 80000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", + "output_cost_per_token": 0.000025, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], + "supports_adaptive_thinking": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, "supports_vision": true }, - "github_copilot/claude-opus-5": { + "github_copilot/claude-opus-5.5": { + "cache_creation_input_token_cost": 0.000005, + "cache_read_input_token_cost": 2e-7, + "input_cost_per_token": 0.000004, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.00002, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], + "supports_adaptive_thinking": true, "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true, + "thinking_always_on": true }, - "github_copilot/claude-sonnet-4": { + "github_copilot/claude-sonnet-4.6": { "cache_creation_input_token_cost": 0.00000375, "cache_read_input_token_cost": 3e-7, + "deprecation_date": "2026-09-01", "input_cost_per_token": 0.000003, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", "output_cost_per_token": 0.000015, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_vision": true - }, - "github_copilot/claude-sonnet-4.5": { - "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], + "supports_adaptive_thinking": true, "supports_function_calling": true, + "supports_legacy_thinking": true, "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, "supports_vision": true }, - "github_copilot/claude-sonnet-4.6": { + "github_copilot/claude-sonnet-5": { + "cache_creation_input_token_cost": 0.0000025, + "cache_read_input_token_cost": 2e-7, + "input_cost_per_token": 0.000002, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.00001, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], + "supports_adaptive_thinking": true, "supports_function_calling": true, "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, "supports_vision": true }, - "github_copilot/claude-sonnet-5": { + "github_copilot/claude-sonnet-5.5": { + "cache_creation_input_token_cost": 0.0000025, + "cache_read_input_token_cost": 1e-7, + "input_cost_per_token": 0.000002, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.00001, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ - "/v1/chat/completions" + "/v1/chat/completions", + "/v1/messages" ], + "supports_adaptive_thinking": true, "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true, + "thinking_always_on": true }, "github_copilot/gemini-3.5-flash": { "litellm_provider": "github_copilot", @@ -31528,25 +32404,57 @@ "supports_vision": true }, "github_copilot/gemini-3.7-flash": { + "cache_read_input_token_cost": 7.5e-8, + "input_cost_per_token": 7.5e-7, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, + "max_input_tokens": 200000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.00000375, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, + "supports_reasoning": true, "supports_vision": true }, "github_copilot/gemini-3.8-flash": { + "cache_read_input_token_cost": 7.5e-8, + "input_cost_per_token": 7.5e-7, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "max_input_tokens": 200000, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.00000375, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_vision": true + }, + "github_copilot/gemini-4-argon": { + "cache_read_input_token_cost": 1e-7, + "input_cost_per_token": 0.000002, + "litellm_provider": "github_copilot", + "max_input_tokens": 200000, + "max_output_tokens": 65536, + "max_tokens": 65536, + "mode": "chat", + "output_cost_per_token": 0.00001, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, + "supports_reasoning": true, "supports_vision": true }, "github_copilot/gpt-3.5-turbo": { @@ -31669,21 +32577,6 @@ "supports_function_calling": true, "supports_parallel_function_calling": true }, - "github_copilot/gpt-5": { - "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/responses" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_response_schema": true, - "supports_vision": true - }, "github_copilot/gpt-5-mini": { "cache_read_input_token_cost": 2.5e-8, "input_cost_per_token": 2.5e-7, @@ -31698,36 +32591,6 @@ "supports_response_schema": true, "supports_vision": true }, - "github_copilot/gpt-5.1": { - "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/responses" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_response_schema": true, - "supports_vision": true - }, - "github_copilot/gpt-5.2": { - "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/responses" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_response_schema": true, - "supports_vision": true - }, "github_copilot/gpt-5.3-codex": { "litellm_provider": "github_copilot", "max_input_tokens": 128000, @@ -31745,18 +32608,25 @@ "supports_vision": true }, "github_copilot/gpt-5.4": { + "cache_read_input_token_cost": 2.5e-7, + "cache_read_input_token_cost_above_272k_tokens": 5e-7, + "input_cost_per_token": 0.0000025, + "input_cost_per_token_above_272k_tokens": 0.000005, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.000015, + "output_cost_per_token_above_272k_tokens": 0.0000225, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" ], "supports_function_calling": true, "supports_parallel_function_calling": true, + "supports_reasoning": true, "supports_response_schema": true, "supports_vision": true }, @@ -31903,48 +32773,72 @@ "supports_vision": true }, "github_copilot/grok-4.5": { + "cache_read_input_token_cost": 5e-7, + "cache_read_input_token_cost_above_200k_tokens": 0.000001, + "input_cost_per_token": 0.000002, + "input_cost_per_token_above_200k_tokens": 0.000004, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "max_input_tokens": 500000, + "max_output_tokens": 500000, + "max_tokens": 500000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.000006, + "output_cost_per_token_above_200k_tokens": 0.000012, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_response_schema": true + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true }, "github_copilot/grok-4.6": { + "cache_read_input_token_cost": 5e-7, + "cache_read_input_token_cost_above_200k_tokens": 0.000001, + "input_cost_per_token": 0.000002, + "input_cost_per_token_above_200k_tokens": 0.000004, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "max_input_tokens": 500000, + "max_output_tokens": 500000, + "max_tokens": 500000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.000006, + "output_cost_per_token_above_200k_tokens": 0.000012, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_response_schema": true + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true }, - "github_copilot/kimi-k2.7-code": { + "github_copilot/grok-4.7": { + "cache_read_input_token_cost": 5e-7, + "cache_read_input_token_cost_above_200k_tokens": 0.000001, + "input_cost_per_token": 0.000002, + "input_cost_per_token_above_200k_tokens": 0.000004, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "max_input_tokens": 500000, + "max_output_tokens": 500000, + "max_tokens": 500000, "mode": "chat", - "source": "https://docs.github.com/en/copilot/reference/ai-models/supported-models", + "output_cost_per_token": 0.000006, + "output_cost_per_token_above_200k_tokens": 0.000012, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_response_schema": true + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true }, - "github_copilot/kimi-k3": { + "github_copilot/kimi-k2.7-code": { "litellm_provider": "github_copilot", "max_input_tokens": 128000, "max_output_tokens": 64000, @@ -31958,37 +32852,23 @@ "supports_parallel_function_calling": true, "supports_response_schema": true }, - "github_copilot/mai-code-1-flash": { - "cache_read_input_token_cost": 7.5e-8, - "input_cost_per_token": 7.5e-7, - "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 0.0000045, - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_response_schema": true - }, - "github_copilot/mai-code-1-flash-internal": { - "cache_read_input_token_cost": 7.5e-8, - "input_cost_per_token": 7.5e-7, + "github_copilot/kimi-k3": { + "cache_read_input_token_cost": 3e-7, + "input_cost_per_token": 0.000003, "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "max_input_tokens": 917504, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.0000045, + "output_cost_per_token": 0.000015, + "source": "https://raw.githubusercontent.com/github/docs/main/data/tables/copilot/models-and-pricing.yml", "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_response_schema": true + "supports_response_schema": true, + "supports_vision": true }, "github_copilot/mai-code-1.1-flash": { "litellm_provider": "github_copilot", @@ -32020,13 +32900,19 @@ }, "global.amazon.nova-2-lite-v1:0": { "cache_read_input_token_cost": 7.5e-8, + "cache_read_input_token_cost_priority": 1.3125e-7, "input_cost_per_token": 3e-7, + "input_cost_per_token_batches": 1.5e-7, + "input_cost_per_token_priority": 5.25e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 1000000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", "output_cost_per_token": 0.0000025, + "output_cost_per_token_batches": 0.00000125, + "output_cost_per_token_priority": 0.000004375, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -32066,7 +32952,7 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": true, + "supports_native_structured_output": false, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -32153,6 +33039,10 @@ "cache_read_input_token_cost": 1e-8, "input_cost_per_token": 1e-7, "litellm_provider": "bedrock_converse", + "token_count_model": { + "model": "global.anthropic.claude-opus-5-5", + "region": "us-east-1" + }, "supports_tool_search": true, "max_input_tokens": 1000000, "max_output_tokens": 128000, @@ -32178,6 +33068,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -32359,6 +33250,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -32454,7 +33346,7 @@ "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, "cache_read_input_token_cost_above_200k_tokens": 6e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "max_input_tokens": 200000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", @@ -32487,6 +33379,7 @@ "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.000012, "cache_read_input_token_cost_above_200k_tokens": 6e-7, + "deprecation_date": "2027-04-08", "litellm_provider": "bedrock_converse", "supports_tool_search": true, "max_input_tokens": 200000, @@ -32534,7 +33427,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -32582,6 +33475,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -32596,6 +33490,10 @@ "cache_read_input_token_cost": 1e-7, "input_cost_per_token": 0.000002, "litellm_provider": "bedrock_converse", + "token_count_model": { + "model": "global.anthropic.claude-opus-5-5", + "region": "us-east-1" + }, "supports_tool_search": true, "max_input_tokens": 1000000, "max_output_tokens": 128000, @@ -32621,6 +33519,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -32642,6 +33541,8 @@ "input_cost_per_token_flex": 0.0000015, "input_cost_per_token_priority": 0.00000525, "litellm_provider": "bedrock_converse", + "token_count_framing": 64, + "token_count_tokenizer": "kimi_k3", "max_input_tokens": 1000000, "max_output_tokens": 131072, "max_tokens": 131072, @@ -32668,7 +33569,8 @@ "output_cost_per_token": 0.000015, "output_cost_per_token_above_272k_tokens": 0.0000225, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", @@ -32699,7 +33601,8 @@ "output_cost_per_token": 0.00003, "output_cost_per_token_above_272k_tokens": 0.000045, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", @@ -32733,11 +33636,12 @@ "output_cost_per_token": 0.0000012, "output_cost_per_token_above_272k_tokens": 0.0000018, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-luna.html", "supported_modalities": [ "text", "image" @@ -32769,11 +33673,12 @@ "output_cost_per_token": 0.00002, "output_cost_per_token_above_272k_tokens": 0.00003, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-sol.html", "supported_modalities": [ "text", "image" @@ -32805,11 +33710,12 @@ "output_cost_per_token": 0.000012, "output_cost_per_token_above_272k_tokens": 0.000018, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-terra.html", "supported_modalities": [ "text", "image" @@ -32847,6 +33753,7 @@ "output_cost_per_token": 0.00005, "output_cost_per_token_above_272k_tokens": 0.000075, "litellm_provider": "bedrock_converse", + "supports_bedrock_converse_tool_result_images": false, "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, @@ -32887,6 +33794,7 @@ "output_cost_per_token": 5e-7, "output_cost_per_token_above_272k_tokens": 7.5e-7, "litellm_provider": "bedrock_converse", + "supports_bedrock_converse_tool_result_images": false, "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, @@ -32901,7 +33809,7 @@ "supports_function_calling": true, "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, - "supports_none_reasoning_effort": false, + "supports_none_reasoning_effort": true, "supports_tool_choice": true, "supports_prompt_caching": true, "supports_reasoning": true, @@ -32925,6 +33833,7 @@ "output_cost_per_token": 0.00001, "output_cost_per_token_above_272k_tokens": 0.000015, "litellm_provider": "bedrock_converse", + "supports_bedrock_converse_tool_result_images": false, "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, @@ -32939,7 +33848,7 @@ "supports_function_calling": true, "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, - "supports_none_reasoning_effort": false, + "supports_none_reasoning_effort": true, "supports_tool_choice": true, "supports_prompt_caching": true, "supports_reasoning": true, @@ -32955,17 +33864,26 @@ "global.openai.gpt-6.1-sol": { "cache_creation_input_token_cost": 0.0000025, "cache_creation_input_token_cost_above_272k_tokens": 0.000005, + "cache_creation_input_token_cost_above_272k_tokens_ultrafast": 0.00003, + "cache_creation_input_token_cost_ultrafast": 0.000015, "cache_read_input_token_cost": 1e-7, "cache_read_input_token_cost_above_272k_tokens": 2e-7, + "cache_read_input_token_cost_above_272k_tokens_ultrafast": 0.0000012, + "cache_read_input_token_cost_ultrafast": 6e-7, "input_cost_per_token": 0.000002, "input_cost_per_token_above_272k_tokens": 0.000004, + "input_cost_per_token_above_272k_tokens_ultrafast": 0.000024, + "input_cost_per_token_ultrafast": 0.000012, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1050000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1000000, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", "output_cost_per_token": 0.00001, "output_cost_per_token_above_272k_tokens": 0.000015, + "output_cost_per_token_above_272k_tokens_ultrafast": 0.00009, + "output_cost_per_token_ultrafast": 0.00006, "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-sol.html", "supported_endpoints": [ "/v1/chat/completions", @@ -32998,6 +33916,17 @@ "supports_video_input": true, "source": "https://aws.amazon.com/bedrock/pricing/" }, + "global.twelvelabs.pegasus-1-5-v1:0": { + "input_cost_per_video_per_second": 0.00049, + "litellm_provider": "bedrock", + "max_input_tokens": 261120, + "max_output_tokens": 98304, + "max_tokens": 98304, + "mode": "chat", + "output_cost_per_token": 0.0000075, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-twelvelabs-pegasus-v1-5.html", + "supports_video_input": true + }, "global.xai.grok-4.6": { "supports_regex_lookaround": false, "input_cost_per_token": 0.000002, @@ -33028,6 +33957,11 @@ }, "global.xai.grok-4.7": { "supports_regex_lookaround": false, + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_response_format": true, "cache_read_input_token_cost": 5e-7, "cache_read_input_token_cost_flex": 2.5e-7, "cache_read_input_token_cost_priority": 8.75e-7, @@ -33035,14 +33969,14 @@ "input_cost_per_token_flex": 0.000001, "input_cost_per_token_priority": 0.0000035, "litellm_provider": "bedrock_converse", - "max_input_tokens": 500000, - "max_output_tokens": 500000, - "max_tokens": 500000, + "max_input_tokens": 524288, + "max_output_tokens": 524288, + "max_tokens": 524288, "mode": "chat", "output_cost_per_token": 0.000006, "output_cost_per_token_flex": 0.000003, "output_cost_per_token_priority": 0.0000105, - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-xai-grok-4-7.html", "supports_function_calling": true, "supports_prompt_caching": false, "supports_reasoning": true, @@ -33249,14 +34183,21 @@ "output_cost_per_token": 0.000002 }, "google.gemma-3-12b-it": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 9e-8, + "input_cost_per_token_batches": 5e-8, + "input_cost_per_token_flex": 5e-8, + "input_cost_per_token_priority": 1.6e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 2.9e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", + "output_cost_per_token_batches": 1.5e-7, + "output_cost_per_token_flex": 1.5e-7, + "output_cost_per_token_priority": 5.1e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, "supports_response_schema": true, @@ -33264,14 +34205,21 @@ "supports_vision": true }, "google.gemma-3-27b-it": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 2.3e-7, + "input_cost_per_token_batches": 1.2e-7, + "input_cost_per_token_flex": 1.2e-7, + "input_cost_per_token_priority": 4e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 3.8e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", + "output_cost_per_token_batches": 1.9e-7, + "output_cost_per_token_flex": 1.9e-7, + "output_cost_per_token_priority": 6.7e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, "supports_response_schema": true, @@ -33279,14 +34227,21 @@ "supports_vision": true }, "google.gemma-3-4b-it": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 4e-8, + "input_cost_per_token_batches": 2e-8, + "input_cost_per_token_flex": 2e-8, + "input_cost_per_token_priority": 7e-8, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 8e-8, - "source": "https://aws.amazon.com/bedrock/pricing/", + "output_cost_per_token_batches": 4e-8, + "output_cost_per_token_flex": 4e-8, + "output_cost_per_token_priority": 1.4e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, "supports_system_messages": true, @@ -33485,6 +34440,7 @@ "gpt-4.1-2025-04-14": { "cache_read_input_token_cost": 5e-7, "cache_read_input_token_cost_priority": 8.75e-7, + "deprecation_date": "2027-01-14", "input_cost_per_token": 0.000002, "input_cost_per_token_priority": 0.0000035, "input_cost_per_token_batches": 0.000001, @@ -33571,6 +34527,7 @@ "gpt-4.1-mini-2025-04-14": { "cache_read_input_token_cost": 1e-7, "cache_read_input_token_cost_priority": 1.75e-7, + "deprecation_date": "2027-01-14", "input_cost_per_token": 4e-7, "input_cost_per_token_priority": 7e-7, "input_cost_per_token_batches": 2e-7, @@ -33736,6 +34693,7 @@ "gpt-4o-2024-08-06": { "cache_read_input_token_cost": 0.00000125, "cache_read_input_token_cost_priority": 0.000002125, + "deprecation_date": "2027-01-14", "input_cost_per_token": 0.0000025, "input_cost_per_token_priority": 0.00000425, "input_cost_per_token_batches": 0.00000125, @@ -33760,6 +34718,7 @@ "gpt-4o-2024-11-20": { "cache_read_input_token_cost": 0.00000125, "cache_read_input_token_cost_priority": 0.000002125, + "deprecation_date": "2027-01-14", "input_cost_per_token": 0.0000025, "input_cost_per_token_priority": 0.00000425, "input_cost_per_token_batches": 0.00000125, @@ -33849,6 +34808,7 @@ "gpt-4o-mini-2024-07-18": { "cache_read_input_token_cost": 7.5e-8, "cache_read_input_token_cost_priority": 1.25e-7, + "deprecation_date": "2027-01-14", "input_cost_per_token": 1.5e-7, "input_cost_per_token_priority": 2.5e-7, "input_cost_per_token_batches": 7.5e-8, @@ -33919,6 +34879,34 @@ "supports_vision": true, "supports_web_search": true }, + "gpt-4o-mini-search-preview-2025-03-11": { + "deprecation_date": "2026-07-23", + "cache_read_input_token_cost": 7.5e-8, + "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7.5e-8, + "litellm_provider": "openai", + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 6e-7, + "output_cost_per_token_batches": 3e-7, + "search_context_cost_per_query": { + "search_context_size_high": 0.025, + "search_context_size_low": 0.025, + "search_context_size_medium": 0.025 + }, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_web_search": true, + "source": "https://developers.openai.com/api/docs/models/gpt-4o-mini-search-preview" + }, "gpt-4o-search-preview": { "cache_read_input_token_cost": 0.00000125, "input_cost_per_token": 0.0000025, @@ -33945,6 +34933,34 @@ "supports_vision": true, "supports_web_search": true }, + "gpt-4o-search-preview-2025-03-11": { + "deprecation_date": "2026-07-23", + "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 0.0000025, + "input_cost_per_token_batches": 0.00000125, + "litellm_provider": "openai", + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 0.00001, + "output_cost_per_token_batches": 0.000005, + "search_context_cost_per_query": { + "search_context_size_high": 0.025, + "search_context_size_low": 0.025, + "search_context_size_medium": 0.025 + }, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_web_search": true, + "source": "https://developers.openai.com/api/docs/models/gpt-4o-search-preview" + }, "gpt-5": { "cache_read_input_token_cost": 1.25e-7, "cache_read_input_token_cost_batches": 6.25e-8, @@ -34551,6 +35567,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_none_reasoning_effort": true, @@ -34604,6 +35621,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_none_reasoning_effort": true, @@ -34727,336 +35745,341 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_none_reasoning_effort": true, - "default_reasoning_effort": "none", - "supports_xhigh_reasoning_effort": true, - "supports_minimal_reasoning_effort": false, - "input_cost_per_token_above_272k_tokens_flex": 0.0000025, - "output_cost_per_token_above_272k_tokens_flex": 0.00001125, - "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-7, - "source": "https://developers.openai.com/api/docs/pricing" - }, - "gpt-5.4-2026-03-05": { - "cache_read_input_token_cost": 2.5e-7, - "cache_read_input_token_cost_above_272k_tokens": 5e-7, - "cache_read_input_token_cost_batches": 1.3e-7, - "cache_read_input_token_cost_above_272k_tokens_batches": 2.5e-7, - "cache_read_input_token_cost_flex": 1.3e-7, - "cache_read_input_token_cost_priority": 5e-7, - "input_cost_per_token": 0.0000025, - "input_cost_per_token_above_272k_tokens": 0.000005, - "input_cost_per_token_flex": 0.00000125, - "input_cost_per_token_batches": 0.00000125, - "input_cost_per_token_above_272k_tokens_batches": 0.0000025, - "input_cost_per_token_priority": 0.000005, - "litellm_provider": "openai", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000015, - "output_cost_per_token_above_272k_tokens": 0.0000225, - "output_cost_per_token_flex": 0.0000075, - "output_cost_per_token_batches": 0.0000075, - "output_cost_per_token_above_272k_tokens_batches": 0.00001125, - "output_cost_per_token_priority": 0.00003, - "regional_processing_uplift_multiplier_eu": 1.1, - "regional_processing_uplift_multiplier_us": 1.1, - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/batch", - "/v1/responses" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_none_reasoning_effort": true, - "default_reasoning_effort": "none", - "supports_xhigh_reasoning_effort": true, - "supports_minimal_reasoning_effort": false, - "input_cost_per_token_above_272k_tokens_flex": 0.0000025, - "output_cost_per_token_above_272k_tokens_flex": 0.00001125, - "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-7, - "source": "https://developers.openai.com/api/docs/pricing" - }, - "gpt-5.4-mini": { - "cache_read_input_token_cost": 7.5e-8, - "cache_read_input_token_cost_batches": 3.75e-8, - "cache_read_input_token_cost_flex": 3.75e-8, - "cache_read_input_token_cost_priority": 1.5e-7, - "input_cost_per_token": 7.5e-7, - "input_cost_per_token_flex": 3.75e-7, - "input_cost_per_token_batches": 3.75e-7, - "input_cost_per_token_priority": 0.0000015, - "litellm_provider": "openai", - "max_input_tokens": 272000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.0000045, - "output_cost_per_token_flex": 0.00000225, - "output_cost_per_token_batches": 0.00000225, - "output_cost_per_token_priority": 0.000009, - "regional_processing_uplift_multiplier_eu": 1.1, - "regional_processing_uplift_multiplier_us": 1.1, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/batch", - "/v1/responses" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_none_reasoning_effort": true, - "default_reasoning_effort": "none", - "source": "https://developers.openai.com/api/docs/pricing", - "supports_xhigh_reasoning_effort": true, - "supports_minimal_reasoning_effort": false - }, - "gpt-5.4-mini-2026-03-17": { - "cache_read_input_token_cost": 7.5e-8, - "cache_read_input_token_cost_batches": 3.75e-8, - "cache_read_input_token_cost_flex": 3.75e-8, - "cache_read_input_token_cost_priority": 1.5e-7, - "input_cost_per_token": 7.5e-7, - "input_cost_per_token_flex": 3.75e-7, - "input_cost_per_token_batches": 3.75e-7, - "input_cost_per_token_priority": 0.0000015, - "litellm_provider": "openai", - "max_input_tokens": 272000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.0000045, - "output_cost_per_token_flex": 0.00000225, - "output_cost_per_token_batches": 0.00000225, - "output_cost_per_token_priority": 0.000009, - "regional_processing_uplift_multiplier_eu": 1.1, - "regional_processing_uplift_multiplier_us": 1.1, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/batch", - "/v1/responses" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_none_reasoning_effort": true, - "default_reasoning_effort": "none", - "source": "https://developers.openai.com/api/docs/pricing", - "supports_xhigh_reasoning_effort": true, - "supports_minimal_reasoning_effort": false - }, - "gpt-5.4-nano": { - "cache_read_input_token_cost": 2e-8, - "cache_read_input_token_cost_batches": 1e-8, - "cache_read_input_token_cost_flex": 1e-8, - "input_cost_per_token": 2e-7, - "input_cost_per_token_flex": 1e-7, - "input_cost_per_token_batches": 1e-7, - "litellm_provider": "openai", - "max_input_tokens": 272000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.00000125, - "output_cost_per_token_flex": 6.25e-7, - "output_cost_per_token_batches": 6.25e-7, - "regional_processing_uplift_multiplier_eu": 1.1, - "regional_processing_uplift_multiplier_us": 1.1, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/batch", - "/v1/responses" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_none_reasoning_effort": true, - "default_reasoning_effort": "none", - "source": "https://developers.openai.com/api/docs/pricing", - "supports_xhigh_reasoning_effort": true, - "supports_minimal_reasoning_effort": false, - "deprecation_date": "2027-04-01" - }, - "gpt-5.4-nano-2026-03-17": { - "cache_read_input_token_cost": 2e-8, - "cache_read_input_token_cost_batches": 1e-8, - "cache_read_input_token_cost_flex": 1e-8, - "input_cost_per_token": 2e-7, - "input_cost_per_token_flex": 1e-7, - "input_cost_per_token_batches": 1e-7, - "litellm_provider": "openai", - "max_input_tokens": 272000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.00000125, - "output_cost_per_token_flex": 6.25e-7, - "output_cost_per_token_batches": 6.25e-7, - "regional_processing_uplift_multiplier_eu": 1.1, - "regional_processing_uplift_multiplier_us": 1.1, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/batch", - "/v1/responses" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_none_reasoning_effort": true, - "default_reasoning_effort": "none", - "source": "https://developers.openai.com/api/docs/pricing", - "supports_xhigh_reasoning_effort": true, - "supports_minimal_reasoning_effort": false - }, - "gpt-5.5": { - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_above_272k_tokens": 0.000001, - "cache_read_input_token_cost_batches": 2.5e-7, - "cache_read_input_token_cost_above_272k_tokens_batches": 5e-7, - "cache_read_input_token_cost_flex": 2.5e-7, - "cache_read_input_token_cost_priority": 0.00000125, - "input_cost_per_token": 0.000005, - "input_cost_per_token_above_272k_tokens": 0.00001, - "input_cost_per_token_flex": 0.0000025, - "input_cost_per_token_batches": 0.0000025, - "input_cost_per_token_above_272k_tokens_batches": 0.000005, - "input_cost_per_token_priority": 0.0000125, - "litellm_provider": "openai", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.00003, - "output_cost_per_token_above_272k_tokens": 0.000045, - "output_cost_per_token_flex": 0.000015, - "output_cost_per_token_batches": 0.000015, - "output_cost_per_token_above_272k_tokens_batches": 0.0000225, - "output_cost_per_token_priority": 0.000075, - "regional_processing_uplift_multiplier_eu": 1.1, - "regional_processing_uplift_multiplier_us": 1.1, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/batch", - "/v1/responses" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, + "supports_tool_search": true, + "supports_vision": true, + "supports_none_reasoning_effort": true, + "default_reasoning_effort": "none", + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false, + "input_cost_per_token_above_272k_tokens_flex": 0.0000025, + "output_cost_per_token_above_272k_tokens_flex": 0.00001125, + "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-7, + "source": "https://developers.openai.com/api/docs/pricing" + }, + "gpt-5.4-2026-03-05": { + "cache_read_input_token_cost": 2.5e-7, + "cache_read_input_token_cost_above_272k_tokens": 5e-7, + "cache_read_input_token_cost_batches": 1.3e-7, + "cache_read_input_token_cost_above_272k_tokens_batches": 2.5e-7, + "cache_read_input_token_cost_flex": 1.3e-7, + "cache_read_input_token_cost_priority": 5e-7, + "input_cost_per_token": 0.0000025, + "input_cost_per_token_above_272k_tokens": 0.000005, + "input_cost_per_token_flex": 0.00000125, + "input_cost_per_token_batches": 0.00000125, + "input_cost_per_token_above_272k_tokens_batches": 0.0000025, + "input_cost_per_token_priority": 0.000005, + "litellm_provider": "openai", + "max_input_tokens": 1050000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.000015, + "output_cost_per_token_above_272k_tokens": 0.0000225, + "output_cost_per_token_flex": 0.0000075, + "output_cost_per_token_batches": 0.0000075, + "output_cost_per_token_above_272k_tokens_batches": 0.00001125, + "output_cost_per_token_priority": 0.00003, + "regional_processing_uplift_multiplier_eu": 1.1, + "regional_processing_uplift_multiplier_us": 1.1, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_tool_search": true, + "supports_vision": true, + "supports_none_reasoning_effort": true, + "default_reasoning_effort": "none", + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false, + "input_cost_per_token_above_272k_tokens_flex": 0.0000025, + "output_cost_per_token_above_272k_tokens_flex": 0.00001125, + "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-7, + "source": "https://developers.openai.com/api/docs/pricing" + }, + "gpt-5.4-mini": { + "cache_read_input_token_cost": 7.5e-8, + "cache_read_input_token_cost_batches": 3.75e-8, + "cache_read_input_token_cost_flex": 3.75e-8, + "cache_read_input_token_cost_priority": 1.5e-7, + "input_cost_per_token": 7.5e-7, + "input_cost_per_token_flex": 3.75e-7, + "input_cost_per_token_batches": 3.75e-7, + "input_cost_per_token_priority": 0.0000015, + "litellm_provider": "openai", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.0000045, + "output_cost_per_token_flex": 0.00000225, + "output_cost_per_token_batches": 0.00000225, + "output_cost_per_token_priority": 0.000009, + "regional_processing_uplift_multiplier_eu": 1.1, + "regional_processing_uplift_multiplier_us": 1.1, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_tool_search": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "default_reasoning_effort": "none", + "source": "https://developers.openai.com/api/docs/pricing", + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false + }, + "gpt-5.4-mini-2026-03-17": { + "cache_read_input_token_cost": 7.5e-8, + "cache_read_input_token_cost_batches": 3.75e-8, + "cache_read_input_token_cost_flex": 3.75e-8, + "cache_read_input_token_cost_priority": 1.5e-7, + "input_cost_per_token": 7.5e-7, + "input_cost_per_token_flex": 3.75e-7, + "input_cost_per_token_batches": 3.75e-7, + "input_cost_per_token_priority": 0.0000015, + "litellm_provider": "openai", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.0000045, + "output_cost_per_token_flex": 0.00000225, + "output_cost_per_token_batches": 0.00000225, + "output_cost_per_token_priority": 0.000009, + "regional_processing_uplift_multiplier_eu": 1.1, + "regional_processing_uplift_multiplier_us": 1.1, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_tool_search": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "default_reasoning_effort": "none", + "source": "https://developers.openai.com/api/docs/pricing", + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false + }, + "gpt-5.4-nano": { + "cache_read_input_token_cost": 2e-8, + "cache_read_input_token_cost_batches": 1e-8, + "cache_read_input_token_cost_flex": 1e-8, + "input_cost_per_token": 2e-7, + "input_cost_per_token_flex": 1e-7, + "input_cost_per_token_batches": 1e-7, + "litellm_provider": "openai", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00000125, + "output_cost_per_token_flex": 6.25e-7, + "output_cost_per_token_batches": 6.25e-7, + "regional_processing_uplift_multiplier_eu": 1.1, + "regional_processing_uplift_multiplier_us": 1.1, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "default_reasoning_effort": "none", + "source": "https://developers.openai.com/api/docs/pricing", + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false, + "deprecation_date": "2027-04-01" + }, + "gpt-5.4-nano-2026-03-17": { + "cache_read_input_token_cost": 2e-8, + "cache_read_input_token_cost_batches": 1e-8, + "cache_read_input_token_cost_flex": 1e-8, + "input_cost_per_token": 2e-7, + "input_cost_per_token_flex": 1e-7, + "input_cost_per_token_batches": 1e-7, + "litellm_provider": "openai", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00000125, + "output_cost_per_token_flex": 6.25e-7, + "output_cost_per_token_batches": 6.25e-7, + "regional_processing_uplift_multiplier_eu": 1.1, + "regional_processing_uplift_multiplier_us": 1.1, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "default_reasoning_effort": "none", + "source": "https://developers.openai.com/api/docs/pricing", + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false + }, + "gpt-5.5": { + "cache_read_input_token_cost": 5e-7, + "cache_read_input_token_cost_above_272k_tokens": 0.000001, + "cache_read_input_token_cost_batches": 2.5e-7, + "cache_read_input_token_cost_above_272k_tokens_batches": 5e-7, + "cache_read_input_token_cost_flex": 2.5e-7, + "cache_read_input_token_cost_priority": 0.00000125, + "input_cost_per_token": 0.000005, + "input_cost_per_token_above_272k_tokens": 0.00001, + "input_cost_per_token_flex": 0.0000025, + "input_cost_per_token_batches": 0.0000025, + "input_cost_per_token_above_272k_tokens_batches": 0.000005, + "input_cost_per_token_priority": 0.0000125, + "litellm_provider": "openai", + "max_input_tokens": 1050000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00003, + "output_cost_per_token_above_272k_tokens": 0.000045, + "output_cost_per_token_flex": 0.000015, + "output_cost_per_token_batches": 0.000015, + "output_cost_per_token_above_272k_tokens_batches": 0.0000225, + "output_cost_per_token_priority": 0.000075, + "regional_processing_uplift_multiplier_eu": 1.1, + "regional_processing_uplift_multiplier_us": 1.1, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_none_reasoning_effort": true, @@ -35119,6 +36142,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_none_reasoning_effort": true, @@ -35138,6 +36162,15 @@ "source": "https://developers.openai.com/api/docs/pricing", "supports_reasoning": true }, + "gpt-5.5-cyber-preview": { + "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 0.0000125, + "litellm_provider": "openai", + "mode": "chat", + "output_cost_per_token": 0.000075, + "source": "https://developers.openai.com/api/docs/pricing", + "supports_reasoning": true + }, "gpt-5.6": { "cache_creation_input_token_cost": 0.000005, "cache_creation_input_token_cost_above_272k_tokens": 0.00001, @@ -35207,6 +36240,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_xhigh_reasoning_effort": true @@ -35320,6 +36354,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_xhigh_reasoning_effort": true @@ -35395,6 +36430,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_xhigh_reasoning_effort": true @@ -35469,6 +36505,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_xhigh_reasoning_effort": true @@ -35553,19 +36590,24 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_xhigh_reasoning_effort": true }, "gpt-6-luna": { "cache_creation_input_token_cost": 1.25e-7, + "cache_creation_input_token_cost_decisions": 1e-7, "cache_creation_input_token_cost_above_272k_tokens": 2.5e-7, + "cache_creation_input_token_cost_above_272k_tokens_decisions": 2e-7, "cache_creation_input_token_cost_above_272k_tokens_flex": 1.25e-7, "cache_creation_input_token_cost_above_272k_tokens_priority": 5e-7, "cache_creation_input_token_cost_flex": 6.25e-8, "cache_creation_input_token_cost_priority": 2.5e-7, "cache_read_input_token_cost": 1e-8, + "cache_read_input_token_cost_decisions": 1e-7, "cache_read_input_token_cost_above_272k_tokens": 2e-8, + "cache_read_input_token_cost_above_272k_tokens_decisions": 2e-7, "cache_read_input_token_cost_above_272k_tokens_flex": 1e-8, "cache_read_input_token_cost_above_272k_tokens_priority": 4e-8, "cache_read_input_token_cost_batches": 5e-9, @@ -35575,7 +36617,9 @@ "cache_read_input_token_cost_flex": 5e-9, "cache_read_input_token_cost_priority": 2e-8, "input_cost_per_token": 1e-7, + "input_cost_per_token_decisions": 1e-7, "input_cost_per_token_above_272k_tokens": 2e-7, + "input_cost_per_token_above_272k_tokens_decisions": 2e-7, "input_cost_per_token_above_272k_tokens_flex": 1e-7, "input_cost_per_token_above_272k_tokens_priority": 4e-7, "input_cost_per_token_batches": 5e-8, @@ -35588,7 +36632,9 @@ "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 5e-7, + "output_cost_per_token_decisions": 0, "output_cost_per_token_above_272k_tokens": 7.5e-7, + "output_cost_per_token_above_272k_tokens_decisions": 0, "output_cost_per_token_above_272k_tokens_flex": 3.75e-7, "output_cost_per_token_above_272k_tokens_priority": 0.0000015, "output_cost_per_token_batches": 2.5e-7, @@ -35630,6 +36676,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_xhigh_reasoning_effort": true @@ -35706,6 +36753,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_xhigh_reasoning_effort": true @@ -35716,25 +36764,31 @@ "cache_creation_input_token_cost_above_272k_tokens_batches": 0.0000025, "cache_creation_input_token_cost_above_272k_tokens_flex": 0.0000025, "cache_creation_input_token_cost_above_272k_tokens_priority": 0.00001, + "cache_creation_input_token_cost_above_272k_tokens_ultrafast": 0.00003, "cache_creation_input_token_cost_batches": 0.00000125, "cache_creation_input_token_cost_flex": 0.00000125, "cache_creation_input_token_cost_priority": 0.000005, + "cache_creation_input_token_cost_ultrafast": 0.000015, "cache_read_input_token_cost": 1e-7, "cache_read_input_token_cost_above_272k_tokens": 2e-7, "cache_read_input_token_cost_above_272k_tokens_batches": 1e-7, "cache_read_input_token_cost_above_272k_tokens_flex": 1e-7, "cache_read_input_token_cost_above_272k_tokens_priority": 4e-7, + "cache_read_input_token_cost_above_272k_tokens_ultrafast": 0.0000012, "cache_read_input_token_cost_batches": 5e-8, "cache_read_input_token_cost_flex": 5e-8, "cache_read_input_token_cost_priority": 2e-7, + "cache_read_input_token_cost_ultrafast": 6e-7, "input_cost_per_token": 0.000002, "input_cost_per_token_above_272k_tokens": 0.000004, "input_cost_per_token_above_272k_tokens_batches": 0.000002, "input_cost_per_token_above_272k_tokens_flex": 0.000002, "input_cost_per_token_above_272k_tokens_priority": 0.000008, + "input_cost_per_token_above_272k_tokens_ultrafast": 0.000024, "input_cost_per_token_batches": 0.000001, "input_cost_per_token_flex": 0.000001, "input_cost_per_token_priority": 0.000004, + "input_cost_per_token_ultrafast": 0.000012, "litellm_provider": "openai", "max_input_tokens": 922000, "max_output_tokens": 128000, @@ -35745,9 +36799,11 @@ "output_cost_per_token_above_272k_tokens_batches": 0.0000075, "output_cost_per_token_above_272k_tokens_flex": 0.0000075, "output_cost_per_token_above_272k_tokens_priority": 0.00003, + "output_cost_per_token_above_272k_tokens_ultrafast": 0.00009, "output_cost_per_token_batches": 0.000005, "output_cost_per_token_flex": 0.000005, "output_cost_per_token_priority": 0.00002, + "output_cost_per_token_ultrafast": 0.00006, "regional_processing_uplift_multiplier_eu": 1.1, "regional_processing_uplift_multiplier_us": 1.1, "search_context_cost_per_query": { @@ -35782,6 +36838,7 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_tool_search": true, "supports_vision": true, "supports_web_search": true, "supports_xhigh_reasoning_effort": true @@ -35934,6 +36991,44 @@ "supports_tool_choice": true, "supports_vision": false }, + "gpt-audio-mini-2025-10-06": { + "deprecation_date": "2026-07-23", + "input_cost_per_audio_token": 0.00001, + "input_cost_per_token": 6e-7, + "litellm_provider": "openai", + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_audio_token": 0.00002, + "output_cost_per_token": 0.0000024, + "source": "https://developers.openai.com/api/docs/models/gpt-audio-mini", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/realtime", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], + "supports_audio_input": true, + "supports_audio_output": true, + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_prompt_caching": false, + "supports_reasoning": false, + "supports_response_schema": false, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, "gpt-audio-mini-2025-12-15": { "input_cost_per_audio_token": 0.00001, "input_cost_per_token": 6e-7, @@ -35972,6 +37067,14 @@ "deprecation_date": "2027-01-20", "source": "https://developers.openai.com/api/docs/pricing" }, + "gpt-rosalind-discovery": { + "cache_read_input_token_cost": 5e-7, + "input_cost_per_token": 0.000005, + "litellm_provider": "openai", + "mode": "chat", + "output_cost_per_token": 0.000025, + "source": "https://developers.openai.com/api/docs/pricing" + }, "gpt-rosalind-research": { "cache_read_input_token_cost": 5e-7, "input_cost_per_token": 0.000005, @@ -36548,6 +37651,8 @@ "input_cost_per_token_flex": 0.00000165, "input_cost_per_token_priority": 0.000005775, "litellm_provider": "bedrock_converse", + "token_count_framing": 64, + "token_count_tokenizer": "kimi_k3", "max_input_tokens": 1000000, "max_output_tokens": 131072, "max_tokens": 131072, @@ -36598,17369 +37703,8093 @@ "supports_system_messages": true, "supports_tool_choice": true }, - "jamba-1.5": { - "input_cost_per_token": 2e-7, - "litellm_provider": "ai21", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 4e-7, - "supports_tool_choice": true - }, - "jamba-1.5-large": { - "input_cost_per_token": 0.000002, - "litellm_provider": "ai21", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 0.000008, - "supports_tool_choice": true - }, - "jamba-1.5-large@001": { - "input_cost_per_token": 0.000002, - "litellm_provider": "ai21", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 0.000008, - "supports_tool_choice": true - }, - "jamba-1.5-mini": { - "input_cost_per_token": 2e-7, - "litellm_provider": "ai21", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 4e-7, - "supports_tool_choice": true - }, - "jamba-1.5-mini@001": { - "input_cost_per_token": 2e-7, - "litellm_provider": "ai21", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 4e-7, - "supports_tool_choice": true - }, - "jamba-large-1.6": { - "input_cost_per_token": 0.000002, - "litellm_provider": "ai21", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 0.000008, - "supports_tool_choice": true - }, - "jamba-large-1.7": { - "input_cost_per_token": 0.000002, - "litellm_provider": "ai21", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, + "inferenceapis/MiniMaxAI/MiniMax-M3": { + "cache_read_input_token_cost": 8e-8, + "input_cost_per_token": 3.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 524288, + "max_tokens": 524288, "mode": "chat", - "output_cost_per_token": 0.000008, - "supports_tool_choice": true + "output_cost_per_token": 0.00000156, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": false }, - "jamba-mini-1.6": { - "input_cost_per_token": 2e-7, - "litellm_provider": "ai21", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, + "inferenceapis/Qwen/Qwen3-VL-235B-A22B-Instruct": { + "input_cost_per_token": 2.6e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 4e-7, - "supports_tool_choice": true + "output_cost_per_token": 0.00000115, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_reasoning": false, + "supports_tool_choice": true, + "supports_vision": true }, - "jamba-mini-1.7": { - "input_cost_per_token": 2e-7, - "litellm_provider": "ai21", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, + "inferenceapis/Qwen/Qwen3.8-Flash": { + "cache_read_input_token_cost": 2e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1000000, + "max_tokens": 1000000, "mode": "chat", - "output_cost_per_token": 4e-7, - "supports_tool_choice": true + "output_cost_per_token": 5e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": false }, - "jp.anthropic.claude-haiku-4-5-20251001-v1:0": { - "cache_creation_input_token_cost": 0.000001375, - "cache_creation_input_token_cost_above_1hr": 0.0000022, - "cache_read_input_token_cost": 1.1e-7, - "input_cost_per_token": 0.0000011, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "inferenceapis/deepseek-ai/DeepSeek-V3": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 1.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.0000055, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_assistant_prefill": true, - "supports_computer_use": true, + "output_cost_per_token": 3.8e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_native_structured_output": true, - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 4096, - "input_cost_per_token_batches": 5.5e-7, - "output_cost_per_token_batches": 0.00000275 + "supports_vision": false }, - "jp.anthropic.claude-haiku-5-5": { - "input_cost_per_token_above_100k_tokens": 5.5e-7, - "output_cost_per_token_above_100k_tokens": 0.00000275, - "cache_creation_input_token_cost_above_100k_tokens": 6.875e-7, - "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 0.0000011, - "cache_read_input_token_cost_above_100k_tokens": 5.5e-8, - "bedrock_converse_supports_strict_tools": false, - "cache_creation_input_token_cost": 1.375e-7, - "cache_creation_input_token_cost_above_1hr": 2.2e-7, - "cache_read_input_token_cost": 1.1e-8, - "input_cost_per_token": 1.1e-7, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "inferenceapis/deepseek-ai/DeepSeek-V4-Flash": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 1.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 5.5e-7, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supports_adaptive_thinking": true, - "supports_mid_conversation_system": true, - "supports_assistant_prefill": false, - "supports_computer_use": true, + "output_cost_per_token": 3.8e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true, - "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": false, - "supports_max_reasoning_effort": true, - "supports_output_config": true, - "bedrock_output_config_effort_ceiling": "xhigh", - "supports_parallel_tool_use_config": true, - "supports_forced_tool_use": true, - "thinking_always_on": false, - "prompt_cache_min_tokens": 512, - "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json" + "supports_vision": false }, - "jp.anthropic.claude-opus-4-7": { - "bedrock_converse_supports_strict_tools": false, - "cache_creation_input_token_cost": 0.000006875, - "cache_creation_input_token_cost_above_1hr": 0.000011, - "cache_read_input_token_cost": 5.5e-7, - "input_cost_per_token": 0.0000055, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "inferenceapis/deepseek-ai/DeepSeek-V4-Flash-0731": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 1.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.0000275, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supports_adaptive_thinking": true, - "supports_assistant_prefill": false, - "supports_computer_use": true, + "output_cost_per_token": 3.8e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true, - "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": false, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true, - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 2048, - "source": "https://aws.amazon.com/bedrock/pricing/" + "supports_vision": false }, - "jp.anthropic.claude-opus-4-8": { - "bedrock_converse_supports_strict_tools": false, - "supports_adaptive_thinking": true, - "supports_mid_conversation_system": true, - "cache_creation_input_token_cost": 0.000006875, - "cache_creation_input_token_cost_above_1hr": 0.000011, - "cache_read_input_token_cost": 5.5e-7, - "input_cost_per_token": 0.0000055, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "inferenceapis/deepseek-ai/DeepSeek-V4-Pro": { + "cache_read_input_token_cost": 1.7e-7, + "input_cost_per_token": 0.00000172, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.0000275, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supports_assistant_prefill": false, - "supports_computer_use": true, + "output_cost_per_token": 0.00000515, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true, - "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": false, - "supports_max_reasoning_effort": true, - "supports_output_config": true, - "bedrock_output_config_effort_ceiling": "xhigh", - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 1024, - "source": "https://aws.amazon.com/bedrock/pricing/" + "supports_vision": false }, - "jp.anthropic.claude-opus-5": { - "bedrock_converse_supports_strict_tools": false, - "supports_adaptive_thinking": true, - "supports_mid_conversation_system": true, - "cache_creation_input_token_cost": 0.000006875, - "cache_creation_input_token_cost_above_1hr": 0.000011, - "cache_read_input_token_cost": 5.5e-7, - "input_cost_per_token": 0.0000055, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "inferenceapis/deepseek-ai/DeepSeek-V4-Pro-0813": { + "cache_read_input_token_cost": 1.7e-7, + "input_cost_per_token": 0.00000172, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.0000275, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supports_assistant_prefill": false, - "supports_computer_use": true, + "output_cost_per_token": 0.00000515, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true, - "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": false, - "supports_max_reasoning_effort": true, - "supports_output_config": true, - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/" + "supports_vision": false }, - "jp.anthropic.claude-opus-5-5": { - "bedrock_converse_supports_strict_tools": false, - "supports_adaptive_thinking": true, - "supports_mid_conversation_system": true, - "cache_creation_input_token_cost": 0.0000055, - "cache_creation_input_token_cost_above_1hr": 0.0000088, - "cache_read_input_token_cost": 2.2e-7, - "input_cost_per_token": 0.0000044, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "inferenceapis/deepseek-ai/DeepSeek-V4.1-Flash": { + "cache_read_input_token_cost": 1e-8, + "input_cost_per_token": 4e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.000022, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supports_assistant_prefill": false, - "supports_computer_use": true, + "output_cost_per_token": 0.0000016, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true, - "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": false, - "supports_max_reasoning_effort": true, - "supports_output_config": true, - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 512, - "thinking_always_on": true, - "supports_forced_tool_use": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "supports_vision": false }, - "jp.anthropic.claude-sonnet-4-5-20250929-v1:0": { - "cache_creation_input_token_cost": 0.000004125, - "cache_creation_input_token_cost_above_1hr": 0.0000066, - "cache_read_input_token_cost": 3.3e-7, - "input_cost_per_token": 0.0000033, - "input_cost_per_token_above_200k_tokens": 0.0000066, - "output_cost_per_token_above_200k_tokens": 0.00002475, - "cache_creation_input_token_cost_above_200k_tokens": 0.00000825, - "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.0000132, - "cache_read_input_token_cost_above_200k_tokens": 6.6e-7, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "inferenceapis/deepseek-chat": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 1.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.0000165, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supports_assistant_prefill": true, - "supports_computer_use": true, + "output_cost_per_token": 3.8e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_native_structured_output": true, - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 1024, - "input_cost_per_token_batches": 0.00000165, - "output_cost_per_token_batches": 0.00000825, - "source": "https://aws.amazon.com/bedrock/pricing/" + "supports_vision": false }, - "jp.anthropic.claude-sonnet-4-6": { - "supports_adaptive_thinking": true, - "supports_legacy_thinking": true, - "cache_creation_input_token_cost": 0.000004125, - "cache_creation_input_token_cost_above_1hr": 0.0000066, - "cache_read_input_token_cost": 3.3e-7, - "input_cost_per_token": 0.0000033, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 1000000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "inferenceapis/deepseek-flash": { + "cache_read_input_token_cost": 1e-8, + "input_cost_per_token": 4e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.0000165, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supports_assistant_prefill": true, - "supports_computer_use": true, + "output_cost_per_token": 0.0000016, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_max_reasoning_effort": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_native_structured_output": true, - "supports_output_config": true, - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 1024, - "source": "https://aws.amazon.com/bedrock/pricing/" + "supports_vision": false }, - "jp.anthropic.claude-sonnet-5": { - "bedrock_converse_supports_strict_tools": false, - "cache_creation_input_token_cost": 0.00000275, - "cache_creation_input_token_cost_above_1hr": 0.0000044, - "cache_read_input_token_cost": 2.2e-7, - "input_cost_per_token": 0.0000022, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "inferenceapis/deepseek-reasoner": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 1.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.000011, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supports_adaptive_thinking": true, - "supports_mid_conversation_system": true, - "supports_assistant_prefill": false, - "supports_computer_use": true, + "output_cost_per_token": 3.8e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true, - "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": false, - "supports_max_reasoning_effort": true, - "supports_output_config": true, - "bedrock_output_config_effort_ceiling": "xhigh", - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 1024, - "source": "https://aws.amazon.com/bedrock/pricing/" + "supports_vision": false }, - "jp.anthropic.claude-sonnet-5-5": { - "bedrock_converse_supports_strict_tools": false, - "cache_creation_input_token_cost": 0.00000275, - "cache_creation_input_token_cost_above_1hr": 0.0000044, - "cache_read_input_token_cost": 1.1e-7, - "input_cost_per_token": 0.0000022, - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "inferenceapis/deepseek-v3": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 1.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.000011, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "supports_adaptive_thinking": true, - "supports_mid_conversation_system": true, - "supports_assistant_prefill": false, - "supports_computer_use": true, + "output_cost_per_token": 3.8e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true, - "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": false, - "supports_max_reasoning_effort": true, - "supports_output_config": true, - "bedrock_output_config_effort_ceiling": "xhigh", - "supports_parallel_tool_use_config": true, - "supports_forced_tool_use": false, - "thinking_always_on": true, - "prompt_cache_min_tokens": 512, - "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json" + "supports_vision": false }, - "kimi-k2-thinking-251104": { - "input_cost_per_token": 0, - "litellm_provider": "volcengine", - "max_input_tokens": 229376, - "max_output_tokens": 32768, - "max_tokens": 32768, + "inferenceapis/deepseek-v4-flash": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 1.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0, - "supports_assistant_prefill": true, + "output_cost_per_token": 3.8e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/deepseek-llama3.3-70b": { - "input_cost_per_token": 2e-7, - "litellm_provider": "lambda_ai", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "inferenceapis/deepseek-v4-pro": { + "cache_read_input_token_cost": 1.7e-7, + "input_cost_per_token": 0.00000172, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 6e-7, + "output_cost_per_token": 0.00000515, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/deepseek-r1-0528": { - "input_cost_per_token": 2e-7, - "litellm_provider": "lambda_ai", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "inferenceapis/deepseek-v4.1-flash": { + "cache_read_input_token_cost": 1e-8, + "input_cost_per_token": 4e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000016, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/deepseek-r1-671b": { - "input_cost_per_token": 8e-7, - "litellm_provider": "lambda_ai", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "inferenceapis/glm-5.3": { + "cache_read_input_token_cost": 3.4e-7, + "input_cost_per_token": 0.00000182, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048575, + "max_tokens": 1048575, "mode": "chat", - "output_cost_per_token": 8e-7, + "output_cost_per_token": 0.00000572, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/deepseek-v3-0324": { + "inferenceapis/glm-5.3-flash": { + "cache_read_input_token_cost": 4e-8, "input_cost_per_token": 2e-7, - "litellm_provider": "lambda_ai", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048575, + "max_tokens": 1048575, "mode": "chat", - "output_cost_per_token": 6e-7, + "output_cost_per_token": 6.6e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/hermes3-405b": { - "input_cost_per_token": 8e-7, - "litellm_provider": "lambda_ai", + "inferenceapis/gpt-oss-120b": { + "input_cost_per_token": 2e-7, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", "output_cost_per_token": 8e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/hermes3-70b": { - "input_cost_per_token": 1.2e-7, - "litellm_provider": "lambda_ai", + "inferenceapis/gpt-oss-20b": { + "input_cost_per_token": 4e-8, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 3e-7, + "output_cost_per_token": 1.9e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/hermes3-8b": { - "input_cost_per_token": 2.5e-8, - "litellm_provider": "lambda_ai", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "inferenceapis/kimi-k2.7-code": { + "cache_read_input_token_cost": 1.8e-7, + "input_cost_per_token": 8.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 4e-8, + "output_cost_per_token": 0.00000442, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/lfm-40b": { - "input_cost_per_token": 1e-7, - "litellm_provider": "lambda_ai", + "inferenceapis/llama-3.1-8b-instant": { + "input_cost_per_token": 2e-7, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 2e-7, + "output_cost_per_token": 5e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": false, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/lfm-7b": { - "input_cost_per_token": 2.5e-8, - "litellm_provider": "lambda_ai", + "inferenceapis/llama-3.3-70b": { + "input_cost_per_token": 2e-7, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 4e-8, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true - }, - "lambda_ai/llama-4-maverick-17b-128e-instruct-fp8": { - "input_cost_per_token": 5e-8, - "litellm_provider": "lambda_ai", - "max_input_tokens": 131072, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 1e-7, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true - }, - "lambda_ai/llama-4-scout-17b-16e-instruct": { - "input_cost_per_token": 5e-8, - "litellm_provider": "lambda_ai", - "max_input_tokens": 16384, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 1e-7, + "output_cost_per_token": 5e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": false, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/llama3.1-405b-instruct-fp8": { - "input_cost_per_token": 8e-7, - "litellm_provider": "lambda_ai", + "inferenceapis/llama-3.3-70b-instruct": { + "input_cost_per_token": 2e-7, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 8e-7, + "output_cost_per_token": 5e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": false, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/llama3.1-70b-instruct-fp8": { - "input_cost_per_token": 1.2e-7, - "litellm_provider": "lambda_ai", + "inferenceapis/llama-3.3-70b-versatile": { + "input_cost_per_token": 2e-7, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 3e-7, + "output_cost_per_token": 5e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": false, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/llama3.1-8b-instruct": { - "input_cost_per_token": 2.5e-8, - "litellm_provider": "lambda_ai", + "inferenceapis/meta-llama/Llama-3.3-70B-Instruct": { + "input_cost_per_token": 2e-7, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 4e-8, + "output_cost_per_token": 5e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": false, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/llama3.1-nemotron-70b-instruct-fp8": { - "input_cost_per_token": 1.2e-7, - "litellm_provider": "lambda_ai", + "inferenceapis/meta-llama/Llama-3.3-70B-Instruct-Turbo": { + "input_cost_per_token": 2e-7, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 3e-7, + "output_cost_per_token": 5e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": false, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/llama3.2-11b-vision-instruct": { - "input_cost_per_token": 1.5e-8, - "litellm_provider": "lambda_ai", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "inferenceapis/minimax-m3": { + "cache_read_input_token_cost": 8e-8, + "input_cost_per_token": 3.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 524288, + "max_tokens": 524288, "mode": "chat", - "output_cost_per_token": 2.5e-8, + "output_cost_per_token": 0.00000156, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": false }, - "lambda_ai/llama3.2-3b-instruct": { - "input_cost_per_token": 1.5e-8, - "litellm_provider": "lambda_ai", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "inferenceapis/moonshotai/Kimi-K2.7-Code": { + "cache_read_input_token_cost": 1.8e-7, + "input_cost_per_token": 8.9e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 2.5e-8, + "output_cost_per_token": 0.00000442, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/llama3.3-70b-instruct-fp8": { - "input_cost_per_token": 1.2e-7, - "litellm_provider": "lambda_ai", + "inferenceapis/openai/gpt-oss-120b": { + "input_cost_per_token": 2e-7, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 3e-7, + "output_cost_per_token": 8e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/qwen25-coder-32b-instruct": { - "input_cost_per_token": 5e-8, - "litellm_provider": "lambda_ai", + "inferenceapis/openai/gpt-oss-20b": { + "input_cost_per_token": 4e-8, + "litellm_provider": "inferenceapis", "max_input_tokens": 131072, - "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 1e-7, + "output_cost_per_token": 1.9e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": false }, - "lambda_ai/qwen3-32b-fp8": { - "input_cost_per_token": 5e-8, - "litellm_provider": "lambda_ai", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "inferenceapis/qwen3-vl": { + "input_cost_per_token": 2.6e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 1e-7, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_reasoning": true, - "supports_system_messages": true, - "supports_tool_choice": true - }, - "lemonade/Gemma-3-4b-it-GGUF": { - "input_cost_per_token": 0, - "litellm_provider": "lemonade", - "max_tokens": 8192, - "max_input_tokens": 128000, - "max_output_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "lemonade/Qwen3-4B-Instruct-2507-GGUF": { - "input_cost_per_token": 0, - "litellm_provider": "lemonade", - "max_tokens": 32768, - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "lemonade/Qwen3-Coder-30B-A3B-Instruct-GGUF": { - "input_cost_per_token": 0, - "litellm_provider": "lemonade", - "max_tokens": 32768, - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "lemonade/gpt-oss-120b-mxfp-GGUF": { - "input_cost_per_token": 0, - "litellm_provider": "lemonade", - "max_tokens": 32768, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "lemonade/gpt-oss-20b-mxfp4-GGUF": { - "input_cost_per_token": 0, - "litellm_provider": "lemonade", - "max_tokens": 32768, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "libertai/deepseek-v4-flash": { - "max_tokens": 200000, - "max_input_tokens": 200000, - "max_output_tokens": 200000, - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 0.00000175, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": false, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/deepseek-v4-flash-thinking": { - "max_tokens": 200000, - "max_input_tokens": 200000, - "max_output_tokens": 200000, - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 0.00000175, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": false, - "supports_reasoning": true, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/gemma-4-31b-it": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": true, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/gemma-4-31b-it-thinking": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": true, - "supports_reasoning": true, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/hermes-3-8b-tee": { - "max_tokens": 16000, - "max_input_tokens": 16000, - "max_output_tokens": 16000, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 6e-7, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": false, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/qwen3.5-122b-a10b": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 0.00000175, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": true, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/qwen3.5-122b-a10b-thinking": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 0.00000175, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": true, - "supports_reasoning": true, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/qwen3.6-27b": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 5e-7, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": true, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/qwen3.6-27b-thinking": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 5e-7, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": true, - "supports_reasoning": true, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/qwen3.6-35b-a3b": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 5e-7, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": true, - "source": "https://docs.libertai.io/apis/text/" - }, - "libertai/qwen3.6-35b-a3b-thinking": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 5e-7, - "litellm_provider": "libertai", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_vision": true, - "supports_reasoning": true, - "source": "https://docs.libertai.io/apis/text/" - }, - "llamagate/codellama-7b": { - "max_tokens": 4096, - "max_input_tokens": 16384, - "max_output_tokens": 4096, - "input_cost_per_token": 6e-8, - "output_cost_per_token": 1.2e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true - }, - "llamagate/deepseek-coder-6.7b": { - "max_tokens": 4096, - "max_input_tokens": 16384, - "max_output_tokens": 4096, - "input_cost_per_token": 6e-8, - "output_cost_per_token": 1.2e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true - }, - "llamagate/deepseek-r1-7b-qwen": { - "max_tokens": 16384, - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "input_cost_per_token": 8e-8, - "output_cost_per_token": 1.5e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "llamagate/deepseek-r1-8b": { - "max_tokens": 16384, - "max_input_tokens": 65536, - "max_output_tokens": 16384, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 2e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "llamagate/dolphin3-8b": { - "max_tokens": 8192, - "max_input_tokens": 128000, - "max_output_tokens": 8192, - "input_cost_per_token": 8e-8, - "output_cost_per_token": 1.5e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true - }, - "llamagate/gemma3-4b": { - "max_tokens": 8192, - "max_input_tokens": 128000, - "max_output_tokens": 8192, - "input_cost_per_token": 3e-8, - "output_cost_per_token": 8e-8, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_vision": true - }, - "llamagate/llama-3.1-8b": { - "max_tokens": 8192, - "max_input_tokens": 131072, - "max_output_tokens": 8192, - "input_cost_per_token": 3e-8, - "output_cost_per_token": 5e-8, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true - }, - "llamagate/llama-3.2-3b": { - "max_tokens": 8192, - "max_input_tokens": 131072, - "max_output_tokens": 8192, - "input_cost_per_token": 4e-8, - "output_cost_per_token": 8e-8, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true - }, - "llamagate/llava-7b": { - "max_tokens": 2048, - "max_input_tokens": 4096, - "max_output_tokens": 2048, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 2e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_response_schema": true, - "supports_vision": true - }, - "llamagate/mistral-7b-v0.3": { - "max_tokens": 8192, - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 1.5e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true - }, - "llamagate/openthinker-7b": { - "max_tokens": 8192, - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "input_cost_per_token": 8e-8, - "output_cost_per_token": 1.5e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "llamagate/qwen2.5-coder-7b": { - "max_tokens": 8192, - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "input_cost_per_token": 6e-8, - "output_cost_per_token": 1.2e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true - }, - "llamagate/qwen3-8b": { - "max_tokens": 8192, - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "input_cost_per_token": 4e-8, - "output_cost_per_token": 1.4e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true - }, - "llamagate/qwen3-vl-8b": { - "max_tokens": 8192, - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 5.5e-7, - "litellm_provider": "llamagate", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_vision": true - }, - "medlm-large": { - "input_cost_per_character": 0.000005, - "litellm_provider": "vertex_ai-language-models", - "max_input_tokens": 8192, - "max_output_tokens": 1024, - "max_tokens": 1024, - "mode": "chat", - "output_cost_per_character": 0.000015, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true - }, - "medlm-medium": { - "input_cost_per_character": 5e-7, - "litellm_provider": "vertex_ai-language-models", - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_character": 0.000001, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true - }, - "meta.llama2-13b-chat-v1": { - "input_cost_per_token": 7.5e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0.000001, - "source": "https://aws.amazon.com/bedrock/pricing/" - }, - "meta.llama2-70b-chat-v1": { - "input_cost_per_token": 0.00000195, - "litellm_provider": "bedrock", - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0.00000256, - "source": "https://aws.amazon.com/bedrock/pricing/" - }, - "meta.llama3-1-405b-instruct-v1:0": { - "input_cost_per_token": 0.0000024, - "litellm_provider": "bedrock", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0.0000024, - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", - "supports_function_calling": true, - "supports_tool_choice": false - }, - "meta.llama3-1-70b-instruct-v1:0": { - "input_cost_per_token": 7.2e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 128000, - "max_output_tokens": 2048, - "max_tokens": 2048, - "mode": "chat", - "output_cost_per_token": 7.2e-7, - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", - "supports_function_calling": true, - "supports_tool_choice": false - }, - "meta.llama3-1-8b-instruct-v1:0": { - "input_cost_per_token": 2.2e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 128000, - "max_output_tokens": 2048, - "max_tokens": 2048, - "mode": "chat", - "output_cost_per_token": 2.2e-7, - "supports_function_calling": true, - "supports_tool_choice": false - }, - "meta.llama3-2-11b-instruct-v1:0": { - "input_cost_per_token": 1.6e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 1.6e-7, - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", - "supports_function_calling": true, - "supports_tool_choice": false, - "supports_vision": true - }, - "meta.llama3-2-1b-instruct-v1:0": { - "input_cost_per_token": 1e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 1e-7, - "supports_function_calling": true, - "supports_tool_choice": false - }, - "meta.llama3-2-3b-instruct-v1:0": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 1.5e-7, - "supports_function_calling": true, - "supports_tool_choice": false - }, - "meta.llama3-2-90b-instruct-v1:0": { - "input_cost_per_token": 7.2e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 7.2e-7, - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", - "supports_function_calling": true, - "supports_tool_choice": false, - "supports_vision": true - }, - "meta.llama3-3-70b-instruct-v1:0": { - "input_cost_per_token": 7.2e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 7.2e-7, - "supports_function_calling": true, - "supports_tool_choice": false - }, - "meta.llama3-70b-instruct-v1:0": { - "input_cost_per_token": 0.00000265, - "litellm_provider": "bedrock", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0.0000035 - }, - "meta.llama3-8b-instruct-v1:0": { - "input_cost_per_token": 3e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 6e-7 - }, - "meta.llama4-maverick-17b-instruct-v1:0": { - "input_cost_per_token": 2.4e-7, - "input_cost_per_token_batches": 1.2e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 9.7e-7, - "output_cost_per_token_batches": 4.85e-7, - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text", - "code" - ], - "supports_function_calling": true, - "supports_tool_choice": false - }, - "meta.llama4-scout-17b-instruct-v1:0": { - "input_cost_per_token": 1.7e-7, - "input_cost_per_token_batches": 8.5e-8, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 6.6e-7, - "output_cost_per_token_batches": 3.3e-7, - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text", - "code" - ], - "supports_function_calling": true, - "supports_tool_choice": false - }, - "meta/muse-spark-1.1": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "meta", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.00000425, - "search_context_cost_per_query": { - "search_context_size_high": 0.0025, - "search_context_size_low": 0.0025, - "search_context_size_medium": 0.0025 - }, - "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/responses", - "/v1/messages" - ], - "supported_modalities": [ - "text", - "image", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_xhigh_reasoning_effort": true - }, - "meta/muse-spark-1.2": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "meta", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.00000425, - "search_context_cost_per_query": { - "search_context_size_high": 0.0025, - "search_context_size_low": 0.0025, - "search_context_size_medium": 0.0025 - }, - "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/responses", - "/v1/messages" - ], - "supported_modalities": [ - "text", - "image", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_xhigh_reasoning_effort": true - }, - "meta/muse-spark-1.2-contributor": { - "cache_read_input_token_cost": 2e-9, - "input_cost_per_token": 1e-7, - "litellm_provider": "meta", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 2e-7, - "search_context_cost_per_query": { - "search_context_size_high": 0.0025, - "search_context_size_low": 0.0025, - "search_context_size_medium": 0.0025 - }, - "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/responses", - "/v1/messages" - ], - "supported_modalities": [ - "text", - "image", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_xhigh_reasoning_effort": true - }, - "meta/muse-spark-1.3": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "meta", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.00000425, - "search_context_cost_per_query": { - "search_context_size_high": 0.0025, - "search_context_size_low": 0.0025, - "search_context_size_medium": 0.0025 - }, - "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/responses", - "/v1/messages" - ], - "supported_modalities": [ - "text", - "image", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_xhigh_reasoning_effort": true - }, - "meta/muse-spark-1.3-contributor": { - "cache_read_input_token_cost": 2e-9, - "input_cost_per_token": 1e-7, - "litellm_provider": "meta", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 2e-7, - "search_context_cost_per_query": { - "search_context_size_high": 0.0025, - "search_context_size_low": 0.0025, - "search_context_size_medium": 0.0025 - }, - "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/responses", - "/v1/messages" - ], - "supported_modalities": [ - "text", - "image", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_xhigh_reasoning_effort": true - }, - "meta_llama/Llama-3.3-70B-Instruct": { - "litellm_provider": "meta_llama", - "max_input_tokens": 128000, - "max_output_tokens": 4028, - "max_tokens": 4028, - "mode": "chat", - "source": "https://ai.developer.meta.com/docs/models", - "supported_modalities": [ - "text" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_tool_choice": true - }, - "meta_llama/Llama-3.3-8B-Instruct": { - "litellm_provider": "meta_llama", - "max_input_tokens": 128000, - "max_output_tokens": 4028, - "max_tokens": 4028, - "mode": "chat", - "source": "https://ai.developer.meta.com/docs/models", - "supported_modalities": [ - "text" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_tool_choice": true - }, - "meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8": { - "litellm_provider": "meta_llama", - "max_input_tokens": 1000000, - "max_output_tokens": 4028, - "max_tokens": 4028, - "mode": "chat", - "source": "https://ai.developer.meta.com/docs/models", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_tool_choice": true - }, - "meta_llama/Llama-4-Scout-17B-16E-Instruct-FP8": { - "litellm_provider": "meta_llama", - "max_input_tokens": 10000000, - "max_output_tokens": 4028, - "max_tokens": 4028, - "mode": "chat", - "source": "https://ai.developer.meta.com/docs/models", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_tool_choice": true - }, - "minimax.minimax-m2": { - "input_cost_per_token": 3e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 0.0000012, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_system_messages": true, - "supports_native_structured_output": true, - "supports_response_schema": true, - "supports_vision": false - }, - "minimax.minimax-m2.1": { - "input_cost_per_token": 3e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 196000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 0.0000012, - "supports_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false - }, - "minimax.minimax-m2.5": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 196000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "supports_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false - }, - "minimax/MiniMax-M2": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "cache_read_input_token_cost": 3e-8, - "cache_creation_input_token_cost": 3.75e-7, - "litellm_provider": "minimax", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_system_messages": true, - "max_input_tokens": 200000, - "max_output_tokens": 8192 - }, - "minimax/MiniMax-M2.1": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "cache_read_input_token_cost": 3e-8, - "cache_creation_input_token_cost": 3.75e-7, - "litellm_provider": "minimax", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_system_messages": true, - "max_input_tokens": 1000000, - "max_output_tokens": 8192 - }, - "minimax/MiniMax-M2.1-lightning": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000024, - "cache_read_input_token_cost": 3e-8, - "cache_creation_input_token_cost": 3.75e-7, - "litellm_provider": "minimax", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_system_messages": true, - "max_input_tokens": 1000000, - "max_output_tokens": 8192 - }, - "minimax/MiniMax-M2.5": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "cache_read_input_token_cost": 3e-8, - "cache_creation_input_token_cost": 3.75e-7, - "litellm_provider": "minimax", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_system_messages": true, - "max_input_tokens": 1000000, - "max_output_tokens": 8192 - }, - "minimax/MiniMax-M2.5-lightning": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000024, - "cache_read_input_token_cost": 3e-8, - "cache_creation_input_token_cost": 3.75e-7, - "litellm_provider": "minimax", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_system_messages": true, - "max_input_tokens": 1000000, - "max_output_tokens": 8192 - }, - "minimax/MiniMax-M3": { - "input_cost_per_token": 3e-7, - "input_cost_per_token_above_512k_tokens": 6e-7, - "output_cost_per_token": 0.0000012, - "output_cost_per_token_above_512k_tokens": 0.0000024, - "cache_read_input_token_cost": 6e-8, - "cache_read_input_token_cost_above_512k_tokens": 1.2e-7, - "litellm_provider": "minimax", - "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_system_messages": true, - "supports_vision": true, - "max_input_tokens": 1000000, - "max_output_tokens": 128000 - }, - "mistral.devstral-2-123b": { - "input_cost_per_token": 4e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 256000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "mode": "chat", - "output_cost_per_token": 0.000002, - "supports_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false - }, - "mistral.magistral-small-2509": { - "input_cost_per_token": 5e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 40000, - "max_tokens": 40000, - "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_vision": true - }, - "mistral.ministral-3-14b-instruct": { - "input_cost_per_token": 2e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_system_messages": true, - "supports_native_structured_output": true, - "supports_response_schema": true, - "supports_vision": true - }, - "mistral.ministral-3-3b-instruct": { - "input_cost_per_token": 1e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 1e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_system_messages": true, - "supports_native_structured_output": true, - "supports_response_schema": true, - "supports_vision": true - }, - "mistral.ministral-3-8b-instruct": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 1.5e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_system_messages": true, - "supports_native_structured_output": true, - "supports_response_schema": true, - "supports_vision": true - }, - "mistral.mistral-7b-instruct-v0:2": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_tool_choice": true - }, - "mistral.mistral-large-2402-v1:0": { - "input_cost_per_token": 0.000004, - "litellm_provider": "bedrock", - "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, - "mode": "chat", - "output_cost_per_token": 0.000012, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_function_calling": true - }, - "mistral.mistral-large-2407-v1:0": { - "input_cost_per_token": 0.000002, - "litellm_provider": "bedrock", - "max_input_tokens": 128000, - "max_output_tokens": 8191, - "max_tokens": 8191, - "mode": "chat", - "output_cost_per_token": 0.000006, - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", - "supports_function_calling": true, - "supports_tool_choice": true - }, - "mistral.mistral-large-3-675b-instruct": { - "input_cost_per_token": 5e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 256000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_system_messages": true, - "supports_native_structured_output": true, - "supports_response_schema": true, - "supports_vision": true - }, - "mistral.mistral-small-2402-v1:0": { - "input_cost_per_token": 0.000001, - "input_cost_per_token_batches": 5e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, - "mode": "chat", - "output_cost_per_token": 0.000003, - "output_cost_per_token_batches": 0.0000015, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_function_calling": true - }, - "mistral.mixtral-8x7b-instruct-v0:1": { - "input_cost_per_token": 4.5e-7, - "litellm_provider": "bedrock", - "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, - "mode": "chat", - "output_cost_per_token": 7e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_tool_choice": true - }, - "mistral.pixtral-large-2502-v1:0": { - "input_cost_per_token": 0.000002, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", - "output_cost_per_token": 0.000006, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_function_calling": true, - "supports_tool_choice": false - }, - "mistral.voxtral-mini-3b-2507": { - "input_cost_per_token": 4e-8, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 4e-8, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": true, - "supports_system_messages": true, - "supports_native_structured_output": true - }, - "mistral.voxtral-small-24b-2507": { - "input_cost_per_token": 1e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 3e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": true, - "supports_system_messages": true, - "supports_native_structured_output": true - }, - "mistral/codestral-2508": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 3e-7, - "litellm_provider": "mistral", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 9e-7, - "source": "https://docs.mistral.ai/models/model-cards/codestral-25-08", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/codestral-latest": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 3e-7, - "litellm_provider": "mistral", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 9e-7, - "source": "https://docs.mistral.ai/models/model-cards/codestral-25-08", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/codestral-mamba-latest": { - "cache_read_input_token_cost": 2.5e-8, - "input_cost_per_token": 2.5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 2.5e-7, - "source": "https://mistral.ai/technology/", - "supports_assistant_prefill": true, - "supports_tool_choice": true - }, - "mistral/devstral-latest": { - "cache_read_input_token_cost": 4e-8, - "input_cost_per_token": 4e-7, - "litellm_provider": "mistral", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://mistral.ai/news/devstral-2-vibe-cli", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/devstral-medium-latest": { - "cache_read_input_token_cost": 4e-8, - "input_cost_per_token": 4e-7, - "litellm_provider": "mistral", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://mistral.ai/news/devstral-2-vibe-cli", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/devstral-small-latest": { - "cache_read_input_token_cost": 1e-8, - "input_cost_per_token": 1e-7, - "litellm_provider": "mistral", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 3e-7, - "source": "https://docs.mistral.ai/models/devstral-small-2-25-12", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/glm-5-2": { - "cache_read_input_token_cost": 1.4e-7, - "input_cost_per_token": 0.0000014, - "litellm_provider": "mistral", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "reasoning_effort_levels": [ - "none", - "minimal", - "low", - "medium", - "high", - "xhigh", - "max" - ], - "source": "https://docs.mistral.ai/models/zai-glm-5-2", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/labs-leanstral-1-5": { - "deprecation_date": "2026-09-30", - "input_cost_per_token": 0, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0, - "source": "https://docs.mistral.ai/models/model-cards/leanstral-1-5", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/labs-leanstral-1-5-1": { - "input_cost_per_token": 0, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0, - "source": "https://docs.mistral.ai/models/model-cards/leanstral-1-5", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/magistral-medium-latest": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/magistral-small-latest": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/ministral-14b-2512": { - "input_cost_per_token": 2e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://docs.mistral.ai/models/ministral-3-14b-25-12", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "cache_read_input_token_cost": 2e-8 - }, - "mistral/ministral-14b-latest": { - "input_cost_per_token": 2e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://docs.mistral.ai/models/ministral-3-14b-25-12", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "cache_read_input_token_cost": 2e-8 - }, - "mistral/ministral-3-14b-2512": { - "cache_read_input_token_cost": 2e-8, - "input_cost_per_token": 2e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://mistral.ai/pricing", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/ministral-3-3b-2512": { - "cache_read_input_token_cost": 1e-8, - "input_cost_per_token": 1e-7, - "litellm_provider": "mistral", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 1e-7, - "source": "https://mistral.ai/pricing", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/ministral-3-8b-2512": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 1.5e-7, - "source": "https://mistral.ai/pricing", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/ministral-3b-2512": { - "input_cost_per_token": 1e-7, - "litellm_provider": "mistral", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 1e-7, - "source": "https://docs.mistral.ai/models/ministral-3-3b-25-12", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "cache_read_input_token_cost": 1e-8 - }, - "mistral/ministral-3b-latest": { - "input_cost_per_token": 1e-7, - "litellm_provider": "mistral", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 1e-7, - "source": "https://docs.mistral.ai/models/ministral-3-3b-25-12", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "cache_read_input_token_cost": 1e-8 - }, - "mistral/ministral-8b-2512": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 1.5e-7, - "source": "https://mistral.ai/pricing", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/ministral-8b-latest": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 1.5e-7, - "source": "https://mistral.ai/pricing", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-code-agent-latest": { - "cache_read_input_token_cost": 4e-8, - "input_cost_per_token": 4e-7, - "litellm_provider": "mistral", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://mistral.ai/news/devstral-2-vibe-cli", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/mistral-code-fim-latest": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 3e-7, - "litellm_provider": "mistral", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 9e-7, - "source": "https://docs.mistral.ai/models/model-cards/codestral-25-08", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/mistral-code-latest": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 3e-7, - "litellm_provider": "mistral", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 9e-7, - "source": "https://docs.mistral.ai/models/model-cards/codestral-25-08", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/mistral-large-2512": { - "cache_read_input_token_cost": 5e-8, - "input_cost_per_token": 5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://docs.mistral.ai/models/mistral-large-3-25-12", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-large-3": { - "cache_read_input_token_cost": 5e-8, - "input_cost_per_token": 5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://docs.mistral.ai/models/mistral-large-3-25-12", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-large-4": { - "cache_read_input_token_cost": 6.8e-8, - "input_cost_per_token": 6.8e-7, - "litellm_provider": "mistral", - "max_input_tokens": 524288, - "max_tokens": 524288, - "mode": "chat", - "output_cost_per_token": 0.00000209, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/mistral-large-4", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-large-4-0": { - "cache_read_input_token_cost": 6.8e-8, - "input_cost_per_token": 6.8e-7, - "litellm_provider": "mistral", - "max_input_tokens": 524288, - "max_tokens": 524288, - "mode": "chat", - "output_cost_per_token": 0.00000209, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/mistral-large-4", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-large-latest": { - "cache_read_input_token_cost": 5e-8, - "input_cost_per_token": 5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://docs.mistral.ai/models/mistral-large-3-25-12", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-medium": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-medium-2604": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-medium-3": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-medium-3-5": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-medium-3.5": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-medium-latest": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-small": { - "cache_read_input_token_cost": 1e-8, - "input_cost_per_token": 1e-7, - "litellm_provider": "mistral", - "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, - "mode": "chat", - "output_cost_per_token": 3e-7, - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/mistral-small-2603": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 6e-7, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-small-latest": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 6e-7, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-tiny": { - "cache_read_input_token_cost": 2.5e-8, - "input_cost_per_token": 2.5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 32000, - "max_output_tokens": 8191, - "max_tokens": 8191, - "mode": "chat", - "output_cost_per_token": 2.5e-7, - "supports_assistant_prefill": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/mistral-vibe-cli-fast": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 6e-7, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-vibe-cli-latest": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/mistral-vibe-cli-with-tools": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "mistral", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "reasoning_effort_levels": [ - "none", - "high" - ], - "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/open-mistral-nemo": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 3e-7, - "litellm_provider": "mistral", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 3e-7, - "source": "https://mistral.ai/technology/", - "supports_assistant_prefill": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/pixtral-large-latest": { - "cache_read_input_token_cost": 2e-7, - "input_cost_per_token": 0.000002, - "litellm_provider": "mistral", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000006, - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "mistral/voxtral-small-2507": { - "cache_read_input_token_cost": 1e-8, - "input_cost_per_token": 1e-7, - "litellm_provider": "mistral", - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 4e-7, - "source": "https://docs.mistral.ai/models/voxtral-small-25-07", - "supports_audio_input": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/voxtral-small-latest": { - "cache_read_input_token_cost": 1e-8, - "input_cost_per_token": 1e-7, - "litellm_provider": "mistral", - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 4e-7, - "source": "https://docs.mistral.ai/models/voxtral-small-25-07", - "supports_audio_input": true, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/zai-glm-5": { - "cache_read_input_token_cost": 1.4e-7, - "input_cost_per_token": 0.0000014, - "litellm_provider": "mistral", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "reasoning_effort_levels": [ - "low", - "high", - "max" - ], - "source": "https://docs.mistral.ai/models/zai-glm-5-3", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/zai-glm-5-2": { - "cache_read_input_token_cost": 1.4e-7, - "deprecation_date": "2026-10-31", - "input_cost_per_token": 0.0000014, - "litellm_provider": "mistral", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "reasoning_effort_levels": [ - "none", - "minimal", - "low", - "medium", - "high", - "xhigh", - "max" - ], - "source": "https://docs.mistral.ai/models/zai-glm-5-2", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/zai-glm-5-3": { - "cache_read_input_token_cost": 1.4e-7, - "input_cost_per_token": 0.0000014, - "litellm_provider": "mistral", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "reasoning_effort_levels": [ - "low", - "high", - "max" - ], - "source": "https://docs.mistral.ai/models/zai-glm-5-3", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "mistral/zai-glm-latest": { - "cache_read_input_token_cost": 1.4e-7, - "input_cost_per_token": 0.0000014, - "litellm_provider": "mistral", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "reasoning_effort_levels": [ - "low", - "high", - "max" - ], - "source": "https://docs.mistral.ai/models/zai-glm-5-3", - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "moonshot.kimi-k2-thinking": { - "input_cost_per_token": 6e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 256000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "output_cost_per_token": 0.0000025, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_system_messages": true, - "supports_native_structured_output": true, - "supports_response_schema": true, - "supports_vision": false - }, - "moonshot/kimi-k2.5": { - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 6e-7, - "litellm_provider": "moonshot", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.000003, - "source": "https://platform.moonshot.ai/docs/guide/kimi-k2-5-quickstart", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true - }, - "moonshot/kimi-k2.6": { - "cache_read_input_token_cost": 1.6e-7, - "input_cost_per_token": 9.5e-7, - "litellm_provider": "moonshot", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://platform.kimi.ai/docs/pricing/chat-k26", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true - }, - "moonshot/kimi-k2.7-code": { - "cache_read_input_token_cost": 1.9e-7, - "input_cost_per_token": 9.5e-7, - "litellm_provider": "moonshot", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://platform.kimi.ai/docs/pricing/chat-k27-code", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true - }, - "moonshot/kimi-k3": { - "cache_read_input_token_cost": 3e-7, - "input_cost_per_token": 0.000003, - "litellm_provider": "moonshot", - "max_input_tokens": 1048576, - "max_output_tokens": 1048576, - "max_tokens": 1048576, - "mode": "chat", - "output_cost_per_token": 0.000015, - "reasoning_effort_levels": [ - "low", - "high", - "max" - ], - "source": "https://platform.kimi.ai/docs/pricing/chat-k3", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true - }, - "moonshot/moonshot-v1-128k": { - "input_cost_per_token": 0.000002, - "litellm_provider": "moonshot", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://platform.moonshot.ai/docs/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "moonshot/moonshot-v1-128k-vision-preview": { - "input_cost_per_token": 0.000002, - "litellm_provider": "moonshot", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://platform.moonshot.ai/docs/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "moonshot/moonshot-v1-32k": { - "input_cost_per_token": 0.000001, - "litellm_provider": "moonshot", - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.000003, - "source": "https://platform.moonshot.ai/docs/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "moonshot/moonshot-v1-32k-vision-preview": { - "input_cost_per_token": 0.000001, - "litellm_provider": "moonshot", - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.000003, - "source": "https://platform.moonshot.ai/docs/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "moonshot/moonshot-v1-8k": { - "input_cost_per_token": 2e-7, - "litellm_provider": "moonshot", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://platform.moonshot.ai/docs/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "moonshot/moonshot-v1-8k-vision-preview": { - "input_cost_per_token": 2e-7, - "litellm_provider": "moonshot", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://platform.moonshot.ai/docs/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "moonshot/moonshot-v1-auto": { - "input_cost_per_token": 0.000002, - "litellm_provider": "moonshot", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://platform.moonshot.ai/docs/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "moonshotai.kimi-k2.5": { - "input_cost_per_token": 6e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 256000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "output_cost_per_token": 0.000003, - "supports_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_response_schema": true - }, - "moonshotai.kimi-k3": { - "supports_regex_lookaround": false, - "cache_creation_input_token_cost": 0.000004125, - "cache_creation_input_token_cost_flex": 0.0000020625, - "cache_creation_input_token_cost_priority": 0.00000721875, - "cache_read_input_token_cost": 3.3e-7, - "cache_read_input_token_cost_flex": 1.65e-7, - "cache_read_input_token_cost_priority": 5.775e-7, - "input_cost_per_token": 0.0000033, - "input_cost_per_token_flex": 0.00000165, - "input_cost_per_token_priority": 0.000005775, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000165, - "output_cost_per_token_flex": 0.00000825, - "output_cost_per_token_priority": 0.000028875, - "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_prompt_cache_breakpoint": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "morph/morph-v3-fast": { - "input_cost_per_token": 8e-7, - "litellm_provider": "morph", - "max_input_tokens": 16000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "output_cost_per_token": 0.0000012, - "supports_function_calling": false, - "supports_parallel_function_calling": false, - "supports_system_messages": true, - "supports_tool_choice": false, - "supports_vision": false - }, - "morph/morph-v3-large": { - "input_cost_per_token": 9e-7, - "litellm_provider": "morph", - "max_input_tokens": 16000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "output_cost_per_token": 0.0000019, - "supports_function_calling": false, - "supports_parallel_function_calling": false, - "supports_system_messages": true, - "supports_tool_choice": false, - "supports_vision": false - }, - "nebius/MiniMaxAI/MiniMax-M2.5": { - "max_tokens": 196608, - "max_input_tokens": 196608, - "max_output_tokens": 196608, - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/MiniMaxAI%2FMiniMax-M2.5" - }, - "nebius/MiniMaxAI/MiniMax-M3": { - "max_tokens": 1048576, - "max_input_tokens": 1048576, - "max_output_tokens": 1048576, - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/MiniMaxAI%2FMiniMax-M3" - }, - "nebius/NousResearch/Hermes-3-Llama-3.1-405B": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/NousResearch/Hermes-4-405B": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/NousResearch%2FHermes-4-405B" - }, - "nebius/NousResearch/Hermes-4-70B": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 1.3e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/NousResearch%2FHermes-4-70B" - }, - "nebius/Qwen/QwQ-32B": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 4.5e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen2-VL-72B-Instruct": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 1.3e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_vision": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen2-VL-7B-Instruct": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 2e-8, - "output_cost_per_token": 6e-8, - "litellm_provider": "nebius", - "mode": "chat", - "supports_vision": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen2.5-32B-Instruct": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 6e-8, - "output_cost_per_token": 2e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen2.5-72B-Instruct": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 1.3e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen2.5-Coder-7B": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 1e-8, - "output_cost_per_token": 3e-8, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen2.5-VL-72B-Instruct": { - "max_tokens": 32000, - "max_input_tokens": 32000, - "max_output_tokens": 32000, - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 7.5e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/Qwen%2FQwen2.5-VL-72B-Instruct" - }, - "nebius/Qwen/Qwen3-14B": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 8e-8, - "output_cost_per_token": 2.4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen3-235B-A22B": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 2e-7, - "output_cost_per_token": 6e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen3-235B-A22B-Instruct-2507": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 2e-7, - "output_cost_per_token": 6e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-235B-A22B-Instruct-2507" - }, - "nebius/Qwen/Qwen3-30B-A3B": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 3e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen3-30B-A3B-Instruct-2507": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 3e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-30B-A3B-Instruct-2507" - }, - "nebius/Qwen/Qwen3-32B": { - "max_tokens": 40960, - "max_input_tokens": 40960, - "max_output_tokens": 40960, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 3e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-32B" - }, - "nebius/Qwen/Qwen3-4B": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 8e-8, - "output_cost_per_token": 2.4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/Qwen/Qwen3-Next-80B-A3B-Thinking": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 0.0000012, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-Next-80B-A3B-Thinking" - }, - "nebius/Qwen/Qwen3.5-397B-A17B": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000036, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3.5-397B-A17B" - }, - "nebius/Qwen/Qwen3.8-27B": { - "input_cost_per_token": 4.5e-7, - "litellm_provider": "nebius", - "max_input_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.000003, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3.8-27B", - "supports_function_calling": true, - "supports_reasoning": true - }, - "nebius/deepseek-ai/DeepSeek-R1": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 8e-7, - "output_cost_per_token": 0.0000024, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://nebius.com/prices" - }, - "nebius/deepseek-ai/DeepSeek-R1-0528": { - "max_tokens": 164000, - "max_input_tokens": 164000, - "max_output_tokens": 164000, - "input_cost_per_token": 8e-7, - "output_cost_per_token": 0.0000024, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://nebius.com/prices" - }, - "nebius/deepseek-ai/DeepSeek-R1-Distill-Llama-70B": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 7.5e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/deepseek-ai/DeepSeek-V3": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 5e-7, - "output_cost_per_token": 0.0000015, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/deepseek-ai/DeepSeek-V3-0324": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 5e-7, - "output_cost_per_token": 0.0000015, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/deepseek-ai/DeepSeek-V4-Flash": { - "max_tokens": 1048576, - "max_input_tokens": 1048576, - "max_output_tokens": 1048576, - "input_cost_per_token": 1.4e-7, - "output_cost_per_token": 2.8e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Flash" - }, - "nebius/deepseek-ai/DeepSeek-V4-Flash-0731": { - "max_tokens": 1024000, - "max_input_tokens": 1024000, - "max_output_tokens": 1024000, - "input_cost_per_token": 1.4e-7, - "output_cost_per_token": 2.8e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Flash-0731" - }, - "nebius/deepseek-ai/DeepSeek-V4-Pro": { - "max_tokens": 1048576, - "max_input_tokens": 1048576, - "max_output_tokens": 1048576, - "input_cost_per_token": 0.00000175, - "output_cost_per_token": 0.0000035, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Pro" - }, - "nebius/deepseek-ai/DeepSeek-V4-Pro-0813": { - "input_cost_per_token": 0.00000132, - "litellm_provider": "nebius", - "max_input_tokens": 979000, - "mode": "chat", - "output_cost_per_token": 0.00000396, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Pro-0813", - "supports_function_calling": true, - "supports_reasoning": true - }, - "nebius/deepseek-ai/DeepSeek-V4.1-Flash": { - "input_cost_per_token": 3e-7, - "litellm_provider": "nebius", - "max_input_tokens": 1048000, - "max_output_tokens": 384000, - "max_tokens": 384000, - "mode": "chat", - "output_cost_per_token": 0.0000012, - "source": "https://tokenfactory.nebius.com/endpoints?modals=endpoint-details&model-id=deepseek-ai/DeepSeek-V4.1-Flash", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_vision": true - }, - "nebius/google/gemma-3-27b-it": { - "max_tokens": 110000, - "max_input_tokens": 110000, - "max_output_tokens": 110000, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 3e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/google%2Fgemma-3-27b-it" - }, - "nebius/meta-llama/Llama-3.3-70B-Instruct": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 1.3e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/meta-llama/Llama-Guard-3-8B": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 2e-8, - "output_cost_per_token": 6e-8, - "litellm_provider": "nebius", - "mode": "chat", - "source": "https://nebius.com/prices" - }, - "nebius/meta-llama/Meta-Llama-3.1-405B-Instruct": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/meta-llama/Meta-Llama-3.1-70B-Instruct": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 1.3e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/meta-llama/Meta-Llama-3.1-8B-Instruct": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 2e-8, - "output_cost_per_token": 6e-8, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/mistralai/Mistral-Nemo-Instruct-2407": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 4e-8, - "output_cost_per_token": 1.2e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/moonshotai/Kimi-K2.6": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 9.5e-7, - "output_cost_per_token": 0.000004, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/moonshotai%2FKimi-K2.6" - }, - "nebius/moonshotai/Kimi-K2.7-Code": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 9.5e-7, - "output_cost_per_token": 0.000004, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/moonshotai%2FKimi-K2.7-Code" - }, - "nebius/moonshotai/Kimi-K3": { - "max_tokens": 1024000, - "max_input_tokens": 1024000, - "max_output_tokens": 1024000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/moonshotai%2FKimi-K3" - }, - "nebius/nvidia/Cosmos3-Super-Reasoner": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 3e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/nvidia%2FCosmos3-Super-Reasoner" - }, - "nebius/nvidia/Llama-3.1-Nemotron-Ultra-253B-v1": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000018, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/nvidia/Llama-3.3-Nemotron-Super-49B-v1": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "source": "https://nebius.com/prices" - }, - "nebius/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000018, - "litellm_provider": "nebius", - "mode": "chat", - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FLlama-3_1-Nemotron-Ultra-253B-v1" - }, - "nebius/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 6e-8, - "output_cost_per_token": 2.4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNVIDIA-Nemotron-3-Nano-30B-A3B" - }, - "nebius/nvidia/Nemotron-3-Nano-Omni": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 6e-8, - "output_cost_per_token": 2.4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3-Nano-Omni" - }, - "nebius/nvidia/Nemotron-3-Ultra-550b-a55b": { - "max_tokens": 1048576, - "max_input_tokens": 1048576, - "max_output_tokens": 1048576, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3-Ultra-550b-a55b" - }, - "nebius/nvidia/Nemotron-3_5-Lightning": { - "max_tokens": 1048576, - "max_input_tokens": 1048576, - "max_output_tokens": 1048576, - "input_cost_per_token": 6e-8, - "output_cost_per_token": 2.4e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3_5-Lightning" - }, - "nebius/nvidia/nemotron-3-super-120b-a12b": { - "max_tokens": 262144, - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "input_cost_per_token": 3e-7, - "output_cost_per_token": 9e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2Fnemotron-3-super-120b-a12b" - }, - "nebius/openai/gpt-oss-120b": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 6e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/openai%2Fgpt-oss-120b" - }, - "nebius/openbmb/MiniCPM-V-4_5": { - "max_tokens": 32000, - "max_input_tokens": 32000, - "max_output_tokens": 32000, - "input_cost_per_token": 6.58e-7, - "output_cost_per_token": 0.00000111, - "litellm_provider": "nebius", - "mode": "chat", - "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/openbmb%2FMiniCPM-V-4_5" - }, - "nebius/zai-org/GLM-5.1": { - "max_tokens": 202752, - "max_input_tokens": 202752, - "max_output_tokens": 202752, - "input_cost_per_token": 0.0000014, - "output_cost_per_token": 0.0000044, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.1" - }, - "nebius/zai-org/GLM-5.2": { - "max_tokens": 1048576, - "max_input_tokens": 1048576, - "max_output_tokens": 1048576, - "input_cost_per_token": 0.0000014, - "output_cost_per_token": 0.0000044, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.2" - }, - "nebius/zai-org/GLM-5.3": { - "input_cost_per_token": 0.0000014, - "litellm_provider": "nebius", - "max_input_tokens": 1024000, - "max_tokens": 1024000, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.3", - "supports_function_calling": true, - "supports_reasoning": true - }, - "nebius/zai-org/GLM-5.3-Flash": { - "max_tokens": 1024000, - "max_input_tokens": 1024000, - "max_output_tokens": 1024000, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 5e-7, - "litellm_provider": "nebius", - "mode": "chat", - "supports_function_calling": true, - "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.3-Flash", - "supports_vision": true - }, - "novita/Sao10K/L3-8B-Stheno-v3.2": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 5e-8, - "output_cost_per_token": 5e-8, - "max_input_tokens": 8192, - "max_output_tokens": 32000, - "max_tokens": 32000, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true - }, - "novita/baichuan/baichuan-m2-32b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 7e-8, - "output_cost_per_token": 7e-8, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "supports_system_messages": true - }, - "novita/baidu/cobuddy": { - "cache_read_input_token_cost": 7e-8, - "input_cost_per_token": 2.8e-7, - "litellm_provider": "novita", - "max_input_tokens": 131072, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.00000113, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/baidu/ernie-4.5-21B-a3b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 7e-8, - "output_cost_per_token": 2.8e-7, - "max_input_tokens": 120000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true - }, - "novita/baidu/ernie-4.5-21B-a3b-thinking": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 7e-8, - "output_cost_per_token": 2.8e-7, - "max_input_tokens": 131072, - "max_output_tokens": 65536, - "max_tokens": 65536, - "supports_system_messages": true, - "supports_reasoning": true - }, - "novita/baidu/ernie-4.5-300b-a47b-paddle": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.8e-7, - "output_cost_per_token": 0.0000011, - "max_input_tokens": 123000, - "max_output_tokens": 12000, - "max_tokens": 12000, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/baidu/ernie-4.5-vl-28b-a3b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.4e-7, - "output_cost_per_token": 5.6e-7, - "max_input_tokens": 30000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_reasoning": true - }, - "novita/baidu/ernie-4.5-vl-28b-a3b-thinking": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3.9e-7, - "output_cost_per_token": 3.9e-7, - "max_input_tokens": 131072, - "max_output_tokens": 65536, - "max_tokens": 65536, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "novita/baidu/ernie-4.5-vl-424b-a47b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 4.2e-7, - "output_cost_per_token": 0.00000125, - "max_input_tokens": 123000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "supports_vision": true, - "supports_system_messages": true, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-ocr": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-8, - "output_cost_per_token": 3e-8, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/deepseek/deepseek-ocr-2": { - "input_cost_per_token": 3e-8, - "litellm_provider": "novita", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 3e-8, - "source": "https://novita.ai/pricing", - "supports_vision": true - }, - "novita/deepseek/deepseek-prover-v2-671b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 7e-7, - "output_cost_per_token": 0.0000025, - "max_input_tokens": 160000, - "max_output_tokens": 160000, - "max_tokens": 160000, - "supports_system_messages": true - }, - "novita/deepseek/deepseek-r1": { - "input_cost_per_token": 0.000004, - "litellm_provider": "novita", - "max_input_tokens": 64000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/deepseek/deepseek-r1-0528": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 7e-7, - "output_cost_per_token": 0.0000025, - "max_input_tokens": 163840, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 3.5e-7, - "input_cost_per_token_cache_hit": 3.5e-7, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-r1-0528-qwen3-8b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 6e-8, - "output_cost_per_token": 9e-8, - "max_input_tokens": 128000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "supports_system_messages": true, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-r1-distill-llama-70b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 8e-7, - "output_cost_per_token": 8e-7, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-r1-distill-qwen-14b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 1.5e-7, - "max_input_tokens": 32768, - "max_output_tokens": 16384, - "max_tokens": 16384, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-r1-distill-qwen-32b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-7, - "output_cost_per_token": 3e-7, - "max_input_tokens": 64000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-r1-turbo": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 7e-7, - "output_cost_per_token": 0.0000025, - "max_input_tokens": 64000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_reasoning": true, - "supports_response_schema": true - }, - "novita/deepseek/deepseek-r1/community": { - "input_cost_per_token": 0.000004, - "litellm_provider": "novita", - "max_input_tokens": 64000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/deepseek/deepseek-v3-0324": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.7e-7, - "output_cost_per_token": 0.00000112, - "max_input_tokens": 163840, - "max_output_tokens": 65536, - "max_tokens": 65536, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 1.35e-7, - "input_cost_per_token_cache_hit": 1.35e-7 - }, - "novita/deepseek/deepseek-v3-turbo": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 4e-7, - "output_cost_per_token": 0.0000013, - "max_input_tokens": 64000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true - }, - "novita/deepseek/deepseek-v3.1": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.7e-7, - "output_cost_per_token": 0.000001, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 1.35e-7, - "input_cost_per_token_cache_hit": 1.35e-7, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-v3.1-terminus": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.7e-7, - "output_cost_per_token": 0.000001, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 1.35e-7, - "input_cost_per_token_cache_hit": 1.35e-7, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-v3.2": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.69e-7, - "output_cost_per_token": 4e-7, - "max_input_tokens": 163840, - "max_output_tokens": 65536, - "max_tokens": 65536, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 1.345e-7, - "input_cost_per_token_cache_hit": 1.345e-7, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-v3.2-exp": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.7e-7, - "output_cost_per_token": 4.1e-7, - "max_input_tokens": 163840, - "max_output_tokens": 65536, - "max_tokens": 65536, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "novita/deepseek/deepseek-v3/community": { - "input_cost_per_token": 8.900000000000001e-7, - "litellm_provider": "novita", - "max_input_tokens": 64000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 8.900000000000001e-7, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/deepseek/deepseek-v4-flash": { - "cache_read_input_token_cost": 2.8e-8, - "input_cost_per_token": 1.4e-7, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 393216, - "max_tokens": 393216, - "mode": "chat", - "output_cost_per_token": 2.8e-7, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/deepseek/deepseek-v4-flash-0731": { - "cache_read_input_token_cost": 2.8e-8, - "input_cost_per_token": 4.4e-7, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 393216, - "max_tokens": 393216, - "mode": "chat", - "output_cost_per_token": 0.00000132, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/deepseek/deepseek-v4-flash-vision-exp": { - "cache_read_input_token_cost": 2.8e-8, - "input_cost_per_token": 4.4e-7, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 393216, - "max_tokens": 393216, - "mode": "chat", - "output_cost_per_token": 0.00000132, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/deepseek/deepseek-v4-pro": { - "cache_read_input_token_cost": 1.35e-7, - "input_cost_per_token": 0.0000016000000000000001, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 393216, - "max_tokens": 393216, - "mode": "chat", - "output_cost_per_token": 0.0000032000000000000003, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/deepseek/deepseek-v4-pro-0813": { - "cache_read_input_token_cost": 1.3200000000000002e-7, - "input_cost_per_token": 0.00000132, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 393216, - "max_tokens": 393216, - "mode": "chat", - "output_cost_per_token": 0.00000396, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/deepseek/deepseek_v3": { - "input_cost_per_token": 8.900000000000001e-7, - "litellm_provider": "novita", - "max_input_tokens": 64000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "output_cost_per_token": 8.900000000000001e-7, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/google/gemma-3-12b-it": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 5e-8, - "output_cost_per_token": 1e-7, - "max_input_tokens": 131072, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/google/gemma-3-27b-it": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.19e-7, - "output_cost_per_token": 2e-7, - "max_input_tokens": 98304, - "max_output_tokens": 16384, - "max_tokens": 16384, - "supports_vision": true, - "supports_system_messages": true - }, - "novita/google/gemma-4-26b-a4b-it": { - "input_cost_per_token": 1.3e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 4.0000000000000003e-7, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/google/gemma-4-31b-it": { - "input_cost_per_token": 1.4e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 4.0000000000000003e-7, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/gryphe/mythomax-l2-13b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 9e-8, - "output_cost_per_token": 9e-8, - "max_input_tokens": 4096, - "max_output_tokens": 3200, - "max_tokens": 3200, - "supports_system_messages": true - }, - "novita/inclusionai/ling-3.0-flash": { - "cache_read_input_token_cost": 1.2e-8, - "input_cost_per_token": 6e-8, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 1.8e-7, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/inclusionai/ling-3.0-flash-fast": { - "cache_read_input_token_cost": 1.2e-8, - "input_cost_per_token": 6e-8, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 1.8e-7, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/kwaipilot/kat-coder-pro": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "max_input_tokens": 256000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 6e-8, - "input_cost_per_token_cache_hit": 6e-8 - }, - "novita/meta-llama/llama-3-70b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 5.1e-7, - "output_cost_per_token": 7.4e-7, - "max_input_tokens": 8192, - "max_output_tokens": 8000, - "max_tokens": 8000, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/meta-llama/llama-3-8b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 4e-8, - "output_cost_per_token": 4e-8, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/meta-llama/llama-3.1-8b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2e-8, - "output_cost_per_token": 5e-8, - "max_input_tokens": 16384, - "max_output_tokens": 16384, - "max_tokens": 16384, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/meta-llama/llama-3.2-1b-instruct": { - "input_cost_per_token": 2e-8, - "litellm_provider": "novita", - "max_input_tokens": 131000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "mode": "chat", - "output_cost_per_token": 2e-8, - "source": "https://api.novita.ai/v3/openai/models", - "supports_response_schema": true, - "supports_vision": false - }, - "novita/meta-llama/llama-3.2-3b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-8, - "output_cost_per_token": 5e-8, - "max_input_tokens": 32768, - "max_output_tokens": 32000, - "max_tokens": 32000, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true - }, - "novita/meta-llama/llama-3.3-70b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.35e-7, - "output_cost_per_token": 4e-7, - "max_input_tokens": 12288, - "max_output_tokens": 12288, - "max_tokens": 12288, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/meta-llama/llama-4-maverick-17b-128e-instruct-fp8": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.7e-7, - "output_cost_per_token": 8.5e-7, - "max_input_tokens": 1048576, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/meta-llama/llama-4-scout-17b-16e-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.8e-7, - "output_cost_per_token": 5.9e-7, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "supports_vision": true, - "supports_system_messages": true - }, - "novita/microsoft/wizardlm-2-8x22b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 6.2e-7, - "output_cost_per_token": 6.2e-7, - "max_input_tokens": 65535, - "max_output_tokens": 8000, - "max_tokens": 8000, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/mindai/macaron-v1-tall": { - "cache_read_input_token_cost": 8e-8, - "input_cost_per_token": 4.5000000000000003e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.0000026, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/mindai/macaron-v1-venti": { - "cache_read_input_token_cost": 3e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000045, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/minimax/minimax-m2": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token_cache_hit": 3e-8, - "supports_reasoning": true, - "supports_response_schema": true - }, - "novita/minimax/minimax-m2.1": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token_cache_hit": 3e-8 - }, - "novita/minimax/minimax-m2.5": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 3e-7, - "litellm_provider": "novita", - "max_input_tokens": 204800, - "max_output_tokens": 131100, - "max_tokens": 131100, - "mode": "chat", - "output_cost_per_token": 0.0000012, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/minimax/minimax-m2.5-highspeed": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 6e-7, - "litellm_provider": "novita", - "max_input_tokens": 204800, - "max_output_tokens": 131100, - "max_tokens": 131100, - "mode": "chat", - "output_cost_per_token": 0.0000024, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/minimax/minimax-m2.7": { - "cache_read_input_token_cost": 6e-8, - "input_cost_per_token": 3e-7, - "litellm_provider": "novita", - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000012, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/minimax/minimax-m2.7-highspeed": { - "cache_read_input_token_cost": 6e-8, - "input_cost_per_token": 6e-7, - "litellm_provider": "novita", - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000024, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/minimax/minimax-m3": { - "cache_read_input_token_cost": 6e-8, - "input_cost_per_token": 3e-7, - "litellm_provider": "novita", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000012, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/minimaxai/minimax-m1-80k": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 5.5e-7, - "output_cost_per_token": 0.0000022, - "max_input_tokens": 1000000, - "max_output_tokens": 40000, - "max_tokens": 40000, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_reasoning": true, - "supports_response_schema": true - }, - "novita/mistralai/mistral-nemo": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 4e-8, - "output_cost_per_token": 1.7e-7, - "max_input_tokens": 60288, - "max_output_tokens": 16000, - "max_tokens": 16000, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/moonshotai/kimi-k2-0905": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000025, - "max_input_tokens": 262144, - "max_output_tokens": 100352, - "max_tokens": 100352, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/moonshotai/kimi-k2-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 5.7e-7, - "output_cost_per_token": 0.0000023, - "max_input_tokens": 131072, - "max_output_tokens": 100352, - "max_tokens": 100352, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/moonshotai/kimi-k2-thinking": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000025, - "max_input_tokens": 262144, - "max_output_tokens": 100352, - "max_tokens": 100352, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true, - "cache_read_input_token_cost": 1.5e-7, - "supports_prompt_caching": true - }, - "novita/moonshotai/kimi-k2.5": { - "cache_read_input_token_cost": 1.0000000000000001e-7, - "input_cost_per_token": 6e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.000003, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/moonshotai/kimi-k2.6": { - "cache_read_input_token_cost": 1.6e-7, - "input_cost_per_token": 8.000000000000001e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.0000034, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/moonshotai/kimi-k2.7-code": { - "cache_read_input_token_cost": 1.9e-7, - "input_cost_per_token": 9.499999999999999e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/moonshotai/kimi-k3": { - "cache_read_input_token_cost": 3e-7, - "input_cost_per_token": 0.000003, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 1048576, - "max_tokens": 1048576, - "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/nousresearch/hermes-2-pro-llama-3-8b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.4e-7, - "output_cost_per_token": 1.4e-7, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/nvidia/nemotron-3-nano-30b-a3b": { - "input_cost_per_token": 5.0000000000000004e-8, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 2.0000000000000002e-7, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/openai/gpt-oss-120b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 5e-8, - "output_cost_per_token": 2.5e-7, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "novita/openai/gpt-oss-20b": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 4e-8, - "output_cost_per_token": 1.5e-7, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_tool_choice": true, - "supports_vision": false, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "novita/paddlepaddle/paddleocr-vl": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2e-8, - "output_cost_per_token": 2e-8, - "max_input_tokens": 16384, - "max_output_tokens": 16384, - "max_tokens": 16384, - "supports_vision": true, - "supports_system_messages": true - }, - "novita/qwen/qwen-2.5-72b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3.8e-7, - "output_cost_per_token": 4e-7, - "max_input_tokens": 32000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen-mt-plus": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 7.5e-7, - "max_input_tokens": 16384, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_system_messages": true - }, - "novita/qwen/qwen2.5-7b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 7e-8, - "output_cost_per_token": 7e-8, - "max_input_tokens": 32000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen2.5-vl-72b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 8e-7, - "output_cost_per_token": 8e-7, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_vision": true, - "supports_system_messages": true - }, - "novita/qwen/qwen3-235b-a22b-fp8": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2e-7, - "output_cost_per_token": 8e-7, - "max_input_tokens": 40960, - "max_output_tokens": 20000, - "max_tokens": 20000, - "supports_system_messages": true, - "supports_reasoning": true, - "supports_response_schema": true - }, - "novita/qwen/qwen3-235b-a22b-instruct-2507": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 9e-8, - "output_cost_per_token": 5.8e-7, - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen3-235b-a22b-thinking-2507": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.000003, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_reasoning": true - }, - "novita/qwen/qwen3-30b-a3b-fp8": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 9e-8, - "output_cost_per_token": 4.5e-7, - "max_input_tokens": 40960, - "max_output_tokens": 20000, - "max_tokens": 20000, - "supports_system_messages": true, - "supports_reasoning": true, - "supports_function_calling": true, - "supports_tool_choice": true - }, - "novita/qwen/qwen3-32b-fp8": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1e-7, - "output_cost_per_token": 4.5e-7, - "max_input_tokens": 40960, - "max_output_tokens": 20000, - "max_tokens": 20000, - "supports_system_messages": true, - "supports_reasoning": true - }, - "novita/qwen/qwen3-4b-fp8": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-8, - "output_cost_per_token": 3e-8, - "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_system_messages": true, - "supports_reasoning": true, - "supports_function_calling": true, - "supports_tool_choice": true - }, - "novita/qwen/qwen3-8b-fp8": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3.5e-8, - "output_cost_per_token": 1.38e-7, - "max_input_tokens": 128000, - "max_output_tokens": 20000, - "max_tokens": 20000, - "supports_system_messages": true, - "supports_reasoning": true - }, - "novita/qwen/qwen3-coder-30b-a3b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 7e-8, - "output_cost_per_token": 2.7e-7, - "max_input_tokens": 160000, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen3-coder-480b-a35b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3.8e-7, - "output_cost_per_token": 0.00000155, - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen3-coder-next": { - "input_cost_per_token": 2.0000000000000002e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/qwen/qwen3-max": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 0.00000211, - "output_cost_per_token": 0.00000845, - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen3-next-80b-a3b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 0.0000015, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "novita/qwen/qwen3-next-80b-a3b-thinking": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 0.0000015, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true - }, - "novita/qwen/qwen3-omni-30b-a3b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 9.7e-7, - "max_input_tokens": 65536, - "max_output_tokens": 16384, - "max_tokens": 16384, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_audio_input": true, - "supports_audio_output": true - }, - "novita/qwen/qwen3-omni-30b-a3b-thinking": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 9.7e-7, - "max_input_tokens": 65536, - "max_output_tokens": 16384, - "max_tokens": 16384, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true, - "supports_reasoning": true, - "supports_audio_input": true - }, - "novita/qwen/qwen3-vl-235b-a22b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000015, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen3-vl-235b-a22b-thinking": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 9.8e-7, - "output_cost_per_token": 0.00000395, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_vision": true, - "supports_system_messages": true, - "supports_reasoning": true, - "supports_function_calling": true, - "supports_tool_choice": true - }, - "novita/qwen/qwen3-vl-30b-a3b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2e-7, - "output_cost_per_token": 7e-7, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen3-vl-30b-a3b-thinking": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2e-7, - "output_cost_per_token": 0.000001, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen3-vl-8b-instruct": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 8e-8, - "output_cost_per_token": 5e-7, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/qwen/qwen3.5-122b-a10b": { - "input_cost_per_token": 4.0000000000000003e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.0000032000000000000003, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/qwen/qwen3.5-27b": { - "input_cost_per_token": 3e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.0000024, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/qwen/qwen3.5-35b-a3b": { - "input_cost_per_token": 2.5e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/qwen/qwen3.5-397b-a17b": { - "input_cost_per_token": 6e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.0000036000000000000003, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/qwen/qwen3.6-27b": { - "input_cost_per_token": 6e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.0000036000000000000003, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/qwen/qwen3.6-35b-a3b": { - "input_cost_per_token": 2.48e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.0000014850000000000002, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/qwen/qwen3.7-max": { - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "novita", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.00000375, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/qwen/qwen3.8-max": { - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.000002, - "litellm_provider": "novita", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000006, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/sao10k/l3-70b-euryale-v2.1": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 0.00000148, - "output_cost_per_token": 0.00000148, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true - }, - "novita/sao10k/l3-8b-lunaris": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 5e-8, - "output_cost_per_token": 5e-8, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/sao10k/l31-70b-euryale-v2.2": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 0.00000148, - "output_cost_per_token": 0.00000148, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true - }, - "novita/skywork/r1v4-lite": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 2e-7, - "output_cost_per_token": 6e-7, - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true - }, - "novita/stepfun/step-3.7-flash": { - "cache_read_input_token_cost": 4e-8, - "input_cost_per_token": 2.0000000000000002e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 256000, - "max_tokens": 256000, - "mode": "chat", - "output_cost_per_token": 0.00000115, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/tencent/hy3": { - "cache_read_input_token_cost": 3.5e-8, - "input_cost_per_token": 1.4e-7, - "litellm_provider": "novita", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 5.8e-7, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/thudm/glm-4-32b-0414": { - "input_cost_per_token": 5.5e-7, - "litellm_provider": "novita", - "max_input_tokens": 32000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "mode": "chat", - "output_cost_per_token": 0.00000166, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/xiaomimimo/mimo-v2-flash": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.1e-7, - "output_cost_per_token": 3.3e-7, - "max_input_tokens": 262144, - "max_output_tokens": 32000, - "max_tokens": 32000, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 2.4e-8, - "input_cost_per_token_cache_hit": 2.4e-8, - "supports_reasoning": true - }, - "novita/xiaomimimo/mimo-v2.5": { - "cache_read_input_token_cost": 3.4e-9, - "input_cost_per_token": 1.6800000000000002e-7, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 3.3600000000000004e-7, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "novita/xiaomimimo/mimo-v2.5-pro": { - "cache_read_input_token_cost": 4.3e-9, - "input_cost_per_token": 5.22e-7, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000001044, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/zai-org/autoglm-phone-9b-multilingual": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3.5e-8, - "output_cost_per_token": 1.38e-7, - "max_input_tokens": 65536, - "max_output_tokens": 65536, - "max_tokens": 65536, - "supports_vision": true, - "supports_system_messages": true - }, - "novita/zai-org/glm-4.5": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000022, - "max_input_tokens": 131072, - "max_output_tokens": 98304, - "max_tokens": 98304, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "cache_read_input_token_cost": 1.1e-7, - "input_cost_per_token_cache_hit": 1.1e-7, - "supports_reasoning": true - }, - "novita/zai-org/glm-4.5-air": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 1.3e-7, - "output_cost_per_token": 8.5e-7, - "max_input_tokens": 131072, - "max_output_tokens": 98304, - "max_tokens": 98304, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_reasoning": true, - "cache_read_input_token_cost": 2.5e-8, - "supports_prompt_caching": true - }, - "novita/zai-org/glm-4.5v": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000018, - "max_input_tokens": 65536, - "max_output_tokens": 16384, - "max_tokens": 16384, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 1.1e-7, - "input_cost_per_token_cache_hit": 1.1e-7, - "supports_reasoning": true - }, - "novita/zai-org/glm-4.6": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 5.5e-7, - "output_cost_per_token": 0.0000022, - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 1.1e-7, - "input_cost_per_token_cache_hit": 1.1e-7, - "supports_reasoning": true - }, - "novita/zai-org/glm-4.6v": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 3e-7, - "output_cost_per_token": 9e-7, - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 5.5e-8, - "input_cost_per_token_cache_hit": 5.5e-8, - "supports_reasoning": true - }, - "novita/zai-org/glm-4.7": { - "litellm_provider": "novita", - "mode": "chat", - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000022, - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, - "supports_system_messages": true, - "supports_response_schema": true, - "cache_read_input_token_cost": 1.1e-7, - "input_cost_per_token_cache_hit": 1.1e-7, - "supports_reasoning": true - }, - "novita/zai-org/glm-4.7-flash": { - "cache_read_input_token_cost": 1e-8, - "input_cost_per_token": 7e-8, - "litellm_provider": "novita", - "max_input_tokens": 200000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 4.0000000000000003e-7, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/zai-org/glm-4.7-h": { - "cache_read_input_token_cost": 1.1e-7, - "input_cost_per_token": 6e-7, - "litellm_provider": "novita", - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000022, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/zai-org/glm-5": { - "cache_read_input_token_cost": 2.0000000000000002e-7, - "input_cost_per_token": 0.000001, - "litellm_provider": "novita", - "max_input_tokens": 202800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000032000000000000003, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/zai-org/glm-5-turbo": { - "cache_read_input_token_cost": 2.4e-7, - "input_cost_per_token": 0.0000012, - "litellm_provider": "novita", - "max_input_tokens": 202800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/zai-org/glm-5.1": { - "cache_read_input_token_cost": 2.6e-7, - "input_cost_per_token": 0.00000138, - "litellm_provider": "novita", - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/zai-org/glm-5.2": { - "cache_read_input_token_cost": 2.6e-7, - "input_cost_per_token": 0.0000014, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "source": "https://novita.ai/pricing", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/zai-org/glm-5.3": { - "cache_read_input_token_cost": 2.6e-7, - "input_cost_per_token": 0.0000014, - "litellm_provider": "novita", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "novita/zai-org/glm-5v-turbo": { - "cache_read_input_token_cost": 2.4e-7, - "input_cost_per_token": 0.0000012, - "litellm_provider": "novita", - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://api.novita.ai/v3/openai/models", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "nscale/Qwen/QwQ-32B": { - "input_cost_per_token": 1.8e-7, - "litellm_provider": "nscale", - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/Qwen/Qwen2.5-Coder-32B-Instruct": { - "input_cost_per_token": 6e-8, - "litellm_provider": "nscale", - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/Qwen/Qwen2.5-Coder-3B-Instruct": { - "input_cost_per_token": 1e-8, - "litellm_provider": "nscale", - "mode": "chat", - "output_cost_per_token": 3e-8, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/Qwen/Qwen2.5-Coder-7B-Instruct": { - "input_cost_per_token": 1e-8, - "litellm_provider": "nscale", - "mode": "chat", - "output_cost_per_token": 3e-8, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-70B": { - "input_cost_per_token": 3.75e-7, - "litellm_provider": "nscale", - "metadata": { - "notes": "Pricing listed as $0.75/1M tokens total. Assumed 50/50 split for input/output." - }, - "mode": "chat", - "output_cost_per_token": 3.75e-7, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-8B": { - "input_cost_per_token": 2.5e-8, - "litellm_provider": "nscale", - "metadata": { - "notes": "Pricing listed as $0.05/1M tokens total. Assumed 50/50 split for input/output." - }, - "mode": "chat", - "output_cost_per_token": 2.5e-8, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": { - "input_cost_per_token": 9e-8, - "litellm_provider": "nscale", - "metadata": { - "notes": "Pricing listed as $0.18/1M tokens total. Assumed 50/50 split for input/output." - }, - "mode": "chat", - "output_cost_per_token": 9e-8, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": { - "input_cost_per_token": 7e-8, - "litellm_provider": "nscale", - "metadata": { - "notes": "Pricing listed as $0.14/1M tokens total. Assumed 50/50 split for input/output." - }, - "mode": "chat", - "output_cost_per_token": 7e-8, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "nscale", - "metadata": { - "notes": "Pricing listed as $0.30/1M tokens total. Assumed 50/50 split for input/output." - }, - "mode": "chat", - "output_cost_per_token": 1.5e-7, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": { - "input_cost_per_token": 2e-7, - "litellm_provider": "nscale", - "metadata": { - "notes": "Pricing listed as $0.40/1M tokens total. Assumed 50/50 split for input/output." - }, - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/meta-llama/Llama-3.1-8B-Instruct": { - "input_cost_per_token": 3e-8, - "litellm_provider": "nscale", - "metadata": { - "notes": "Pricing listed as $0.06/1M tokens total. Assumed 50/50 split for input/output." - }, - "mode": "chat", - "output_cost_per_token": 3e-8, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/meta-llama/Llama-3.3-70B-Instruct": { - "input_cost_per_token": 2e-7, - "litellm_provider": "nscale", - "metadata": { - "notes": "Pricing listed as $0.40/1M tokens total. Assumed 50/50 split for input/output." - }, - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/meta-llama/Llama-4-Scout-17B-16E-Instruct": { - "input_cost_per_token": 9e-8, - "litellm_provider": "nscale", - "mode": "chat", - "output_cost_per_token": 2.9e-7, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nscale/mistralai/mixtral-8x22b-instruct-v0.1": { - "input_cost_per_token": 6e-7, - "litellm_provider": "nscale", - "metadata": { - "notes": "Pricing listed as $1.20/1M tokens total. Assumed 50/50 split for input/output." - }, - "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" - }, - "nvidia.nemotron-nano-12b-v2": { - "input_cost_per_token": 2e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_vision": true - }, - "nvidia.nemotron-nano-3-30b": { - "input_cost_per_token": 6e-8, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 256000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 2.4e-7, - "supports_function_calling": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_native_structured_output": true, - "supports_response_schema": true, - "supports_vision": false - }, - "nvidia.nemotron-nano-9b-v2": { - "input_cost_per_token": 6e-8, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 8000, - "max_tokens": 8000, - "mode": "chat", - "output_cost_per_token": 2.3e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_vision": false - }, - "nvidia.nemotron-super-3-120b": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 256000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "mode": "chat", - "output_cost_per_token": 6.5e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "o1": { - "cache_read_input_token_cost": 0.0000075, - "deprecation_date": "2026-10-23", - "input_cost_per_token": 0.000015, - "input_cost_per_token_batches": 0.0000075, - "litellm_provider": "openai", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "chat", - "output_cost_per_token": 0.00006, - "output_cost_per_token_batches": 0.00003, - "source": "https://developers.openai.com/api/docs/pricing", - "supports_function_calling": true, - "supports_parallel_function_calling": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "o1-2024-12-17": { - "cache_read_input_token_cost": 0.0000075, - "deprecation_date": "2026-10-23", - "input_cost_per_token": 0.000015, - "input_cost_per_token_batches": 0.0000075, - "litellm_provider": "openai", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "chat", - "output_cost_per_token": 0.00006, - "output_cost_per_token_batches": 0.00003, - "source": "https://developers.openai.com/api/docs/pricing", - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "o3": { - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_flex": 2.5e-7, - "cache_read_input_token_cost_priority": 8.75e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_batches": 0.000001, - "input_cost_per_token_flex": 0.000001, - "input_cost_per_token_priority": 0.0000035, - "litellm_provider": "openai", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "chat", - "output_cost_per_token": 0.000008, - "output_cost_per_token_batches": 0.000004, - "output_cost_per_token_flex": 0.000004, - "output_cost_per_token_priority": 0.000014, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "source": "https://developers.openai.com/api/docs/pricing", - "supported_endpoints": [ - "/v1/responses", - "/v1/chat/completions", - "/v1/completions", - "/v1/batch" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "o3-2025-04-16": { - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_flex": 2.5e-7, - "cache_read_input_token_cost_priority": 8.75e-7, - "deprecation_date": "2026-12-11", - "input_cost_per_token": 0.000002, - "input_cost_per_token_batches": 0.000001, - "input_cost_per_token_flex": 0.000001, - "input_cost_per_token_priority": 0.0000035, - "litellm_provider": "openai", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "chat", - "output_cost_per_token": 0.000008, - "output_cost_per_token_batches": 0.000004, - "output_cost_per_token_flex": 0.000004, - "output_cost_per_token_priority": 0.000014, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "source": "https://developers.openai.com/api/docs/pricing", - "supported_endpoints": [ - "/v1/responses", - "/v1/chat/completions", - "/v1/completions", - "/v1/batch" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "o3-mini": { - "cache_read_input_token_cost": 5.5e-7, - "deprecation_date": "2026-10-23", - "input_cost_per_token": 0.0000011, - "input_cost_per_token_batches": 5.5e-7, - "litellm_provider": "openai", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "output_cost_per_token_batches": 0.0000022, - "source": "https://developers.openai.com/api/docs/pricing", - "supports_function_calling": true, - "supports_parallel_function_calling": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "o3-mini-2025-01-31": { - "cache_read_input_token_cost": 5.5e-7, - "deprecation_date": "2026-10-23", - "input_cost_per_token": 0.0000011, - "input_cost_per_token_batches": 5.5e-7, - "litellm_provider": "openai", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "output_cost_per_token_batches": 0.0000022, - "source": "https://developers.openai.com/api/docs/pricing", - "supports_function_calling": true, - "supports_parallel_function_calling": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "o4-mini": { - "cache_read_input_token_cost": 2.75e-7, - "cache_read_input_token_cost_flex": 1.38e-7, - "cache_read_input_token_cost_priority": 5e-7, - "deprecation_date": "2026-10-23", - "input_cost_per_token": 0.0000011, - "input_cost_per_token_batches": 5.5e-7, - "input_cost_per_token_flex": 5.5e-7, - "input_cost_per_token_priority": 0.000002, - "litellm_provider": "openai", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "output_cost_per_token_batches": 0.0000022, - "output_cost_per_token_flex": 0.0000022, - "output_cost_per_token_priority": 0.000008, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "source": "https://developers.openai.com/api/docs/pricing", - "supports_function_calling": true, - "supports_parallel_function_calling": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "o4-mini-2025-04-16": { - "cache_read_input_token_cost": 2.75e-7, - "cache_read_input_token_cost_flex": 1.38e-7, - "cache_read_input_token_cost_priority": 5e-7, - "deprecation_date": "2026-10-23", - "input_cost_per_token": 0.0000011, - "input_cost_per_token_batches": 5.5e-7, - "input_cost_per_token_flex": 5.5e-7, - "input_cost_per_token_priority": 0.000002, - "litellm_provider": "openai", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, - "mode": "chat", - "output_cost_per_token": 0.0000044, - "output_cost_per_token_batches": 0.0000022, - "output_cost_per_token_flex": 0.0000022, - "output_cost_per_token_priority": 0.000008, - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "source": "https://developers.openai.com/api/docs/pricing", - "supports_function_calling": true, - "supports_parallel_function_calling": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "oci/cohere.command-a-03-2025": { - "input_cost_per_token": 0.00000156, - "litellm_provider": "oci", - "max_input_tokens": 256000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 0.00000156, - "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/cohere.command-a-reasoning": { - "input_cost_per_token": 0.00000156, - "litellm_provider": "oci", - "max_input_tokens": 256000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0.00000156, - "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", - "supports_function_calling": false, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/cohere.command-a-reasoning-08-2025": { - "input_cost_per_token": 0.00000156, - "litellm_provider": "oci", - "max_input_tokens": 256000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 0.00000156, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false - }, - "oci/cohere.command-a-translate-08-2025": { - "input_cost_per_token": 9e-8, - "litellm_provider": "oci", - "max_input_tokens": 256000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 9e-8, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": false, - "supports_response_schema": false - }, - "oci/cohere.command-a-vision": { - "input_cost_per_token": 0.00000156, - "litellm_provider": "oci", - "max_input_tokens": 256000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0.00000156, - "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true, - "supports_vision": true - }, - "oci/cohere.command-a-vision-07-2025": { - "input_cost_per_token": 0.00000156, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 0.00000156, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_vision": true - }, - "oci/cohere.command-latest": { - "input_cost_per_token": 0.00000156, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 0.00000156, - "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/cohere.command-plus-latest": { - "input_cost_per_token": 0.00000156, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 0.00000156, - "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/cohere.command-r-08-2024": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 1.5e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false - }, - "oci/cohere.command-r-plus-08-2024": { - "input_cost_per_token": 0.00000156, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 0.00000156, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false - }, - "oci/google.gemini-2.5-flash": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "oci", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_native_streaming": true, - "supports_image_size": false, - "supports_reasoning": true, - "supports_system_messages": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "supports_video_input": true - }, - "oci/google.gemini-2.5-flash-lite": { - "input_cost_per_token": 7.5e-8, - "litellm_provider": "oci", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 3e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_native_streaming": true, - "supports_image_size": false, - "supports_reasoning": false, - "supports_system_messages": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "supports_video_input": true - }, - "oci/google.gemini-2.5-pro": { - "input_cost_per_token": 0.00000125, - "litellm_provider": "oci", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.00001, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_native_streaming": true, - "supports_reasoning": true, - "supports_system_messages": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "supports_video_input": true - }, - "oci/meta.llama-3.1-405b-instruct": { - "input_cost_per_token": 0.00001068, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 0.00001068, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/meta.llama-3.1-70b-instruct": { - "input_cost_per_token": 7.2e-7, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 7.2e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/meta.llama-3.1-8b-instruct": { - "input_cost_per_token": 7.2e-7, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 7.2e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/meta.llama-3.2-11b-vision-instruct": { - "input_cost_per_token": 0.000002, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_vision": true - }, - "oci/meta.llama-3.2-90b-vision-instruct": { - "input_cost_per_token": 0.000002, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true, - "supports_vision": true - }, - "oci/meta.llama-3.3-70b-instruct": { - "input_cost_per_token": 7.2e-7, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 7.2e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/meta.llama-3.3-70b-instruct-fp8-dynamic": { - "input_cost_per_token": 7.2e-7, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, - "mode": "chat", - "output_cost_per_token": 7.2e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false - }, - "oci/meta.llama-4-maverick-17b-128e-instruct-fp8": { - "input_cost_per_token": 7.2e-7, - "litellm_provider": "oci", - "max_input_tokens": 1048576, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 7.2e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true, - "supports_vision": true - }, - "oci/meta.llama-4-scout-17b-16e-instruct": { - "input_cost_per_token": 7.2e-7, - "litellm_provider": "oci", - "max_input_tokens": 10485760, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 7.2e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/openai.gpt-5": { - "input_cost_per_token": 0.00000125, - "litellm_provider": "oci", - "max_input_tokens": 272000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.00001, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true - }, - "oci/openai.gpt-5-mini": { - "input_cost_per_token": 2.5e-7, - "litellm_provider": "oci", - "max_input_tokens": 272000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true - }, - "oci/openai.gpt-5-nano": { - "input_cost_per_token": 5e-8, - "litellm_provider": "oci", - "max_input_tokens": 272000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 4e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true - }, - "oci/xai.grok-3": { - "input_cost_per_token": 0.000003, - "litellm_provider": "oci", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/xai.grok-3-fast": { - "input_cost_per_token": 0.000005, - "litellm_provider": "oci", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000025, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/xai.grok-3-mini": { - "input_cost_per_token": 3e-7, - "litellm_provider": "oci", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 5e-7, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/xai.grok-3-mini-fast": { - "input_cost_per_token": 6e-7, - "litellm_provider": "oci", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/xai.grok-4": { - "input_cost_per_token": 0.000003, - "litellm_provider": "oci", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false, - "supports_native_streaming": true - }, - "oci/xai.grok-4-fast": { - "input_cost_per_token": 0.000005, - "litellm_provider": "oci", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000025, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false - }, - "oci/xai.grok-4.1-fast": { - "input_cost_per_token": 0.000005, - "litellm_provider": "oci", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000025, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false - }, - "oci/xai.grok-4.20": { - "input_cost_per_token": 0.000003, - "litellm_provider": "oci", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false - }, - "oci/xai.grok-4.20-multi-agent": { - "input_cost_per_token": 0.000003, - "litellm_provider": "oci", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false - }, - "oci/xai.grok-code-fast-1": { - "input_cost_per_token": 0.000005, - "litellm_provider": "oci", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.000025, - "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", - "supports_function_calling": true, - "supports_response_schema": false - }, - "ollama/codegeex4": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": false - }, - "ollama/deepseek-coder-v2-instruct": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/deepseek-coder-v2-lite-instruct": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/deepseek-v3.1:671b-cloud": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 163840, - "max_output_tokens": 163840, - "max_tokens": 163840, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/gpt-oss:120b-cloud": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/gpt-oss:20b-cloud": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/internlm2_5-20b-chat": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/llama2": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0 - }, - "ollama/llama2:13b": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0 - }, - "ollama/llama2:70b": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0 - }, - "ollama/llama2:7b": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0 - }, - "ollama/llama3": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0 - }, - "ollama/llama3.1": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/llama3:70b": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0 - }, - "ollama/llama3:8b": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0 - }, - "ollama/mistral-7B-Instruct-v0.1": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/mistral-7B-Instruct-v0.2": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/mistral-large-instruct-2407": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 65536, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/mixtral-8x22B-Instruct-v0.1": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 65536, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/mixtral-8x7B-Instruct-v0.1": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "ollama/qwen3-coder:480b-cloud": { - "input_cost_per_token": 0, - "litellm_provider": "ollama", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "output_cost_per_token": 0, - "supports_function_calling": true - }, - "openai.gpt-6-luna": { - "input_cost_per_token": 1e-7, - "input_cost_per_token_above_272k_tokens": 2e-7, - "cache_creation_input_token_cost": 1.25e-7, - "cache_creation_input_token_cost_above_272k_tokens": 2.5e-7, - "cache_read_input_token_cost": 1e-8, - "cache_read_input_token_cost_above_272k_tokens": 2e-8, - "output_cost_per_token": 5e-7, - "output_cost_per_token_above_272k_tokens": 7.5e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": false, - "supports_none_reasoning_effort": false, - "supports_tool_choice": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_xhigh_reasoning_effort": true, - "supports_vision": true, - "supports_sampling_params": false, - "source": "https://aws.amazon.com/bedrock/pricing/" - }, - "openai.gpt-6-sol": { - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_272k_tokens": 4e-7, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000015, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": false, - "supports_none_reasoning_effort": false, - "supports_tool_choice": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_xhigh_reasoning_effort": true, - "supports_vision": true, - "supports_sampling_params": false, - "source": "https://aws.amazon.com/bedrock/pricing/" - }, - "openai.gpt-6.1-sol": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 1e-7, - "cache_read_input_token_cost_above_272k_tokens": 2e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 1050000, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000015, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-sol.html", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": false, - "supports_none_reasoning_effort": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_sampling_params": false, - "supports_xhigh_reasoning_effort": true - }, - "openai.gpt-oss-120b-1:0": { - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "openai.gpt-oss-20b-1:0": { - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, - "input_cost_per_token": 7e-8, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 3e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "openai.gpt-oss-safeguard-120b": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_vision": false - }, - "openai.gpt-oss-safeguard-20b": { - "input_cost_per_token": 7e-8, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_vision": false - }, - "openai/container": { - "code_interpreter_cost_per_session": 0.03, - "litellm_provider": "openai", - "mode": "chat" - }, - "opencode-go/deepseek-v4-flash": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1000000, - "max_output_tokens": 384000, - "max_tokens": 384000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "opencode-go/deepseek-v4-pro": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1000000, - "max_output_tokens": 384000, - "max_tokens": 384000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "opencode-go/glm-5": { - "litellm_provider": "opencode-go", - "max_input_tokens": 200000, - "max_output_tokens": 128000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true - }, - "opencode-go/glm-5.1": { - "litellm_provider": "opencode-go", - "max_input_tokens": 200000, - "max_output_tokens": 128000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true - }, - "opencode-go/glm-5.2": { - "litellm_provider": "opencode-go", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_reasoning": true - }, - "opencode-go/glm-5.3": { - "litellm_provider": "opencode-go", - "max_input_tokens": 200000, - "max_output_tokens": 128000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true - }, - "opencode-go/gpt-5.6-luna": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_function_calling": true, - "supports_minimal_reasoning_effort": false, - "supports_native_streaming": true, - "supports_none_reasoning_effort": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_xhigh_reasoning_effort": true - }, - "opencode-go/grok-4.5": { - "litellm_provider": "opencode-go", - "max_input_tokens": 500000, - "max_output_tokens": 500000, - "max_tokens": 500000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "opencode-go/hy3": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "opencode-go/hy3-preview": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_native_streaming": true, - "supports_parallel_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "opencode-go/kimi-k2.5": { - "litellm_provider": "opencode-go", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true - }, - "opencode-go/kimi-k2.6": { - "litellm_provider": "opencode-go", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true - }, - "opencode-go/kimi-k2.7-code": { - "litellm_provider": "opencode-go", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_reasoning": true - }, - "opencode-go/kimi-k3": { - "litellm_provider": "opencode-go", - "max_input_tokens": 262144, - "max_output_tokens": 262144, - "max_tokens": 262144, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true - }, - "opencode-go/mimo-v2-omni": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_audio_input": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true - }, - "opencode-go/mimo-v2-pro": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1048576, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "opencode-go/mimo-v2.5": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_audio_input": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true - }, - "opencode-go/mimo-v2.5-pro": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1048576, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "opencode-go/minimax-m2.5": { - "litellm_provider": "opencode-go", - "max_input_tokens": 196608, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_computer_use": false, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": false - }, - "opencode-go/minimax-m2.7": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1000192, - "max_output_tokens": 1000192, - "max_tokens": 1000192, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_assistant_prefill": true, - "supports_function_calling": true, - "supports_reasoning": false - }, - "opencode-go/minimax-m3": { - "litellm_provider": "opencode-go", - "max_input_tokens": 512000, - "max_output_tokens": 512000, - "max_tokens": 512000, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "opencode-go/qwen3.5-plus": { - "litellm_provider": "opencode-go", - "max_input_tokens": 991808, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "opencode-go/qwen3.6-plus": { - "litellm_provider": "opencode-go", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "opencode-go/qwen3.7-max": { - "litellm_provider": "opencode-go", - "max_input_tokens": 991808, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "opencode-go/qwen3.7-plus": { - "litellm_provider": "opencode-go", - "max_input_tokens": 991808, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "opencode-go/qwen3.8-max": { - "litellm_provider": "opencode-go", - "max_input_tokens": 991808, - "max_output_tokens": 65536, - "max_tokens": 65536, - "mode": "chat", - "source": "https://opencode.ai", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true - }, - "openrouter/aion-labs/aion-2.0": { - "cache_read_input_token_cost": 2e-7, - "input_cost_per_token": 8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.0000016, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/aion-labs/aion-3.0": { - "cache_read_input_token_cost": 7.5e-7, - "input_cost_per_token": 0.000003, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.000006, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/aion-labs/aion-3.0-mini": { - "cache_read_input_token_cost": 1.8e-7, - "input_cost_per_token": 7e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.0000014, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/aion-labs/aion-3.5": { - "cache_read_input_token_cost": 7.5e-7, - "input_cost_per_token": 0.000003, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.000006, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/aion-labs/aion-3.5-mini": { - "cache_read_input_token_cost": 1.8e-7, - "input_cost_per_token": 7e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.0000014, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/aion-labs/aion-rp-llama-3.1-8b": { - "input_cost_per_token": 8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 29491, - "max_tokens": 29491, - "mode": "chat", - "output_cost_per_token": 0.0000016, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/amazon/nova-2-lite-v1": { - "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65535, - "max_tokens": 65535, - "mode": "chat", - "output_cost_per_token": 0.0000025, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/amazon/nova-lite-v1": { - "input_cost_per_token": 6e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 300000, - "max_output_tokens": 5120, - "max_tokens": 5120, - "mode": "chat", - "output_cost_per_token": 2.4e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/amazon/nova-micro-v1": { - "input_cost_per_token": 3.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 5120, - "max_tokens": 5120, - "mode": "chat", - "output_cost_per_token": 1.4e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/amazon/nova-premier-v1": { - "cache_read_input_token_cost": 6.25e-7, - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "mode": "chat", - "output_cost_per_token": 0.0000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/amazon/nova-pro-v1": { - "input_cost_per_token": 8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 300000, - "max_output_tokens": 5120, - "max_tokens": 5120, - "mode": "chat", - "output_cost_per_token": 0.0000032, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/anthracite-org/magnum-v4-72b": { - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 4096, - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/anthropic/claude-3-haiku": { - "cache_creation_input_token_cost": 3e-7, - "cache_creation_input_token_cost_above_1hr": 5e-7, - "cache_read_input_token_cost": 3e-8, - "input_cost_per_image": 0.0004, - "input_cost_per_token": 2.5e-7, - "litellm_provider": "openrouter", - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0.00000125, - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true, - "max_input_tokens": 200000, - "max_output_tokens": 4096, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_web_search": true - }, - "openrouter/anthropic/claude-3.5-sonnet": { - "input_cost_per_token": 0.000003, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 0.000015, - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "openrouter/anthropic/claude-3.7-sonnet": { - "input_cost_per_image": 0.0048, - "input_cost_per_token": 0.000003, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000015, - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true - }, - "openrouter/anthropic/claude-fable-5": { - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00005, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "supports_sampling_params": false, - "supports_adaptive_thinking": true, - "thinking_always_on": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 0.000001, - "supports_prompt_caching": true, - "cache_creation_input_token_cost": 0.0000125, - "cache_creation_input_token_cost_above_1hr": 0.00002, - "prompt_cache_min_tokens": 512, - "supports_web_search": true - }, - "openrouter/anthropic/claude-fable-5.1": { - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00005, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "supports_sampling_params": false, - "supports_adaptive_thinking": true, - "thinking_always_on": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 2.5e-7, - "supports_prompt_caching": true, - "cache_creation_input_token_cost": 0.0000125, - "cache_creation_input_token_cost_above_1hr": 0.00002, - "prompt_cache_min_tokens": 512, - "supports_web_search": true - }, - "openrouter/anthropic/claude-fable-5.1:batch": { - "cache_creation_input_token_cost": 0.00000625, - "cache_creation_input_token_cost_above_1hr": 0.00001, - "cache_read_input_token_cost": 1.25e-7, - "input_cost_per_token": 0.000005, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000025, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-fable-5:batch": { - "cache_creation_input_token_cost": 0.00000625, - "cache_creation_input_token_cost_above_1hr": 0.00001, - "cache_read_input_token_cost": 5e-7, - "input_cost_per_token": 0.000005, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000025, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-haiku-4.5": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_1hr": 0.000002, - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 0.000005, - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true, - "prompt_cache_min_tokens": 4096, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": true, - "supports_response_schema": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-haiku-4.5:batch": { - "cache_creation_input_token_cost": 6.25e-7, - "cache_creation_input_token_cost_above_1hr": 0.000001, - "cache_read_input_token_cost": 5e-8, - "input_cost_per_token": 5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 0.0000025, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-haiku-5.5": { - "input_cost_per_token": 1e-7, - "output_cost_per_token": 5e-7, - "cache_read_input_token_cost": 1e-8, - "cache_creation_input_token_cost": 1.25e-7, - "cache_creation_input_token_cost_above_1hr": 2e-7, - "input_cost_per_token_above_100k_tokens": 5e-7, - "output_cost_per_token_above_100k_tokens": 0.0000025, - "cache_read_input_token_cost_above_100k_tokens": 5e-8, - "cache_creation_input_token_cost_above_100k_tokens": 6.25e-7, - "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "search_context_cost_per_query": { - "search_context_size_high": 0.01, - "search_context_size_low": 0.01, - "search_context_size_medium": 0.01 - }, - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_web_search": true, - "supports_adaptive_thinking": true, - "prompt_cache_min_tokens": 512, - "supports_sampling_params": false - }, - "openrouter/anthropic/claude-opus-4.1": { - "input_cost_per_image": 0.0048, - "cache_creation_input_token_cost": 0.00001875, - "cache_creation_input_token_cost_above_1hr": 0.00003, - "cache_read_input_token_cost": 0.0000015, - "input_cost_per_token": 0.000015, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "mode": "chat", - "output_cost_per_token": 0.000075, - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true, - "prompt_cache_min_tokens": 1024, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": true, - "supports_response_schema": false, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-4.1:batch": { - "cache_creation_input_token_cost": 0.000009375, - "cache_creation_input_token_cost_above_1hr": 0.000015, - "cache_read_input_token_cost": 7.5e-7, - "input_cost_per_token": 0.0000075, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 32000, - "max_tokens": 32000, - "mode": "chat", - "output_cost_per_token": 0.0000375, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-4.5": { - "cache_creation_input_token_cost": 0.00000625, - "cache_creation_input_token_cost_above_1hr": 0.00001, - "cache_read_input_token_cost": 5e-7, - "input_cost_per_token": 0.000005, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 0.000025, - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_output_config": true, - "prompt_cache_min_tokens": 4096, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": true, - "supports_response_schema": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-4.5:batch": { - "cache_creation_input_token_cost": 0.000003125, - "cache_creation_input_token_cost_above_1hr": 0.000005, - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 0.0000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-4.6": { - "supports_adaptive_thinking": true, - "supports_legacy_thinking": true, - "cache_creation_input_token_cost": 0.00000625, - "cache_creation_input_token_cost_above_1hr": 0.00001, - "cache_read_input_token_cost": 5e-7, - "input_cost_per_token": 0.000005, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000025, - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_max_reasoning_effort": true, - "supports_tool_choice": true, - "supports_vision": true, - "prompt_cache_min_tokens": 4096, - "supports_response_schema": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-4.6:batch": { - "cache_creation_input_token_cost": 0.000003125, - "cache_creation_input_token_cost_above_1hr": 0.000005, - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.0000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-4.7": { - "supports_adaptive_thinking": true, - "cache_creation_input_token_cost": 0.00000625, - "cache_creation_input_token_cost_above_1hr": 0.00001, - "cache_read_input_token_cost": 5e-7, - "input_cost_per_token": 0.000005, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000025, - "supports_assistant_prefill": false, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_max_reasoning_effort": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_xhigh_reasoning_effort": true, - "prompt_cache_min_tokens": 2048, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-4.7:batch": { - "cache_creation_input_token_cost": 0.000003125, - "cache_creation_input_token_cost_above_1hr": 0.000005, - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.0000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-4.8": { - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "supports_sampling_params": false, - "supports_adaptive_thinking": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 5e-7, - "supports_prompt_caching": true, - "cache_creation_input_token_cost": 0.00000625, - "cache_creation_input_token_cost_above_1hr": 0.00001, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-4.8:batch": { - "cache_creation_input_token_cost": 0.000003125, - "cache_creation_input_token_cost_above_1hr": 0.000005, - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.0000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-5": { - "prompt_cache_min_tokens": 512, - "supports_adaptive_thinking": true, - "cache_creation_input_token_cost": 0.00000625, - "cache_creation_input_token_cost_above_1hr": 0.00001, - "cache_read_input_token_cost": 5e-7, - "input_cost_per_token": 0.000005, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000025, - "source": "https://openrouter.ai/api/v1/models", - "supports_assistant_prefill": false, - "supports_audio_input": false, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_max_reasoning_effort": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_xhigh_reasoning_effort": true - }, - "openrouter/anthropic/claude-opus-5.5": { - "cache_creation_input_token_cost": 0.000005, - "cache_creation_input_token_cost_above_1hr": 0.000008, - "cache_read_input_token_cost": 2e-7, - "input_cost_per_token": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.00002, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-5.5:batch": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_1hr": 0.000004, - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.00001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-opus-5:batch": { - "cache_creation_input_token_cost": 0.000003125, - "cache_creation_input_token_cost_above_1hr": 0.000005, - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.0000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-sonnet-4": { - "input_cost_per_image": 0.0048, - "cache_creation_input_token_cost": 0.00000375, - "cache_creation_input_token_cost_above_1hr": 0.000006, - "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, - "cache_read_input_token_cost": 3e-7, - "cache_read_input_token_cost_above_200k_tokens": 6e-7, - "input_cost_per_token": 0.000003, - "input_cost_per_token_above_200k_tokens": 0.000006, - "output_cost_per_token_above_200k_tokens": 0.0000225, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 0.000015, - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true, - "prompt_cache_min_tokens": 1024, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": true, - "supports_response_schema": false, - "supports_web_search": true - }, - "openrouter/anthropic/claude-sonnet-4.5": { - "input_cost_per_image": 0.0048, - "cache_creation_input_token_cost": 0.00000375, - "cache_creation_input_token_cost_above_1hr": 0.000006, - "cache_read_input_token_cost": 3e-7, - "input_cost_per_token": 0.000003, - "input_cost_per_token_above_200k_tokens": 0.000006, - "output_cost_per_token_above_200k_tokens": 0.0000225, - "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, - "cache_read_input_token_cost_above_200k_tokens": 6e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 0.000015, - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true, - "prompt_cache_min_tokens": 1024, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": true, - "supports_response_schema": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-sonnet-4.5:batch": { - "cache_creation_input_token_cost": 0.000001875, - "cache_creation_input_token_cost_above_1hr": 0.000003, - "cache_creation_input_token_cost_above_200k_tokens": 0.00000375, - "cache_read_input_token_cost": 1.5e-7, - "cache_read_input_token_cost_above_200k_tokens": 3e-7, - "input_cost_per_token": 0.0000015, - "input_cost_per_token_above_200k_tokens": 0.000003, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 64000, - "max_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "output_cost_per_token_above_200k_tokens": 0.00001125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-sonnet-4.6": { - "supports_adaptive_thinking": true, - "supports_legacy_thinking": true, - "cache_creation_input_token_cost": 0.00000375, - "cache_creation_input_token_cost_above_1hr": 0.000006, - "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, - "cache_read_input_token_cost": 3e-7, - "cache_read_input_token_cost_above_200k_tokens": 6e-7, - "input_cost_per_token": 0.000003, - "input_cost_per_token_above_200k_tokens": 0.000006, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000015, - "output_cost_per_token_above_200k_tokens": 0.0000225, - "source": "https://openrouter.ai/api/v1/models", - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_max_reasoning_effort": true, - "supports_tool_choice": true, - "supports_vision": true, - "prompt_cache_min_tokens": 1024, - "supports_audio_input": false, - "supports_pdf_input": true, - "supports_response_schema": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-sonnet-4.6:batch": { - "cache_creation_input_token_cost": 0.000001875, - "cache_creation_input_token_cost_above_1hr": 0.000003, - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.0000015, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/anthropic/claude-sonnet-5": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.00001, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "supports_sampling_params": false, - "supports_adaptive_thinking": true, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.00000115, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, + "supports_native_streaming": true, + "supports_reasoning": false, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 2e-7, - "supports_prompt_caching": true, - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_1hr": 0.000004, - "supports_web_search": true + "supports_vision": true }, - "openrouter/anthropic/claude-sonnet-5.5": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.00001, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "inferenceapis/qwen3-vl-235b": { + "input_cost_per_token": 2.6e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "supports_sampling_params": false, - "supports_adaptive_thinking": true, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.00000115, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, + "supports_native_streaming": true, + "supports_reasoning": false, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 2e-7, - "supports_prompt_caching": true, - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_1hr": 0.000004, - "supports_web_search": true + "supports_vision": true }, - "openrouter/anthropic/claude-sonnet-5.5:batch": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_1hr": 0.000002, - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", + "inferenceapis/qwen3.8-flash": { + "cache_read_input_token_cost": 2e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "inferenceapis", "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "max_tokens": 1000000, "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 5e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/anthropic/claude-sonnet-5:batch": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_1hr": 0.000002, - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "inferenceapis/zai-org/GLM-5.3": { + "cache_read_input_token_cost": 3.4e-7, + "input_cost_per_token": 0.00000182, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048575, + "max_tokens": 1048575, "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000572, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true - }, - "openrouter/apodex/apodex-1.1-mini:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": false }, - "openrouter/arcee-ai/trinity-large-thinking": { - "cache_read_input_token_cost": 6e-8, - "input_cost_per_token": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 80000, - "max_tokens": 80000, + "inferenceapis/zai-org/GLM-5.3-Flash": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 2e-7, + "litellm_provider": "inferenceapis", + "max_input_tokens": 1048575, + "max_tokens": 1048575, "mode": "chat", - "output_cost_per_token": 8e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 6.6e-7, + "source": "https://inferenceapis.com/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], "supports_function_calling": true, - "supports_pdf_input": false, + "supports_native_streaming": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/baidu/ernie-4.5-vl-424b-a47b": { - "deprecation_date": "2026-10-08", - "input_cost_per_token": 4.2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 123000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "output_cost_per_token": 0.00000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/bytedance-seed/seed-1.6": { - "deprecation_date": "2026-11-11", - "input_cost_per_token": 2.5e-7, - "input_cost_per_token_above_128k_tokens": 5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.000002, - "output_cost_per_token_above_128k_tokens": 0.000004, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/bytedance-seed/seed-1.6-flash": { - "deprecation_date": "2026-11-11", - "input_cost_per_token": 7.5e-8, - "input_cost_per_token_above_128k_tokens": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 3e-7, - "output_cost_per_token_above_128k_tokens": 8e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/bytedance-seed/seed-2-1-turbo": { - "input_cost_per_token": 5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "jamba-1.5": { + "input_cost_per_token": 2e-7, + "litellm_provider": "ai21", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 0.0000025, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 4e-7, + "supports_tool_choice": true }, - "openrouter/bytedance-seed/seed-2.0-code": { - "deprecation_date": "2026-11-11", - "input_cost_per_token": 5e-7, - "input_cost_per_token_above_128k_tokens": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 131072, - "max_tokens": 131072, + "jamba-1.5-large": { + "input_cost_per_token": 0.000002, + "litellm_provider": "ai21", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 0.000003, - "output_cost_per_token_above_128k_tokens": 0.000006, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 0.000008, + "supports_tool_choice": true }, - "openrouter/bytedance-seed/seed-2.0-lite": { - "input_cost_per_token": 2.5e-7, - "input_cost_per_token_above_128k_tokens": 5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 131072, - "max_tokens": 131072, + "jamba-1.5-large@001": { + "input_cost_per_token": 0.000002, + "litellm_provider": "ai21", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 0.000002, - "output_cost_per_token_above_128k_tokens": 0.000004, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 0.000008, + "supports_tool_choice": true }, - "openrouter/bytedance-seed/seed-2.0-mini": { - "input_cost_per_token": 1e-7, - "input_cost_per_token_above_128k_tokens": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 131072, - "max_tokens": 131072, + "jamba-1.5-mini": { + "input_cost_per_token": 2e-7, + "litellm_provider": "ai21", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", "output_cost_per_token": 4e-7, - "output_cost_per_token_above_128k_tokens": 8e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/bytedance/ui-tars-1.5-7b": { - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 2048, - "max_tokens": 2048, - "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/cognitivecomputations/dolphin-mistral-24b-venice-edition": { + "jamba-1.5-mini@001": { "input_cost_per_token": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "output_cost_per_token": 9e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/cohere/command-a": { - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", + "litellm_provider": "ai21", "max_input_tokens": 256000, - "max_output_tokens": 8192, - "max_tokens": 8192, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 0.00001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 4e-7, + "supports_tool_choice": true }, - "openrouter/cohere/command-a-plus": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 192000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "jamba-large-1.6": { + "input_cost_per_token": 0.000002, + "litellm_provider": "ai21", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 0.000008, + "supports_tool_choice": true }, - "openrouter/cohere/command-r-08-2024": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, + "jamba-large-1.7": { + "input_cost_per_token": 0.000002, + "litellm_provider": "ai21", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 0.000008, + "supports_tool_choice": true }, - "openrouter/cohere/command-r-plus-08-2024": { - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, + "jamba-mini-1.6": { + "input_cost_per_token": 2e-7, + "litellm_provider": "ai21", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 0.00001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 4e-7, + "supports_tool_choice": true }, - "openrouter/cohere/command-r7b-12-2024": { - "input_cost_per_token": 3.75e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 4000, - "max_tokens": 4000, + "jamba-mini-1.7": { + "input_cost_per_token": 2e-7, + "litellm_provider": "ai21", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 1.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 4e-7, + "supports_tool_choice": true }, - "openrouter/cohere/north-mini-code:free": { - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 256000, + "jp.amazon.nova-2-lite-v1:0": { + "cache_read_input_token_cost": 9.9e-8, + "cache_read_input_token_cost_flex": 4.95e-8, + "cache_read_input_token_cost_priority": 1.7325e-7, + "input_cost_per_token": 3.96e-7, + "input_cost_per_token_batches": 1.98e-7, + "input_cost_per_token_flex": 1.98e-7, + "input_cost_per_token_priority": 6.93e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/deepseek/deepseek-chat": { - "input_cost_per_token": 2.574e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 163840, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", - "output_cost_per_token": 0.0000010287, - "supports_prompt_caching": false, - "supports_tool_choice": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/deepseek/deepseek-chat-v3-0324": { - "cache_read_input_token_cost": 1.1e-7, - "input_cost_per_token": 2.9e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 163840, - "max_output_tokens": 115200, - "max_tokens": 115200, - "mode": "chat", - "output_cost_per_token": 0.00000114, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000003311, + "output_cost_per_token_batches": 0.0000016555, + "output_cost_per_token_flex": 0.0000016555, + "output_cost_per_token_priority": 0.00000579425, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/ap-northeast-1/index.json", "supports_function_calling": true, - "supports_pdf_input": false, + "supports_pdf_input": true, "supports_prompt_caching": true, - "supports_reasoning": false, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_video_input": true, + "supports_vision": true }, - "openrouter/deepseek/deepseek-chat-v3.1": { - "input_cost_per_token": 2.5e-7, - "input_cost_per_token_cache_hit": 2e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 163840, - "max_output_tokens": 32768, - "max_tokens": 32768, + "jp.anthropic.claude-haiku-4-5-20251001-v1:0": { + "cache_creation_input_token_cost": 0.000001375, + "cache_creation_input_token_cost_above_1hr": 0.0000022, + "cache_read_input_token_cost": 1.1e-7, + "input_cost_per_token": 0.0000011, + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", - "output_cost_per_token": 9.5e-7, + "output_cost_per_token": 0.0000055, + "source": "https://aws.amazon.com/bedrock/pricing/", "supports_assistant_prefill": true, + "supports_computer_use": true, "supports_function_calling": true, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_tool_choice": true, - "cache_read_input_token_cost": 1.3e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": false, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true, + "supports_native_structured_output": true, + "supports_parallel_tool_use_config": true, + "prompt_cache_min_tokens": 4096, + "input_cost_per_token_batches": 5.5e-7, + "output_cost_per_token_batches": 0.00000275 }, - "openrouter/deepseek/deepseek-r1": { - "cache_read_input_token_cost": 1.4e-7, - "input_cost_per_token": 7e-7, - "input_cost_per_token_cache_hit": 1.4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 64000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "jp.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-7, + "output_cost_per_token_above_100k_tokens": 0.00000275, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-7, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 0.0000011, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-8, + "bedrock_converse_supports_strict_tools": false, + "cache_creation_input_token_cost": 1.375e-7, + "cache_creation_input_token_cost_above_1hr": 2.2e-7, + "cache_read_input_token_cost": 1.1e-8, + "input_cost_per_token": 1.1e-7, + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.0000025, - "source": "https://openrouter.ai/api/v1/models", - "supports_assistant_prefill": true, - "supports_audio_input": false, + "output_cost_per_token": 5.5e-7, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_mid_conversation_system": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh", + "supports_parallel_tool_use_config": true, + "supports_forced_tool_use": true, + "thinking_always_on": false, + "prompt_cache_min_tokens": 512, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json" }, - "openrouter/deepseek/deepseek-r1-0528": { - "cache_read_input_token_cost": 3.5e-7, - "input_cost_per_token": 5e-7, - "input_cost_per_token_cache_hit": 1.4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 163840, - "max_output_tokens": 32768, - "max_tokens": 32768, + "jp.anthropic.claude-opus-4-7": { + "bedrock_converse_supports_strict_tools": false, + "cache_creation_input_token_cost": 0.000006875, + "cache_creation_input_token_cost_above_1hr": 0.000011, + "cache_read_input_token_cost": 5.5e-7, + "input_cost_per_token": 0.0000055, + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.00000215, - "source": "https://openrouter.ai/api/v1/models", - "supports_assistant_prefill": true, - "supports_audio_input": false, + "output_cost_per_token": 0.0000275, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": false, + "supports_max_reasoning_effort": true, + "supports_minimal_reasoning_effort": true, + "supports_parallel_tool_use_config": true, + "prompt_cache_min_tokens": 2048, + "source": "https://aws.amazon.com/bedrock/pricing/" }, - "openrouter/deepseek/deepseek-r1-distill-llama-70b": { - "deprecation_date": "2026-09-28", - "input_cost_per_token": 8e-7, - "output_cost_per_token": 8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 8192, - "max_output_tokens": 7372, - "max_tokens": 7372, + "jp.anthropic.claude-opus-4-8": { + "bedrock_converse_supports_strict_tools": false, + "supports_adaptive_thinking": true, + "supports_mid_conversation_system": true, + "cache_creation_input_token_cost": 0.000006875, + "cache_creation_input_token_cost_above_1hr": 0.000011, + "cache_read_input_token_cost": 5.5e-7, + "input_cost_per_token": 0.0000055, + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": false, + "output_cost_per_token": 0.0000275, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_vision": false, - "supports_web_search": false + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh", + "supports_parallel_tool_use_config": true, + "prompt_cache_min_tokens": 1024, + "source": "https://aws.amazon.com/bedrock/pricing/" }, - "openrouter/deepseek/deepseek-v3.1-terminus": { - "cache_read_input_token_cost": 1.35e-7, - "deprecation_date": "2026-09-28", - "input_cost_per_token": 2.7e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 163840, - "max_output_tokens": 147456, - "max_tokens": 147456, + "jp.anthropic.claude-opus-5": { + "bedrock_converse_supports_strict_tools": false, + "supports_adaptive_thinking": true, + "supports_mid_conversation_system": true, + "cache_creation_input_token_cost": 0.000006875, + "cache_creation_input_token_cost_above_1hr": 0.000011, + "cache_read_input_token_cost": 5.5e-7, + "input_cost_per_token": 0.0000055, + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.000001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000275, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": false, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "supports_parallel_tool_use_config": true, + "prompt_cache_min_tokens": 512, + "source": "https://aws.amazon.com/bedrock/pricing/" }, - "openrouter/deepseek/deepseek-v3.2": { - "cache_read_input_token_cost": 2.8e-8, - "deprecation_date": "2026-09-28", - "input_cost_per_token": 2.8e-7, - "input_cost_per_token_cache_hit": 2.8e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 163840, - "max_output_tokens": 65536, - "max_tokens": 65536, + "jp.anthropic.claude-opus-5-5": { + "bedrock_converse_supports_strict_tools": false, + "supports_adaptive_thinking": true, + "supports_mid_conversation_system": true, + "cache_creation_input_token_cost": 0.0000055, + "cache_creation_input_token_cost_above_1hr": 0.0000088, + "cache_read_input_token_cost": 2.2e-7, + "input_cost_per_token": 0.0000044, + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 4.2e-7, - "supports_assistant_prefill": true, + "output_cost_per_token": 0.000022, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, "supports_function_calling": true, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_tool_choice": true, "supports_response_schema": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": false, - "supports_vision": false, - "supports_web_search": false + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": false, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "supports_parallel_tool_use_config": true, + "prompt_cache_min_tokens": 512, + "thinking_always_on": true, + "supports_forced_tool_use": false, + "source": "https://aws.amazon.com/bedrock/pricing/" }, - "openrouter/deepseek/deepseek-v3.2-exp": { - "cache_read_input_token_cost": 2e-8, - "deprecation_date": "2026-09-28", - "input_cost_per_token": 2.7e-7, - "input_cost_per_token_cache_hit": 2e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 163840, - "max_output_tokens": 147456, - "max_tokens": 147456, + "jp.anthropic.claude-sonnet-4-5-20250929-v1:0": { + "cache_creation_input_token_cost": 0.000004125, + "cache_creation_input_token_cost_above_1hr": 0.0000066, + "cache_read_input_token_cost": 3.3e-7, + "input_cost_per_token": 0.0000033, + "input_cost_per_token_above_200k_tokens": 0.0000066, + "output_cost_per_token_above_200k_tokens": 0.00002475, + "cache_creation_input_token_cost_above_200k_tokens": 0.00000825, + "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.0000132, + "cache_read_input_token_cost_above_200k_tokens": 6.6e-7, + "deprecation_date": "2027-04-08", + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", - "output_cost_per_token": 4.1e-7, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000165, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, "supports_assistant_prefill": true, - "supports_audio_input": false, + "supports_computer_use": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_native_structured_output": true, + "supports_parallel_tool_use_config": true, + "prompt_cache_min_tokens": 1024, + "input_cost_per_token_batches": 0.00000165, + "output_cost_per_token_batches": 0.00000825, + "source": "https://aws.amazon.com/bedrock/pricing/" }, - "openrouter/deepseek/deepseek-v4-flash": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 3e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "jp.anthropic.claude-sonnet-4-6": { + "supports_adaptive_thinking": true, + "supports_legacy_thinking": true, + "cache_creation_input_token_cost": 0.000004125, + "cache_creation_input_token_cost_above_1hr": 0.0000066, + "cache_read_input_token_cost": 3.3e-7, + "input_cost_per_token": 0.0000033, + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 1000000, + "max_output_tokens": 64000, + "max_tokens": 64000, "mode": "chat", - "output_cost_per_token": 0.00000128, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000165, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_max_reasoning_effort": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_native_structured_output": true, + "supports_output_config": true, + "supports_parallel_tool_use_config": true, + "prompt_cache_min_tokens": 1024, + "source": "https://aws.amazon.com/bedrock/pricing/" }, - "openrouter/deepseek/deepseek-v4-flash-0731": { - "cache_read_input_token_cost": 1.63e-8, - "input_cost_per_token": 1.63e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "jp.anthropic.claude-sonnet-5": { + "bedrock_converse_supports_strict_tools": false, + "cache_creation_input_token_cost": 0.00000275, + "cache_creation_input_token_cost_above_1hr": 0.0000044, + "cache_read_input_token_cost": 2.2e-7, + "input_cost_per_token": 0.0000022, + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.00000128, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000011, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_mid_conversation_system": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": false, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh", + "supports_parallel_tool_use_config": true, + "prompt_cache_min_tokens": 1024, + "source": "https://aws.amazon.com/bedrock/pricing/" }, - "openrouter/deepseek/deepseek-v4-flash-vision-exp": { - "input_cost_per_token": 2.156e-7, - "output_cost_per_token": 6.468e-7, - "cache_read_input_token_cost": 6.86e-9, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 262144, - "max_tokens": 262144, + "jp.anthropic.claude-sonnet-5-5": { + "bedrock_converse_supports_strict_tools": false, + "cache_creation_input_token_cost": 0.00000275, + "cache_creation_input_token_cost_above_1hr": 0.0000044, + "cache_read_input_token_cost": 1.1e-7, + "input_cost_per_token": 0.0000022, + "litellm_provider": "bedrock_converse", + "supports_tool_search": true, + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000011, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_mid_conversation_system": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh", + "supports_parallel_tool_use_config": true, + "supports_forced_tool_use": false, + "thinking_always_on": true, + "prompt_cache_min_tokens": 512, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json" }, - "openrouter/deepseek/deepseek-v4-pro": { - "cache_read_input_token_cost": 1.74e-8, - "input_cost_per_token": 2.088e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 384000, - "max_tokens": 384000, + "kimi-k2-thinking-251104": { + "input_cost_per_token": 0, + "litellm_provider": "volcengine", + "max_input_tokens": 229376, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", - "output_cost_per_token": 4.176e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0, + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/deepseek/deepseek-v4-pro-0813": { - "cache_read_input_token_cost": 4.4e-8, - "input_cost_per_token": 0.00000132, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 393216, - "max_tokens": 393216, + "lambda_ai/deepseek-llama3.3-70b": { + "input_cost_per_token": 2e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "off_peak_pricing": { - "input_cost_per_token": 6.6e-7, - "output_cost_per_token": 0.00000198, - "cache_read_input_token_cost": 2.2e-8, - "windows": [ - { - "hours_utc": "00:00-00:00", - "weekdays": [ - "saturday", - "sunday" - ] - }, - { - "hours_utc": "00:00-01:00", - "weekdays": [ - "monday", - "tuesday", - "wednesday", - "thursday", - "friday" - ] - }, - { - "hours_utc": "04:00-06:00", - "weekdays": [ - "monday", - "tuesday", - "wednesday", - "thursday", - "friday" - ] - }, - { - "hours_utc": "10:00-00:00", - "weekdays": [ - "monday", - "tuesday", - "wednesday", - "thursday", - "friday" - ] - } - ] - }, - "output_cost_per_token": 0.00000396, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 6e-7, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_tool_choice": true + }, + "lambda_ai/deepseek-r1-0528": { + "input_cost_per_token": 2e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 6e-7, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_tool_choice": true + }, + "lambda_ai/deepseek-r1-671b": { + "input_cost_per_token": 8e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 8e-7, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_tool_choice": true + }, + "lambda_ai/deepseek-v3-0324": { + "input_cost_per_token": 2e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 6e-7, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/deepseek/deepseek-v4.1-flash": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 4.4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "lambda_ai/hermes3-405b": { + "input_cost_per_token": 8e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 3e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 8e-7, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/deepseek/deepseek-v4.1-flash:batch": { - "cache_read_input_token_cost": 3.36e-9, - "input_cost_per_token": 1.12e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, + "lambda_ai/hermes3-70b": { + "input_cost_per_token": 1.2e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 3.36e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 3e-7, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/dots-studio/dots-3-note-preview:free": { - "deprecation_date": "2026-12-31", - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 512000, - "max_output_tokens": 460800, - "max_tokens": 460800, + "lambda_ai/hermes3-8b": { + "input_cost_per_token": 2.5e-8, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 4e-8, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/fireworks/ember-1": { - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_read_input_token_cost": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "lambda_ai/lfm-40b": { + "input_cost_per_token": 1e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 2e-7, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_parallel_function_calling": false, - "supports_pdf_input": false, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-2.5-flash": { - "cache_creation_input_token_cost": 8.33333333333333e-8, - "cache_read_input_audio_token_cost": 1e-7, - "input_cost_per_audio_token": 0.000001, - "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65535, - "max_tokens": 65535, + "lambda_ai/lfm-7b": { + "input_cost_per_token": 2.5e-8, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.0000025, - "supports_audio_output": false, + "output_cost_per_token": 4e-8, "supports_function_calling": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_image_size": false, - "cache_read_input_token_cost": 3e-8, - "deprecation_date": "2026-10-20", - "supports_prompt_caching": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, - "supports_pdf_input": true, - "supports_reasoning": true, - "supports_web_search": true, - "supports_video_input": true + "supports_tool_choice": true }, - "openrouter/google/gemini-2.5-flash-image": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000025, - "cache_read_input_token_cost": 3e-8, - "cache_creation_input_token_cost": 8.33333333333333e-8, - "cache_read_input_audio_token_cost": 1e-7, - "deprecation_date": "2027-03-15", - "input_cost_per_audio_token": 0.000001, - "output_cost_per_image_token": 0.00003, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, + "lambda_ai/llama-4-maverick-17b-128e-instruct-fp8": { + "input_cost_per_token": 5e-8, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_tool_choice": false, - "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": true + "output_cost_per_token": 1e-7, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-2.5-flash-lite": { - "cache_creation_input_token_cost": 8.33333333333333e-8, - "cache_read_input_audio_token_cost": 3e-8, - "input_cost_per_token": 1e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65535, - "max_tokens": 65535, + "lambda_ai/llama-4-scout-17b-16e-instruct": { + "input_cost_per_token": 5e-8, + "litellm_provider": "lambda_ai", + "max_input_tokens": 16384, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 1e-7, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "cache_read_input_token_cost": 1e-8, - "deprecation_date": "2026-10-20", - "input_cost_per_audio_token": 3e-7, - "supports_prompt_caching": true, - "supports_web_search": true, - "supports_video_input": true + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-2.5-flash-lite:batch": { - "cache_read_input_audio_token_cost": 3e-8, - "cache_read_input_token_cost": 1e-8, - "input_cost_per_audio_token": 1.5e-7, - "input_cost_per_token": 5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65535, - "max_tokens": 65535, + "lambda_ai/llama3.1-405b-instruct-fp8": { + "input_cost_per_token": 8e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 8e-7, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-2.5-flash:batch": { - "cache_read_input_audio_token_cost": 1e-7, - "cache_read_input_token_cost": 3e-8, - "deprecation_date": "2026-10-20", - "input_cost_per_audio_token": 5e-7, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65535, - "max_tokens": 65535, + "lambda_ai/llama3.1-70b-instruct-fp8": { + "input_cost_per_token": 1.2e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.00000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 3e-7, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-2.5-pro": { - "cache_creation_input_token_cost": 3.75e-7, - "cache_read_input_audio_token_cost": 1.25e-7, - "input_cost_per_audio_token": 0.00000125, - "input_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "lambda_ai/llama3.1-8b-instruct": { + "input_cost_per_token": 2.5e-8, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.00001, - "supports_audio_output": false, + "output_cost_per_token": 4e-8, "supports_function_calling": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_system_messages": true, - "supports_tool_choice": true, - "supports_vision": true, - "cache_read_input_token_cost": 1.25e-7, - "cache_read_input_token_cost_above_200k_tokens": 2.5e-7, - "deprecation_date": "2026-10-20", - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token_above_200k_tokens": 0.000015, - "supports_prompt_caching": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, - "supports_pdf_input": true, - "supports_reasoning": true, - "supports_web_search": true, - "supports_video_input": true + "supports_tool_choice": true }, - "openrouter/google/gemini-2.5-pro-preview": { - "input_cost_per_token": 0.00000125, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 1.25e-7, - "cache_creation_input_token_cost": 3.75e-7, - "cache_read_input_audio_token_cost": 1.25e-7, - "input_cost_per_audio_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token_above_200k_tokens": 0.000015, - "cache_read_input_token_cost_above_200k_tokens": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "lambda_ai/llama3.1-nemotron-70b-instruct-fp8": { + "input_cost_per_token": 1.2e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 3e-7, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-2.5-pro-preview-05-06": { - "input_cost_per_token": 0.00000125, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 1.25e-7, - "cache_creation_input_token_cost": 3.75e-7, - "input_cost_per_audio_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token_above_200k_tokens": 0.000015, - "cache_read_input_token_cost_above_200k_tokens": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65535, - "max_tokens": 65535, + "lambda_ai/llama3.2-11b-vision-instruct": { + "input_cost_per_token": 1.5e-8, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/google/gemini-2.5-pro-preview-05-06", + "output_cost_per_token": 2.5e-8, "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "supports_prompt_caching": true + "supports_vision": true }, - "openrouter/google/gemini-2.5-pro:batch": { - "cache_read_input_audio_token_cost": 1.25e-7, - "cache_read_input_token_cost": 1.25e-7, - "cache_read_input_token_cost_above_200k_tokens": 2.5e-7, - "deprecation_date": "2026-10-20", - "input_cost_per_audio_token": 6.25e-7, - "input_cost_per_token": 6.25e-7, - "input_cost_per_token_above_200k_tokens": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "lambda_ai/llama3.2-3b-instruct": { + "input_cost_per_token": 1.5e-8, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.000005, - "output_cost_per_token_above_200k_tokens": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 2.5e-8, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-3-flash-preview": { - "cache_creation_input_token_cost": 8.33333333333333e-8, - "cache_read_input_audio_token_cost": 1e-7, - "cache_read_input_token_cost": 5e-8, - "input_cost_per_audio_token": 0.000001, - "input_cost_per_token": 5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "lambda_ai/llama3.3-70b-instruct-fp8": { + "input_cost_per_token": 1.2e-7, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_reasoning_token": 0.000003, - "output_cost_per_token": 0.000003, - "rpm": 2000, - "source": "https://openrouter.ai/api/v1/models", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/completions", - "/v1/batch" - ], - "supported_modalities": [ - "text", - "image", - "audio", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": true, - "supports_audio_output": false, + "output_cost_per_token": 3e-7, "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, "supports_system_messages": true, - "supports_tool_choice": true, - "supports_url_context": true, - "supports_vision": true, - "supports_web_search": true, - "tpm": 800000, - "supports_video_input": true + "supports_tool_choice": true }, - "openrouter/google/gemini-3-flash-preview:batch": { - "input_cost_per_audio_token": 5e-7, - "input_cost_per_token": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "lambda_ai/qwen25-coder-32b-instruct": { + "input_cost_per_token": 5e-8, + "litellm_provider": "lambda_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 1e-7, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_parallel_function_calling": true, + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-3-pro-image": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000012, - "cache_read_input_token_cost": 2e-7, - "cache_creation_input_token_cost": 3.75e-7, - "cache_read_input_audio_token_cost": 2e-7, - "input_cost_per_audio_token": 0.000002, - "output_cost_per_image_token": 0.00012, - "litellm_provider": "openrouter", + "lambda_ai/qwen3-32b-fp8": { + "input_cost_per_token": 5e-8, + "litellm_provider": "lambda_ai", "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 1e-7, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, + "supports_parallel_function_calling": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-3-pro-image-preview": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000012, - "cache_read_input_token_cost": 2e-7, - "cache_creation_input_token_cost": 3.75e-7, - "cache_read_input_audio_token_cost": 2e-7, - "input_cost_per_audio_token": 0.000002, - "output_cost_per_image_token": 0.00012, - "litellm_provider": "openrouter", - "max_input_tokens": 65536, - "max_output_tokens": 32768, - "max_tokens": 32768, + "lemonade/Gemma-3-4b-it-GGUF": { + "input_cost_per_token": 0, + "litellm_provider": "lemonade", + "max_tokens": 8192, + "max_input_tokens": 128000, + "max_output_tokens": 8192, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_tool_choice": false, - "supports_reasoning": true, + "output_cost_per_token": 0, + "supports_function_calling": true, "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_tool_choice": true }, - "openrouter/google/gemini-3-pro-preview": { - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_200k_tokens": 4e-7, - "cache_creation_input_token_cost_above_200k_tokens": 2.5e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_200k_tokens": 0.000004, - "input_cost_per_token_batches": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65535, - "max_tokens": 65535, + "lemonade/Qwen3-4B-Instruct-2507-GGUF": { + "input_cost_per_token": 0, + "litellm_provider": "lemonade", + "max_tokens": 32768, + "max_input_tokens": 262144, + "max_output_tokens": 32768, "mode": "chat", - "output_cost_per_token": 0.000012, - "output_cost_per_token_above_200k_tokens": 0.000018, - "output_cost_per_token_batches": 0.000006, - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/completions", - "/v1/batch" - ], - "supported_modalities": [ - "text", - "image", - "audio", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": true, + "output_cost_per_token": 0, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true, - "supports_web_search": true + "supports_tool_choice": true + }, + "lemonade/Qwen3-Coder-30B-A3B-Instruct-GGUF": { + "input_cost_per_token": 0, + "litellm_provider": "lemonade", + "max_tokens": 32768, + "max_input_tokens": 262144, + "max_output_tokens": 32768, + "mode": "chat", + "output_cost_per_token": 0, + "supports_function_calling": true, + "supports_response_schema": true, + "supports_tool_choice": true }, - "openrouter/google/gemini-3.1-flash-image": { - "input_cost_per_token": 5e-7, - "output_cost_per_token": 0.000003, - "output_cost_per_image_token": 0.00006, - "litellm_provider": "openrouter", + "lemonade/gpt-oss-120b-mxfp-GGUF": { + "input_cost_per_token": 0, + "litellm_provider": "lemonade", + "max_tokens": 32768, "max_input_tokens": 131072, "max_output_tokens": 32768, - "max_tokens": 32768, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": false, - "supports_reasoning": true, + "output_cost_per_token": 0, + "supports_function_calling": true, "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": true + "supports_tool_choice": true }, - "openrouter/google/gemini-3.1-flash-image-preview": { - "input_cost_per_token": 5e-7, - "output_cost_per_token": 0.000003, - "output_cost_per_image_token": 0.00006, - "litellm_provider": "openrouter", - "max_input_tokens": 65536, - "max_output_tokens": 58982, - "max_tokens": 58982, + "lemonade/gpt-oss-20b-mxfp4-GGUF": { + "input_cost_per_token": 0, + "litellm_provider": "lemonade", + "max_tokens": 32768, + "max_input_tokens": 131072, + "max_output_tokens": 32768, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": false, - "supports_reasoning": true, + "output_cost_per_token": 0, + "supports_function_calling": true, "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": true + "supports_tool_choice": true }, - "openrouter/google/gemini-3.1-flash-lite": { - "cache_creation_input_token_cost": 8.33333333333333e-8, - "cache_read_input_audio_token_cost": 5e-8, - "cache_read_input_token_cost": 2.5e-8, - "input_cost_per_audio_token": 5e-7, + "libertai/deepseek-v4-flash": { + "max_tokens": 200000, + "max_input_tokens": 200000, + "max_output_tokens": 200000, "input_cost_per_token": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "output_cost_per_token": 0.00000175, + "litellm_provider": "libertai", "mode": "chat", - "output_cost_per_reasoning_token": 0.0000015, - "output_cost_per_token": 0.0000015, - "rpm": 2000, - "source": "https://openrouter.ai/api/v1/models", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/completions", - "/v1/batch" - ], - "supported_modalities": [ - "text", - "image", - "audio", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": true, - "supports_audio_output": false, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, "supports_tool_choice": true, - "supports_url_context": true, - "supports_video_input": true, - "supports_vision": true, - "supports_web_search": true, - "tpm": 800000 + "supports_system_messages": true, + "supports_vision": false, + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.1-flash-lite-image": { + "libertai/deepseek-v4-flash-thinking": { + "max_tokens": 200000, + "max_input_tokens": 200000, + "max_output_tokens": 200000, "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 0.0000015, - "output_cost_per_image_token": 0.00003, - "litellm_provider": "openrouter", - "max_input_tokens": 65536, - "max_output_tokens": 58982, - "max_tokens": 58982, + "output_cost_per_token": 0.00000175, + "litellm_provider": "libertai", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": false, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": false, "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.1-flash-lite-preview": { - "cache_creation_input_token_cost": 8.33333333333333e-8, - "cache_read_input_audio_token_cost": 5e-8, - "cache_read_input_token_cost": 2.5e-8, - "input_cost_per_audio_token": 5e-7, - "input_cost_per_token": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "libertai/gemma-4-31b-it": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "libertai", "mode": "chat", - "output_cost_per_reasoning_token": 0.0000015, - "output_cost_per_token": 0.0000015, - "rpm": 2000, - "source": "https://openrouter.ai/api/v1/models", - "supported_endpoints": [ - "/v1/chat/completions", - "/v1/completions", - "/v1/batch" - ], - "supported_modalities": [ - "text", - "image", - "audio", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": true, - "supports_audio_output": false, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, "supports_tool_choice": true, - "supports_url_context": true, - "supports_video_input": true, + "supports_system_messages": true, "supports_vision": true, - "supports_web_search": true, - "tpm": 800000 + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.1-flash-lite:batch": { - "cache_read_input_audio_token_cost": 2.5e-8, - "cache_read_input_token_cost": 1.25e-8, - "input_cost_per_audio_token": 2.5e-7, - "input_cost_per_token": 1.25e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "libertai/gemma-4-31b-it-thinking": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "libertai", "mode": "chat", - "output_cost_per_token": 7.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, + "supports_system_messages": true, "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_reasoning": true, + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.1-pro-preview": { - "cache_creation_input_token_cost": 3.75e-7, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_200k_tokens": 4e-7, - "cache_creation_input_token_cost_above_200k_tokens": 2.5e-7, - "cache_read_input_audio_token_cost": 2e-7, - "input_cost_per_audio_token": 0.000002, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_200k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "libertai/hermes-3-8b-tee": { + "max_tokens": 16000, + "max_input_tokens": 16000, + "max_output_tokens": 16000, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 6e-7, + "litellm_provider": "libertai", "mode": "chat", - "output_cost_per_token": 0.000012, - "output_cost_per_token_above_200k_tokens": 0.000018, - "source": "https://openrouter.ai/api/v1/models", - "supported_modalities": [ - "text", - "image", - "audio", - "video" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_system_messages": true, + "supports_vision": false, + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.1-pro-preview-customtools": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000012, - "cache_read_input_token_cost": 2e-7, - "cache_creation_input_token_cost": 3.75e-7, - "cache_read_input_audio_token_cost": 2e-7, - "input_cost_per_audio_token": 0.000002, - "input_cost_per_token_above_200k_tokens": 0.000004, - "output_cost_per_token_above_200k_tokens": 0.000018, - "cache_read_input_token_cost_above_200k_tokens": 4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "libertai/qwen3.5-122b-a10b": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 0.00000175, + "litellm_provider": "libertai", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_system_messages": true, "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "supports_prompt_caching": true, - "supports_web_search": true, - "supports_video_input": true + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.1-pro-preview:batch": { - "input_cost_per_audio_token": 0.000001, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_200k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "libertai/qwen3.5-122b-a10b-thinking": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 0.00000175, + "litellm_provider": "libertai", "mode": "chat", - "output_cost_per_token": 0.000006, - "output_cost_per_token_above_200k_tokens": 0.000009, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, + "supports_system_messages": true, "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_reasoning": true, + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.5-flash": { - "cache_creation_input_token_cost": 8.33333333333333e-8, - "cache_read_input_audio_token_cost": 3e-7, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000009, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "libertai/qwen3.6-27b": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 5e-7, + "litellm_provider": "libertai", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_system_messages": true, "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_audio_token": 0.000003, - "supports_prompt_caching": true, - "supports_web_search": true, - "supports_video_input": true + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.5-flash-lite": { - "cache_creation_input_token_cost": 8.33333333333333e-8, - "cache_read_input_audio_token_cost": 3e-8, - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "libertai/qwen3.6-27b-thinking": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 5e-7, + "litellm_provider": "libertai", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_system_messages": true, "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "cache_read_input_token_cost": 3e-8, - "input_cost_per_audio_token": 3e-7, - "supports_prompt_caching": true, - "supports_web_search": true, - "supports_video_input": true + "supports_reasoning": true, + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.5-flash-lite:batch": { - "cache_read_input_audio_token_cost": 1.5e-8, - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_audio_token": 1.5e-7, + "libertai/qwen3.6-35b-a3b": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "output_cost_per_token": 5e-7, + "litellm_provider": "libertai", "mode": "chat", - "output_cost_per_token": 0.00000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, + "supports_system_messages": true, "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.5-flash:batch": { - "cache_read_input_audio_token_cost": 1.5e-7, - "cache_read_input_token_cost": 7.5e-8, - "input_cost_per_audio_token": 0.0000015, - "input_cost_per_token": 7.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "libertai/qwen3.6-35b-a3b-thinking": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 5e-7, + "litellm_provider": "libertai", "mode": "chat", - "output_cost_per_token": 0.0000045, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, + "supports_system_messages": true, "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_reasoning": true, + "source": "https://docs.libertai.io/apis/text/" }, - "openrouter/google/gemini-3.6-flash": { - "cache_creation_input_token_cost": 4.16666666666667e-8, - "cache_read_input_audio_token_cost": 7.5e-8, - "input_cost_per_token": 7.5e-7, - "output_cost_per_token": 0.00000375, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "llamagate/codellama-7b": { + "max_tokens": 4096, + "max_input_tokens": 16384, + "max_output_tokens": 4096, + "input_cost_per_token": 6e-8, + "output_cost_per_token": 1.2e-7, + "litellm_provider": "llamagate", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "cache_read_input_token_cost": 7.5e-8, - "input_cost_per_audio_token": 7.5e-7, - "supports_prompt_caching": true, - "supports_web_search": true, - "supports_video_input": true + "supports_response_schema": true }, - "openrouter/google/gemini-3.6-flash:batch": { - "cache_creation_input_token_cost": 4.16666666666667e-8, - "cache_read_input_audio_token_cost": 3.75e-8, - "cache_read_input_token_cost": 3.75e-8, - "input_cost_per_audio_token": 3.75e-7, - "input_cost_per_token": 3.75e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "llamagate/deepseek-coder-6.7b": { + "max_tokens": 4096, + "max_input_tokens": 16384, + "max_output_tokens": 4096, + "input_cost_per_token": 6e-8, + "output_cost_per_token": 1.2e-7, + "litellm_provider": "llamagate", "mode": "chat", - "output_cost_per_token": 0.000001875, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_response_schema": true }, - "openrouter/google/gemini-3.7-flash": { - "cache_creation_input_token_cost": 4.16666666666667e-8, - "cache_read_input_audio_token_cost": 7.5e-8, - "input_cost_per_token": 7.5e-7, - "output_cost_per_token": 0.00000375, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "llamagate/deepseek-r1-7b-qwen": { + "max_tokens": 16384, + "max_input_tokens": 131072, + "max_output_tokens": 16384, + "input_cost_per_token": 8e-8, + "output_cost_per_token": 1.5e-7, + "litellm_provider": "llamagate", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "cache_read_input_token_cost": 7.5e-8, - "input_cost_per_audio_token": 7.5e-7, - "supports_prompt_caching": true, - "supports_web_search": true, - "supports_video_input": true + "supports_reasoning": true }, - "openrouter/google/gemini-3.7-flash:batch": { - "cache_creation_input_token_cost": 4.16666666666667e-8, - "cache_read_input_audio_token_cost": 3.75e-8, - "cache_read_input_token_cost": 3.75e-8, - "input_cost_per_audio_token": 3.75e-7, - "input_cost_per_token": 3.75e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "llamagate/deepseek-r1-8b": { + "max_tokens": 16384, + "max_input_tokens": 65536, + "max_output_tokens": 16384, + "input_cost_per_token": 1e-7, + "output_cost_per_token": 2e-7, + "litellm_provider": "llamagate", "mode": "chat", - "output_cost_per_token": 0.000001875, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_reasoning": true }, - "openrouter/google/gemini-3.8-flash": { - "cache_creation_input_token_cost": 4.16666666666667e-8, - "cache_read_input_audio_token_cost": 7.5e-8, - "input_cost_per_token": 7.5e-7, - "output_cost_per_token": 0.00000375, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "llamagate/dolphin3-8b": { + "max_tokens": 8192, + "max_input_tokens": 128000, + "max_output_tokens": 8192, + "input_cost_per_token": 8e-8, + "output_cost_per_token": 1.5e-7, + "litellm_provider": "llamagate", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "cache_read_input_token_cost": 7.5e-8, - "input_cost_per_audio_token": 7.5e-7, - "supports_prompt_caching": true, - "supports_web_search": true, - "supports_video_input": true + "supports_response_schema": true }, - "openrouter/google/gemini-3.8-flash:batch": { - "cache_creation_input_token_cost": 4.16666666666667e-8, - "cache_read_input_audio_token_cost": 3.75e-8, - "cache_read_input_token_cost": 3.75e-8, - "input_cost_per_audio_token": 3.75e-7, - "input_cost_per_token": 3.75e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "llamagate/gemma3-4b": { + "max_tokens": 8192, + "max_input_tokens": 128000, + "max_output_tokens": 8192, + "input_cost_per_token": 3e-8, + "output_cost_per_token": 8e-8, + "litellm_provider": "llamagate", "mode": "chat", - "output_cost_per_token": 0.000001875, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true, - "supports_video_input": true + "supports_vision": true }, - "openrouter/google/gemini-nano-banana-2.1": { - "input_cost_per_token": 0.0000015, - "litellm_provider": "openrouter", - "max_input_tokens": 65536, - "max_output_tokens": 58982, - "max_tokens": 58982, + "llamagate/llama-3.1-8b": { + "max_tokens": 8192, + "max_input_tokens": 131072, + "max_output_tokens": 8192, + "input_cost_per_token": 3e-8, + "output_cost_per_token": 5e-8, + "litellm_provider": "llamagate", + "mode": "chat", + "supports_function_calling": true, + "supports_response_schema": true + }, + "llamagate/llama-3.2-3b": { + "max_tokens": 8192, + "max_input_tokens": 131072, + "max_output_tokens": 8192, + "input_cost_per_token": 4e-8, + "output_cost_per_token": 8e-8, + "litellm_provider": "llamagate", "mode": "chat", - "output_cost_per_image_token": 0.00003, - "output_cost_per_token": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_response_schema": true }, - "openrouter/google/gemma-2-27b-it": { - "input_cost_per_token": 6.5e-7, - "output_cost_per_token": 6.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 8192, - "max_output_tokens": 2048, + "llamagate/llava-7b": { "max_tokens": 2048, + "max_input_tokens": 4096, + "max_output_tokens": 2048, + "input_cost_per_token": 1e-7, + "output_cost_per_token": 2e-7, + "litellm_provider": "llamagate", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": false, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/google/gemma-3-12b-it": { - "input_cost_per_token": 5e-8, + "llamagate/mistral-7b-v0.3": { + "max_tokens": 8192, + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "input_cost_per_token": 1e-7, "output_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "litellm_provider": "llamagate", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false + "supports_response_schema": true }, - "openrouter/google/gemma-3-27b-it": { + "llamagate/openthinker-7b": { + "max_tokens": 8192, + "max_input_tokens": 32768, + "max_output_tokens": 8192, "input_cost_per_token": 8e-8, - "output_cost_per_token": 4.5e-7, - "cache_read_input_token_cost": 4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "output_cost_per_token": 1.5e-7, + "litellm_provider": "llamagate", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false + "supports_reasoning": true }, - "openrouter/google/gemma-3-4b-it": { - "input_cost_per_token": 5e-8, - "output_cost_per_token": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "llamagate/qwen2.5-coder-7b": { + "max_tokens": 8192, + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "input_cost_per_token": 6e-8, + "output_cost_per_token": 1.2e-7, + "litellm_provider": "llamagate", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": false, - "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false + "supports_function_calling": true, + "supports_response_schema": true }, - "openrouter/google/gemma-4-26b-a4b-it": { - "cache_read_input_token_cost": 4.25e-8, - "input_cost_per_token": 7.65e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "llamagate/qwen3-8b": { + "max_tokens": 8192, + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "input_cost_per_token": 4e-8, + "output_cost_per_token": 1.4e-7, + "litellm_provider": "llamagate", "mode": "chat", - "output_cost_per_token": 2.55e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_response_schema": true }, - "openrouter/google/gemma-4-26b-a4b-it:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "llamagate/qwen3-vl-8b": { + "max_tokens": 8192, + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 5.5e-7, + "litellm_provider": "llamagate", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/google/gemma-4-31b-it": { - "input_cost_per_token": 9e-8, - "output_cost_per_token": 3.4e-7, - "cache_read_input_token_cost": 5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 16384, - "max_tokens": 16384, + "medlm-large": { + "input_cost_per_character": 0.000005, + "litellm_provider": "vertex_ai-language-models", + "max_input_tokens": 8192, + "max_output_tokens": 1024, + "max_tokens": 1024, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_character": 0.000015, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_tool_choice": true + }, + "medlm-medium": { + "input_cost_per_character": 5e-7, + "litellm_provider": "vertex_ai-language-models", + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_character": 0.000001, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_tool_choice": true + }, + "meta.llama2-13b-chat-v1": { + "input_cost_per_token": 7.5e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "max_tokens": 4096, + "mode": "chat", + "output_cost_per_token": 0.000001, + "source": "https://aws.amazon.com/bedrock/pricing/" + }, + "meta.llama2-70b-chat-v1": { + "input_cost_per_token": 0.00000195, + "litellm_provider": "bedrock", + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "max_tokens": 4096, + "mode": "chat", + "output_cost_per_token": 0.00000256, + "source": "https://aws.amazon.com/bedrock/pricing/" + }, + "meta.llama3-1-405b-instruct-v1:0": { + "deprecation_date": "2026-07-07", + "input_cost_per_token": 0.0000024, + "litellm_provider": "bedrock", + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "max_tokens": 4096, + "mode": "chat", + "output_cost_per_token": 0.0000024, + "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_tool_choice": false }, - "openrouter/google/gemma-4-31b-it:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "meta.llama3-1-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 7.2e-7, + "input_cost_per_token_batches": 3.6e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 7.2e-7, + "output_cost_per_token_batches": 3.6e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false + "supports_tool_choice": false }, - "openrouter/gryphe/mythomax-l2-13b": { - "input_cost_per_token": 8e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 8192, - "max_output_tokens": 3686, - "max_tokens": 3686, + "meta.llama3-1-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 2.2e-7, + "input_cost_per_token_batches": 1.1e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 1.1e-7, - "supports_tool_choice": false, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 2.2e-7, + "output_cost_per_token_batches": 1.1e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_function_calling": true, + "supports_tool_choice": false }, - "openrouter/ibm-granite/granite-4.0-h-micro": { - "input_cost_per_token": 1.7e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131000, - "max_output_tokens": 117900, - "max_tokens": 117900, + "meta.llama3-2-11b-instruct-v1:0": { + "deprecation_date": "2026-07-07", + "input_cost_per_token": 1.6e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 1.12e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, + "output_cost_per_token": 1.6e-7, + "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", + "supports_function_calling": true, "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/ibm-granite/granite-4.2-8b": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 6e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "meta.llama3-2-1b-instruct-v1:0": { + "deprecation_date": "2026-07-07", + "input_cost_per_token": 1e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 2.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 1e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-2-1b-instruct.md", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": false }, - "openrouter/inception/mercury-2": { - "cache_read_input_token_cost": 2.5e-8, - "input_cost_per_token": 2.5e-7, - "litellm_provider": "openrouter", + "meta.llama3-2-3b-instruct-v1:0": { + "deprecation_date": "2026-07-07", + "input_cost_per_token": 1.5e-7, + "litellm_provider": "bedrock", "max_input_tokens": 128000, - "max_output_tokens": 50000, - "max_tokens": 50000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 7.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 1.5e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-2-3b-instruct.md", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": false }, - "openrouter/inception/mercury-2.5": { - "cache_read_input_token_cost": 4e-9, - "input_cost_per_token": 4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 260000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "meta.llama3-2-90b-instruct-v1:0": { + "deprecation_date": "2026-07-07", + "input_cost_per_token": 7.2e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 1.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 7.2e-7, + "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": false, + "supports_vision": true }, - "openrouter/inclusionai/ling-3.0-flash": { - "cache_read_input_token_cost": 4.2e-9, - "input_cost_per_token": 2.1e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "meta.llama3-3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 7.2e-7, + "input_cost_per_token_batches": 3.6e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 6.3e-8, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 7.2e-7, + "output_cost_per_token_batches": 3.6e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": false }, - "openrouter/inclusionai/ling-3.0-flash-fin": { - "cache_read_input_token_cost": 8.4e-9, - "input_cost_per_token": 4.2e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "meta.llama3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 0.00000265, + "litellm_provider": "bedrock", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 1.232e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000035, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-70b-instruct.md" + }, + "meta.llama3-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 3e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 6e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-8b-instruct.md" + }, + "meta.llama4-maverick-17b-instruct-v1:0": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 2.4e-7, + "input_cost_per_token_batches": 1.2e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 9.7e-7, + "output_cost_per_token_batches": 4.85e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-4-maverick-17b-instruct.md", + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ], + "supports_function_calling": true, + "supports_tool_choice": false + }, + "meta.llama4-scout-17b-instruct-v1:0": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 1.7e-7, + "input_cost_per_token_batches": 8.5e-8, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 10000000, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 6.6e-7, + "output_cost_per_token_batches": 3.3e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-4-scout-17b-instruct.md", + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ], + "supports_function_calling": true, + "supports_tool_choice": false + }, + "meta/muse-spark-1.1": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.00000125, + "litellm_provider": "meta", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 0.00000425, + "search_context_cost_per_query": { + "search_context_size_high": 0.0025, + "search_context_size_low": 0.0025, + "search_context_size_medium": 0.0025 + }, + "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, + "supports_minimal_reasoning_effort": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_web_search": true, + "supports_xhigh_reasoning_effort": true }, - "openrouter/inclusionai/ling-3.0-flash-fin:free": { - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "meta/muse-spark-1.2": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.00000125, + "litellm_provider": "meta", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000425, + "search_context_cost_per_query": { + "search_context_size_high": 0.0025, + "search_context_size_low": 0.0025, + "search_context_size_medium": 0.0025 + }, + "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_minimal_reasoning_effort": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": false, + "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_web_search": true, + "supports_xhigh_reasoning_effort": true }, - "openrouter/inclusionai/ling-3.0-flash-sante:free": { - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "meta/muse-spark-1.2-contributor": { + "cache_read_input_token_cost": 2e-9, + "input_cost_per_token": 1e-7, + "litellm_provider": "meta", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 2e-7, + "search_context_cost_per_query": { + "search_context_size_high": 0.0025, + "search_context_size_low": 0.0025, + "search_context_size_medium": 0.0025 + }, + "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_minimal_reasoning_effort": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": false, + "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_web_search": true, + "supports_xhigh_reasoning_effort": true }, - "openrouter/inclusionai/ling-3.0-flash-vl": { - "cache_read_input_token_cost": 4.2e-9, - "input_cost_per_token": 2.1e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "meta/muse-spark-1.3": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.00000125, + "litellm_provider": "meta", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 6.16e-8, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000425, + "search_context_cost_per_query": { + "search_context_size_high": 0.0025, + "search_context_size_low": 0.0025, + "search_context_size_medium": 0.0025 + }, + "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, + "supports_minimal_reasoning_effort": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": false + "supports_web_search": true, + "supports_xhigh_reasoning_effort": true }, - "openrouter/inclusionai/ling-3.0-flash-vl:free": { - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "meta/muse-spark-1.3-contributor": { + "cache_read_input_token_cost": 2e-9, + "input_cost_per_token": 1e-7, + "litellm_provider": "meta", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 2e-7, + "search_context_cost_per_query": { + "search_context_size_high": 0.0025, + "search_context_size_low": 0.0025, + "search_context_size_medium": 0.0025 + }, + "source": "https://ai.developer.meta.com/docs/pricing-rate-limits", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_minimal_reasoning_effort": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": false, + "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": false + "supports_web_search": true, + "supports_xhigh_reasoning_effort": true }, - "openrouter/inclusionai/ling-3.1-flash": { - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "meta/muse-wonder-1.0": { + "cache_read_input_token_cost": 2e-7, + "input_cost_per_token": 0.000004, + "litellm_provider": "meta", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00002, + "search_context_cost_per_query": { + "search_context_size_high": 0.0025, + "search_context_size_low": 0.0025, + "search_context_size_medium": 0.0025 + }, + "source": "https://dev.meta.ai/docs/pricing-rate-limits", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses", + "/v1/messages" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_max_reasoning_effort": true, + "supports_minimal_reasoning_effort": true, + "supports_none_reasoning_effort": false, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": false, + "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_web_search": true, + "supports_xhigh_reasoning_effort": true }, - "openrouter/inference-net/schematron-v2-small": { - "cache_read_input_token_cost": 5e-8, - "input_cost_per_token": 5e-8, - "litellm_provider": "openrouter", + "meta_llama/Llama-3.3-70B-Instruct": { + "litellm_provider": "meta_llama", "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, + "max_output_tokens": 4028, + "max_tokens": 4028, "mode": "chat", - "output_cost_per_token": 2.3e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "source": "https://ai.developer.meta.com/docs/models", + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_tool_choice": true }, - "openrouter/inference-net/schematron-v2-turbo": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 3e-8, - "litellm_provider": "openrouter", + "meta_llama/Llama-3.3-8B-Instruct": { + "litellm_provider": "meta_llama", "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, + "max_output_tokens": 4028, + "max_tokens": 4028, "mode": "chat", - "output_cost_per_token": 1.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "source": "https://ai.developer.meta.com/docs/models", + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_tool_choice": true }, - "openrouter/kwaipilot/kat-coder-pro-v2.5": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 7.4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8": { + "litellm_provider": "meta_llama", + "max_input_tokens": 1000000, + "max_output_tokens": 4028, + "max_tokens": 4028, "mode": "chat", - "output_cost_per_token": 0.00000296, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://ai.developer.meta.com/docs/models", + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/liquid/lfm-2.5-2.6b:free": { - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 65536, - "max_output_tokens": 8192, - "max_tokens": 8192, + "meta_llama/Llama-4-Scout-17B-16E-Instruct-FP8": { + "litellm_provider": "meta_llama", + "max_input_tokens": 10000000, + "max_output_tokens": 4028, + "max_tokens": 4028, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://ai.developer.meta.com/docs/models", + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/mancer/weaver": { - "input_cost_per_token": 4e-7, - "litellm_provider": "openrouter", - "max_tokens": 6000, + "microsoft_365_copilot/chat": { + "litellm_provider": "microsoft_365_copilot", "mode": "chat", - "output_cost_per_token": 7.5e-7, - "supports_tool_choice": false, - "max_input_tokens": 8000, - "max_output_tokens": 6000, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_mid_conversation_system": true, + "input_cost_per_token": 0, + "output_cost_per_token": 0 }, - "openrouter/meituan/longcat-2.0": { - "cache_read_input_token_cost": 6e-9, + "minimax.minimax-m2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048756, - "max_output_tokens": 262144, - "max_tokens": 262144, + "input_cost_per_token_batches": 1.5e-7, + "input_cost_per_token_flex": 1.5e-7, + "input_cost_per_token_priority": 5.2e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.0000012, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token_batches": 6e-7, + "output_cost_per_token_flex": 6e-7, + "output_cost_per_token_priority": 0.0000021, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_system_messages": true, + "supports_native_structured_output": true, + "supports_response_schema": true, + "supports_vision": false }, - "openrouter/meta-llama/llama-3-70b-instruct": { - "input_cost_per_token": 5.9e-7, - "litellm_provider": "openrouter", + "minimax.minimax-m2.1": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 3e-7, + "input_cost_per_token_batches": 1.5e-7, + "input_cost_per_token_flex": 1.5e-7, + "input_cost_per_token_priority": 5.25e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 196000, + "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", - "output_cost_per_token": 7.9e-7, - "supports_tool_choice": true, - "max_input_tokens": 8192, - "max_output_tokens": 8000 - }, - "openrouter/meta-llama/llama-3.1-70b-instruct": { - "input_cost_per_token": 4e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000012, + "output_cost_per_token_batches": 6e-7, + "output_cost_per_token_flex": 6e-7, + "output_cost_per_token_priority": 0.0000021, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_system_messages": true, "supports_tool_choice": true, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_audio_input": false, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/meta-llama/llama-3.1-8b-instruct": { - "input_cost_per_token": 5e-8, - "output_cost_per_token": 8e-8, - "cache_read_input_token_cost": 2.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "minimax.minimax-m2.5": { + "input_cost_per_token": 3e-7, + "input_cost_per_token_batches": 1.5e-7, + "input_cost_per_token_flex": 1.5e-7, + "input_cost_per_token_priority": 5.25e-7, + "output_cost_per_token": 0.0000012, + "output_cost_per_token_flex": 6e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 196000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token_batches": 6e-7, + "output_cost_per_token_priority": 0.0000021, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false - }, - "openrouter/meta-llama/llama-3.2-1b-instruct": { - "input_cost_per_token": 2.7e-8, - "output_cost_per_token": 2.01e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 60000, - "max_output_tokens": 54000, - "max_tokens": 54000, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/meta-llama/llama-3.2-3b-instruct": { - "input_cost_per_token": 5e-8, - "output_cost_per_token": 3.3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": false, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/meta-llama/llama-3.3-70b-instruct": { - "cache_read_input_token_cost": 1.1e-7, - "input_cost_per_token": 2.2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "minimax/MiniMax-M2": { + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000012, + "cache_read_input_token_cost": 3e-8, + "cache_creation_input_token_cost": 3.75e-7, + "litellm_provider": "minimax", "mode": "chat", - "output_cost_per_token": 5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "max_input_tokens": 200000, + "max_output_tokens": 8192 }, - "openrouter/meta-llama/llama-4-maverick": { - "cache_read_input_token_cost": 5e-8, - "input_cost_per_token": 1.875e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 16384, - "max_tokens": 16384, + "minimax/MiniMax-M2.1": { + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000012, + "cache_read_input_token_cost": 3e-8, + "cache_creation_input_token_cost": 3.75e-7, + "litellm_provider": "minimax", "mode": "chat", - "output_cost_per_token": 6.525e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "max_input_tokens": 1000000, + "max_output_tokens": 8192 }, - "openrouter/meta-llama/llama-4-scout": { - "input_cost_per_token": 1e-7, - "output_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1310720, - "max_output_tokens": 16384, - "max_tokens": 16384, + "minimax/MiniMax-M2.1-lightning": { + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000024, + "cache_read_input_token_cost": 3e-8, + "cache_creation_input_token_cost": 3.75e-7, + "litellm_provider": "minimax", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/meta-llama/llama-guard-4-12b": { - "input_cost_per_token": 1.8e-7, - "output_cost_per_token": 1.8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 163840, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": false, - "supports_response_schema": false, - "supports_vision": true, - "supports_web_search": false + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "max_input_tokens": 1000000, + "max_output_tokens": 8192 }, - "openrouter/meta/muse-glimmer-30b": { - "cache_read_input_token_cost": 4e-8, + "minimax/MiniMax-M2.5": { "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", "output_cost_per_token": 0.0000012, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "cache_read_input_token_cost": 3e-8, + "cache_creation_input_token_cost": 3.75e-7, + "litellm_provider": "minimax", + "mode": "chat", "supports_function_calling": true, - "supports_pdf_input": false, + "supports_tool_choice": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_system_messages": true, + "max_input_tokens": 1000000, + "max_output_tokens": 8192 }, - "openrouter/meta/muse-spark-1.1": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "minimax/MiniMax-M2.5-lightning": { + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000024, + "cache_read_input_token_cost": 3e-8, + "cache_creation_input_token_cost": 3.75e-7, + "litellm_provider": "minimax", "mode": "chat", - "output_cost_per_token": 0.00000425, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, + "supports_tool_choice": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "max_input_tokens": 1000000, + "max_output_tokens": 8192 }, - "openrouter/meta/muse-spark-1.2": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "minimax/MiniMax-M3": { + "input_cost_per_token": 3e-7, + "input_cost_per_token_above_512k_tokens": 6e-7, + "output_cost_per_token": 0.0000012, + "output_cost_per_token_above_512k_tokens": 0.0000024, + "cache_read_input_token_cost": 6e-8, + "cache_read_input_token_cost_above_512k_tokens": 1.2e-7, + "litellm_provider": "minimax", "mode": "chat", - "output_cost_per_token": 0.00000425, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, + "supports_tool_choice": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, + "supports_system_messages": true, "supports_vision": true, - "supports_web_search": true + "max_input_tokens": 1000000, + "max_output_tokens": 128000 }, - "openrouter/meta/muse-spark-1.2-contributor": { - "cache_read_input_token_cost": 2e-9, - "input_cost_per_token": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "mistral.devstral-2-123b": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 4e-7, + "input_cost_per_token_batches": 2e-7, + "input_cost_per_token_flex": 2e-7, + "input_cost_per_token_priority": 7e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 256000, + "max_output_tokens": 32000, + "max_tokens": 32000, "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 0.000002, + "output_cost_per_token_batches": 0.000001, + "output_cost_per_token_flex": 0.000001, + "output_cost_per_token_priority": 0.0000035, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_audio_input": false, + "supports_response_schema": true, + "supports_vision": false }, - "openrouter/meta/muse-spark-1.3": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "mistral.magistral-small-2509": { + "input_cost_per_token": 5e-7, + "input_cost_per_token_batches": 2.5e-7, + "input_cost_per_token_flex": 2.5e-7, + "input_cost_per_token_priority": 8.8e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 40000, + "max_tokens": 40000, "mode": "chat", - "output_cost_per_token": 0.00000425, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 0.0000015, + "output_cost_per_token_batches": 7.5e-7, + "output_cost_per_token_flex": 7.5e-7, + "output_cost_per_token_priority": 0.00000263, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_vision": true }, - "openrouter/meta/muse-spark-1.3-contributor": { - "cache_read_input_token_cost": 2e-9, - "input_cost_per_token": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "mistral.ministral-3-14b-instruct": { + "input_cost_per_token": 2e-7, + "input_cost_per_token_batches": 1e-7, + "input_cost_per_token_flex": 1e-7, + "input_cost_per_token_priority": 3.5e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token_batches": 1e-7, + "output_cost_per_token_flex": 1e-7, + "output_cost_per_token_priority": 3.5e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, + "supports_system_messages": true, + "supports_native_structured_output": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/microsoft/phi-4": { - "input_cost_per_token": 7e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 16384, - "max_output_tokens": 14745, - "max_tokens": 14745, + "mistral.ministral-3-3b-instruct": { + "input_cost_per_token": 1e-7, + "input_cost_per_token_batches": 5e-8, + "input_cost_per_token_flex": 5e-8, + "input_cost_per_token_priority": 1.7e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", - "output_cost_per_token": 1.4e-7, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 1e-7, + "output_cost_per_token_batches": 5e-8, + "output_cost_per_token_flex": 5e-8, + "output_cost_per_token_priority": 1.7e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_function_calling": true, + "supports_system_messages": true, + "supports_native_structured_output": true, "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/microsoft/wizardlm-2-8x22b": { - "input_cost_per_token": 6.2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 65535, + "mistral.ministral-3-8b-instruct": { + "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7e-8, + "input_cost_per_token_flex": 7e-8, + "input_cost_per_token_priority": 2.6e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", - "output_cost_per_token": 6.2e-7, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 1.5e-7, + "output_cost_per_token_batches": 7e-8, + "output_cost_per_token_flex": 7e-8, + "output_cost_per_token_priority": 2.6e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_function_calling": true, + "supports_system_messages": true, + "supports_native_structured_output": true, "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/minimax/minimax-01": { - "input_cost_per_token": 2e-7, - "output_cost_per_token": 0.0000011, - "litellm_provider": "openrouter", - "max_input_tokens": 1000192, - "max_output_tokens": 900172, - "max_tokens": 900172, + "mistral.mistral-7b-instruct-v0:2": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 1.5e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 32000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mistral-7b-instruct.md", + "supports_tool_choice": true }, - "openrouter/minimax/minimax-m1": { - "input_cost_per_token": 5.5e-7, - "output_cost_per_token": 0.0000022, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 40000, - "max_tokens": 40000, + "mistral.mistral-large-2402-v1:0": { + "input_cost_per_token": 0.000004, + "litellm_provider": "bedrock", + "max_input_tokens": 32000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000012, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mistral-large.md", + "supports_function_calling": true + }, + "mistral.mistral-large-2407-v1:0": { + "input_cost_per_token": 0.000002, + "litellm_provider": "bedrock", + "max_input_tokens": 128000, + "max_output_tokens": 8191, + "max_tokens": 8191, + "mode": "chat", + "output_cost_per_token": 0.000006, + "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/minimax/minimax-m2": { - "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 204800, - "max_output_tokens": 176947, - "max_tokens": 176947, + "mistral.mistral-large-3-675b-instruct": { + "input_cost_per_token": 5e-7, + "input_cost_per_token_batches": 2.5e-7, + "input_cost_per_token_flex": 2.5e-7, + "input_cost_per_token_priority": 8.8e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 256000, + "max_output_tokens": 32000, + "max_tokens": 32000, "mode": "chat", - "output_cost_per_token": 0.0000012, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000015, + "output_cost_per_token_batches": 7.5e-7, + "output_cost_per_token_flex": 7.5e-7, + "output_cost_per_token_priority": 0.00000263, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, + "supports_system_messages": true, + "supports_native_structured_output": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/minimax/minimax-m2-her": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "cache_read_input_token_cost": 3e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 65536, - "max_output_tokens": 2048, - "max_tokens": 2048, + "mistral.mistral-small-2402-v1:0": { + "input_cost_per_token": 0.000001, + "input_cost_per_token_batches": 5e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 32000, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_tool_choice": false, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_web_search": false + "output_cost_per_token": 0.000003, + "output_cost_per_token_batches": 0.0000015, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mistral-small.md", + "supports_function_calling": true }, - "openrouter/minimax/minimax-m2.1": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "cache_creation_input_token_cost": 0, + "mistral.mixtral-8x7b-instruct-v0:1": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 4.5e-7, + "litellm_provider": "bedrock", + "max_input_tokens": 32000, + "max_output_tokens": 4096, + "max_tokens": 4096, + "mode": "chat", + "output_cost_per_token": 7e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-mixtral-8x7b-instruct.md", + "supports_tool_choice": true + }, + "mistral.pixtral-large-2502-v1:0": { + "input_cost_per_token": 0.000002, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 0.000006, + "source": "https://aws.amazon.com/bedrock/pricing/", + "supports_function_calling": true, + "supports_tool_choice": false + }, + "mistral.voxtral-mini-3b-2507": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 4e-8, + "input_cost_per_token_batches": 2e-8, + "input_cost_per_token_flex": 2e-8, + "input_cost_per_token_priority": 7e-8, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 4e-8, + "output_cost_per_token_batches": 2e-8, + "output_cost_per_token_flex": 2e-8, + "output_cost_per_token_priority": 7e-8, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_audio_input": true, + "supports_system_messages": true, + "supports_native_structured_output": true + }, + "mistral.voxtral-small-24b-2507": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 1e-7, + "input_cost_per_token_batches": 5e-8, + "input_cost_per_token_flex": 5e-8, + "input_cost_per_token_priority": 1.7e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 3e-7, + "output_cost_per_token_batches": 1.5e-7, + "output_cost_per_token_flex": 1.5e-7, + "output_cost_per_token_priority": 5.2e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_audio_input": true, + "supports_system_messages": true, + "supports_native_structured_output": true + }, + "mistral/codestral-2508": { "cache_read_input_token_cost": 3e-8, - "deprecation_date": "2026-10-08", - "litellm_provider": "openrouter", - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, + "input_cost_per_token": 3e-7, + "litellm_provider": "mistral", + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 9e-7, + "source": "https://docs.mistral.ai/models/model-cards/codestral-25-08", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_computer_use": false, - "supports_pdf_input": false, "supports_response_schema": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/minimax/minimax-m2.5": { - "input_cost_per_token": 2.7e-7, - "output_cost_per_token": 0.00000108, - "cache_read_input_token_cost": 2.7e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 204800, + "mistral/codestral-latest": { + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "mistral", + "max_input_tokens": 128000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 9e-7, + "source": "https://docs.mistral.ai/models/model-cards/codestral-25-08", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_computer_use": false, - "supports_pdf_input": false, "supports_response_schema": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/minimax/minimax-m2.7": { - "cache_read_input_token_cost": 4.2e-8, - "input_cost_per_token": 2.1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 204800, - "max_output_tokens": 176947, - "max_tokens": 176947, + "mistral/codestral-mamba-latest": { + "cache_read_input_token_cost": 2.5e-8, + "input_cost_per_token": 2.5e-7, + "litellm_provider": "mistral", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 8.4e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 2.5e-7, + "source": "https://mistral.ai/technology/", + "supports_assistant_prefill": true, + "supports_tool_choice": true + }, + "mistral/devstral-latest": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 4e-7, + "litellm_provider": "mistral", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, + "mode": "chat", + "output_cost_per_token": 0.000002, + "source": "https://mistral.ai/news/devstral-2-vibe-cli", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/minimax/minimax-m2.7:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 196608, - "max_output_tokens": 176947, - "max_tokens": 176947, + "mistral/devstral-medium-latest": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 4e-7, + "litellm_provider": "mistral", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "source": "https://openrouter.ai/minimax/minimax-m2.7:free", + "output_cost_per_token": 0.000002, + "source": "https://mistral.ai/news/devstral-2-vibe-cli", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false + "supports_tool_choice": true }, - "openrouter/minimax/minimax-m3": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000012, - "litellm_provider": "openrouter", + "mistral/devstral-small-latest": { + "cache_read_input_token_cost": 1e-8, + "input_cost_per_token": 1e-7, + "litellm_provider": "mistral", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, + "mode": "chat", + "output_cost_per_token": 3e-7, + "source": "https://docs.mistral.ai/models/devstral-small-2-25-12", + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_response_schema": true, + "supports_tool_choice": true + }, + "mistral/glm-5-2": { + "cache_read_input_token_cost": 1.4e-7, + "input_cost_per_token": 0.0000014, + "litellm_provider": "mistral", "max_input_tokens": 1048576, - "max_output_tokens": 512000, - "max_tokens": 512000, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000044, + "reasoning_effort_levels": [ + "none", + "minimal", + "low", + "medium", + "high", + "xhigh", + "max" + ], + "source": "https://docs.mistral.ai/models/zai-glm-5-2", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "cache_read_input_token_cost": 6e-8, - "supports_audio_input": false, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/minimax/minimax-m3:free": { + "mistral/labs-leanstral-1-5": { + "deprecation_date": "2026-09-30", "input_cost_per_token": 0, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "source": "https://docs.mistral.ai/models/model-cards/leanstral-1-5", + "supports_function_calling": true, + "supports_response_schema": true, + "supports_tool_choice": true + }, + "mistral/labs-leanstral-1-5-1": { + "input_cost_per_token": 0, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/minimax/minimax-m3:free", + "output_cost_per_token": 0, + "source": "https://docs.mistral.ai/models/model-cards/leanstral-1-5", "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true + "supports_tool_choice": true }, - "openrouter/mistralai/codestral-2508": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 9e-7, - "cache_read_input_token_cost": 3e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 256000, - "max_output_tokens": 204800, - "max_tokens": 204800, + "mistral/magistral-medium-latest": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000075, + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_tool_choice": true, + "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/mistralai/codestral-2508:batch": { + "mistral/magistral-small-latest": { "cache_read_input_token_cost": 1.5e-8, "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 256000, - "max_output_tokens": 204800, - "max_tokens": 204800, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 4.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 6e-7, + "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/devstral-2512": { - "cache_read_input_token_cost": 4e-8, - "input_cost_per_image": 0, - "input_cost_per_token": 4e-7, - "litellm_provider": "openrouter", + "mistral/ministral-14b-2512": { + "input_cost_per_token": 2e-7, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 2e-7, + "source": "https://docs.mistral.ai/models/ministral-3-14b-25-12", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "cache_read_input_token_cost": 2e-8 + }, + "mistral/ministral-14b-latest": { + "input_cost_per_token": 2e-7, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 2e-7, + "source": "https://docs.mistral.ai/models/ministral-3-14b-25-12", + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "cache_read_input_token_cost": 2e-8 }, - "openrouter/mistralai/ministral-14b-2512": { + "mistral/ministral-3-14b-2512": { "cache_read_input_token_cost": 2e-8, - "input_cost_per_image": 0, "input_cost_per_token": 2e-7, - "litellm_provider": "openrouter", + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://mistral.ai/pricing", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/ministral-3b-2512": { + "mistral/ministral-3-3b-2512": { "cache_read_input_token_cost": 1e-8, - "input_cost_per_image": 0, "input_cost_per_token": 1e-7, - "litellm_provider": "openrouter", + "litellm_provider": "mistral", "max_input_tokens": 131072, - "max_output_tokens": 104857, - "max_tokens": 104857, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", "output_cost_per_token": 1e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://mistral.ai/pricing", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/ministral-8b-2512": { + "mistral/ministral-3-8b-2512": { "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_image": 0, "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", "output_cost_per_token": 1.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://mistral.ai/pricing", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/ministral-8b-2512:batch": { - "cache_read_input_token_cost": 7.5e-9, - "input_cost_per_token": 7.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "mistral/ministral-3b-2512": { + "input_cost_per_token": 1e-7, + "litellm_provider": "mistral", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 7.5e-8, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 1e-7, + "source": "https://docs.mistral.ai/models/ministral-3-3b-25-12", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": false - }, - "openrouter/mistralai/mistral-7b-instruct": { - "input_cost_per_token": 1.3e-7, - "litellm_provider": "openrouter", - "max_tokens": 8191, - "mode": "chat", - "output_cost_per_token": 1.3e-7, - "supports_tool_choice": true, - "max_input_tokens": 32768, - "max_output_tokens": 8191 - }, - "openrouter/mistralai/mistral-large": { - "cache_read_input_token_cost": 2e-7, - "input_cost_per_token": 0.000002, - "litellm_provider": "openrouter", - "max_tokens": 102400, - "mode": "chat", - "output_cost_per_token": 0.000006, - "supports_tool_choice": true, - "max_input_tokens": 128000, - "max_output_tokens": 102400, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "cache_read_input_token_cost": 1e-8 }, - "openrouter/mistralai/mistral-large-2407": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, - "cache_read_input_token_cost": 2e-7, - "litellm_provider": "openrouter", + "mistral/ministral-3b-latest": { + "input_cost_per_token": 1e-7, + "litellm_provider": "mistral", "max_input_tokens": 131072, - "max_output_tokens": 104857, - "max_tokens": 104857, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false - }, - "openrouter/mistralai/mistral-large-2512": { - "cache_read_input_token_cost": 5e-8, - "input_cost_per_token": 5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 1e-7, + "source": "https://docs.mistral.ai/models/ministral-3-3b-25-12", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": false + "cache_read_input_token_cost": 1e-8 }, - "openrouter/mistralai/mistral-large-2512:batch": { - "cache_read_input_token_cost": 2.5e-8, - "input_cost_per_token": 2.5e-7, - "litellm_provider": "openrouter", + "mistral/ministral-8b-2512": { + "cache_read_input_token_cost": 1.5e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 7.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 1.5e-7, + "source": "https://mistral.ai/pricing", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/mistral-large-4-0": { - "cache_read_input_token_cost": 7e-8, - "input_cost_per_token": 6.8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 524288, + "mistral/ministral-8b-latest": { + "cache_read_input_token_cost": 1.5e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "mistral", + "max_input_tokens": 262144, "max_output_tokens": 262144, "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.00000209, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 1.5e-7, + "source": "https://mistral.ai/pricing", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/mistral-medium-3": { - "input_cost_per_token": 4e-7, - "output_cost_per_token": 0.000002, + "mistral/mistral-code-agent-latest": { "cache_read_input_token_cost": 4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 104857, - "max_tokens": 104857, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false - }, - "openrouter/mistralai/mistral-medium-3-5": { - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.0000075, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "input_cost_per_token": 4e-7, + "litellm_provider": "mistral", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000002, + "source": "https://mistral.ai/news/devstral-2-vibe-cli", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/mistralai/mistral-medium-3-5:batch": { - "input_cost_per_token": 7.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "mistral/mistral-code-fim-latest": { + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "mistral", + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.00000375, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 9e-7, + "source": "https://docs.mistral.ai/models/model-cards/codestral-25-08", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/mistralai/mistral-medium-3.1": { - "input_cost_per_token": 4e-7, - "output_cost_per_token": 0.000002, - "cache_read_input_token_cost": 4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 104857, - "max_tokens": 104857, + "mistral/mistral-code-latest": { + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "mistral", + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 9e-7, + "source": "https://docs.mistral.ai/models/model-cards/codestral-25-08", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_tool_choice": true, "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/mistralai/mistral-medium-3.1:batch": { - "cache_read_input_token_cost": 2e-8, - "input_cost_per_token": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 104857, - "max_tokens": 104857, + "mistral/mistral-large-2512": { + "cache_read_input_token_cost": 5e-8, + "input_cost_per_token": 5e-7, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.000001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000015, + "source": "https://docs.mistral.ai/models/mistral-large-3-25-12", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/mistral-nemo": { - "input_cost_per_token": 1.9e-8, - "output_cost_per_token": 3e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "mistral/mistral-large-3": { + "cache_read_input_token_cost": 5e-8, + "input_cost_per_token": 5e-7, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000015, + "source": "https://docs.mistral.ai/models/mistral-large-3-25-12", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/mistralai/mistral-saba": { - "input_cost_per_token": 2e-7, - "output_cost_per_token": 6e-7, - "cache_read_input_token_cost": 2e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 26214, - "max_tokens": 26214, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false - }, - "openrouter/mistralai/mistral-small-24b-instruct-2501": { - "input_cost_per_token": 5e-8, - "output_cost_per_token": 8e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": false, - "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/mistral-small-2603": { - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 6e-7, - "cache_read_input_token_cost": 1.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "mistral/mistral-large-4": { + "cache_read_input_token_cost": 6.8e-8, + "input_cost_per_token": 6.8e-7, + "litellm_provider": "mistral", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000209, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/mistral-large-4", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/mistralai/mistral-small-2603:batch": { - "cache_read_input_token_cost": 7.5e-9, - "input_cost_per_token": 7.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 209715, - "max_tokens": 209715, + "mistral/mistral-large-4-0": { + "cache_read_input_token_cost": 6.8e-8, + "input_cost_per_token": 6.8e-7, + "litellm_provider": "mistral", + "max_input_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 3e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000209, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/mistral-large-4", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/mistralai/mistral-small-3.1-24b-instruct": { - "input_cost_per_token": 3.51e-7, - "litellm_provider": "openrouter", - "max_tokens": 102400, - "mode": "chat", - "output_cost_per_token": 5.55e-7, - "supports_tool_choice": true, - "max_input_tokens": 128000, - "max_output_tokens": 102400, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/mistral-small-3.2-24b-instruct": { - "input_cost_per_token": 9.375e-8, - "litellm_provider": "openrouter", - "max_tokens": 16384, + "mistral/mistral-large-latest": { + "cache_read_input_token_cost": 5e-8, + "input_cost_per_token": 5e-7, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 2.5e-7, - "supports_tool_choice": true, - "max_input_tokens": 256000, - "max_output_tokens": 16384, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000015, + "source": "https://docs.mistral.ai/models/mistral-large-3-25-12", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/mistralai/mixtral-8x22b-instruct": { - "cache_read_input_token_cost": 2e-7, - "input_cost_per_token": 0.000002, - "litellm_provider": "openrouter", - "max_tokens": 52428, - "mode": "chat", - "output_cost_per_token": 0.000006, "supports_tool_choice": true, - "max_input_tokens": 65536, - "max_output_tokens": 52428, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/mistralai/voxtral-small-24b-2507": { - "input_cost_per_token": 1e-7, - "output_cost_per_token": 3e-7, - "cache_read_input_token_cost": 1e-8, - "input_cost_per_audio_token": 0.0001, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 26214, - "max_tokens": 26214, + "mistral/mistral-medium": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000075, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_tool_choice": true, + "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_pdf_input": true, - "supports_audio_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false - }, - "openrouter/moonshotai/kimi-k2": { - "input_cost_per_token": 5.7e-7, - "output_cost_per_token": 0.0000023, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 98304, - "max_tokens": 98304, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/moonshotai/kimi-k2-0905": { - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", + "mistral/mistral-medium-2604": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 98304, - "max_tokens": 98304, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000075, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": true, + "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/moonshotai/kimi-k2-thinking": { - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000025, + "mistral/mistral-medium-3": { "cache_read_input_token_cost": 1.5e-7, - "litellm_provider": "openrouter", + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 98304, - "max_tokens": 98304, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000075, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/moonshotai/kimi-k2.5": { - "cache_read_input_token_cost": 7e-8, - "input_cost_per_token": 4.5e-7, - "litellm_provider": "openrouter", + "mistral/mistral-medium-3-5": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.00000225, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000075, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/moonshotai/kimi-k2.6": { - "cache_read_input_token_cost": 1.6e-7, - "input_cost_per_token": 9.5e-7, - "litellm_provider": "openrouter", + "mistral/mistral-medium-3.5": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000075, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/moonshotai/kimi-k2.7-code": { - "cache_read_input_token_cost": 1.8e-7, - "input_cost_per_token": 6.712e-7, - "litellm_provider": "openrouter", + "mistral/mistral-medium-latest": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.00000335, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000075, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/moonshotai/kimi-k3": { - "cache_read_input_token_cost": 2.6e-7, - "input_cost_per_token": 7.9e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "mistral/mistral-small": { + "cache_read_input_token_cost": 1e-8, + "input_cost_per_token": 1e-7, + "litellm_provider": "mistral", + "max_input_tokens": 32000, + "max_output_tokens": 8191, + "max_tokens": 8191, "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 3e-7, + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/moonshotai/kimi-k3:batch": { - "cache_read_input_token_cost": 2.28e-7, - "input_cost_per_token": 0.00000228, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 16384, - "max_tokens": 16384, + "mistral/mistral-small-2603": { + "cache_read_input_token_cost": 1.5e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.0000114, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 6e-7, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/morph/morph-v3-fast": { - "input_cost_per_token": 8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 81920, - "max_output_tokens": 38000, - "max_tokens": 38000, - "mode": "chat", - "output_cost_per_token": 0.0000012, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/morph/morph-v3-large": { - "input_cost_per_token": 9e-7, - "litellm_provider": "openrouter", + "mistral/mistral-small-latest": { + "cache_read_input_token_cost": 1.5e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 131072, - "max_tokens": 131072, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.0000019, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "output_cost_per_token": 6e-7, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03", + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/nex-agi/nex-n2.5-mini": { - "cache_read_input_token_cost": 2.5e-9, - "input_cost_per_token": 2.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "mistral/mistral-tiny": { + "cache_read_input_token_cost": 2.5e-8, + "input_cost_per_token": 2.5e-7, + "litellm_provider": "mistral", + "max_input_tokens": 32000, + "max_output_tokens": 8191, + "max_tokens": 8191, "mode": "chat", - "output_cost_per_token": 1e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, + "output_cost_per_token": 2.5e-7, + "supports_assistant_prefill": true, "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/nex-agi/nex-n2.5-mini:free": { - "deprecation_date": "2026-09-25", - "input_cost_per_token": 0, - "litellm_provider": "openrouter", + "mistral/mistral-vibe-cli-fast": { + "cache_read_input_token_cost": 1.5e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 6e-7, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/nex-agi/nex-n2.5-pro": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 7.5e-8, - "litellm_provider": "openrouter", + "mistral/mistral-vibe-cli-latest": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 2.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000075, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/nex-agi/nex-n2.5-pro:free": { - "deprecation_date": "2026-09-25", - "input_cost_per_token": 0, - "litellm_provider": "openrouter", + "mistral/mistral-vibe-cli-with-tools": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000075, + "reasoning_effort_levels": [ + "none", + "high" + ], + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/nousresearch/hermes-3-llama-3.1-405b": { - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "mistral/open-mistral-nemo": { + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "mistral", + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.000001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "output_cost_per_token": 3e-7, + "source": "https://mistral.ai/technology/", + "supports_assistant_prefill": true, "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/nousresearch/hermes-3-llama-3.1-70b": { - "input_cost_per_token": 7e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "mistral/pixtral-large-latest": { + "cache_read_input_token_cost": 2e-7, + "input_cost_per_token": 0.000002, + "litellm_provider": "mistral", + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 7e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "output_cost_per_token": 0.000006, + "supports_assistant_prefill": true, + "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/nousresearch/hermes-4-405b": { - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "mistral/voxtral-small-2507": { + "cache_read_input_token_cost": 1e-8, + "input_cost_per_token": 1e-7, + "litellm_provider": "mistral", + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", - "output_cost_per_token": 0.000003, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, + "output_cost_per_token": 4e-7, + "source": "https://docs.mistral.ai/models/voxtral-small-25-07", + "supports_audio_input": true, + "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/nvidia/nemotron-3-nano-30b-a3b": { - "input_cost_per_token": 5e-8, - "output_cost_per_token": 2e-7, - "cache_read_input_token_cost": 3e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "mistral/voxtral-small-latest": { + "cache_read_input_token_cost": 1e-8, + "input_cost_per_token": 1e-7, + "litellm_provider": "mistral", + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 4e-7, + "source": "https://docs.mistral.ai/models/voxtral-small-25-07", + "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 256000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "mistral/zai-glm-5": { + "cache_read_input_token_cost": 1.4e-7, + "input_cost_per_token": 0.0000014, + "litellm_provider": "mistral", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000044, + "reasoning_effort_levels": [ + "low", + "high", + "max" + ], + "source": "https://docs.mistral.ai/models/zai-glm-5-3", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_vision": true, - "supports_audio_input": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_response_schema": false, - "supports_web_search": false + "supports_response_schema": true, + "supports_tool_choice": true }, - "openrouter/nvidia/nemotron-3-super-120b-a12b": { - "input_cost_per_token": 8e-8, - "output_cost_per_token": 4.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "mistral/zai-glm-5-2": { + "cache_read_input_token_cost": 1.4e-7, + "deprecation_date": "2026-10-31", + "input_cost_per_token": 0.0000014, + "litellm_provider": "mistral", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000044, + "reasoning_effort_levels": [ + "none", + "minimal", + "low", + "medium", + "high", + "xhigh", + "max" + ], + "source": "https://docs.mistral.ai/models/zai-glm-5-2", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/nvidia/nemotron-3-super-120b-a12b:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "mistral/zai-glm-5-3": { + "cache_read_input_token_cost": 1.4e-7, + "input_cost_per_token": 0.0000014, + "litellm_provider": "mistral", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000044, + "reasoning_effort_levels": [ + "low", + "high", + "max" + ], + "source": "https://docs.mistral.ai/models/zai-glm-5-3", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/nvidia/nemotron-3-ultra-550b-a55b": { - "input_cost_per_token": 5e-7, - "output_cost_per_token": 0.0000022, - "cache_read_input_token_cost": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 16384, - "max_tokens": 16384, + "mistral/zai-glm-latest": { + "cache_read_input_token_cost": 1.4e-7, + "input_cost_per_token": 0.0000014, + "litellm_provider": "mistral", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000044, + "reasoning_effort_levels": [ + "low", + "high", + "max" + ], + "source": "https://docs.mistral.ai/models/zai-glm-5-3", + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/nvidia/nemotron-3-ultra-550b-a55b:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "moonshot.kimi-k2-thinking": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 6e-7, + "input_cost_per_token_batches": 3e-7, + "input_cost_per_token_priority": 0.00000105, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 256000, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000025, + "output_cost_per_token_batches": 0.00000125, + "output_cost_per_token_priority": 0.00000438, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/nvidia/nemotron-3.5-content-safety": { - "input_cost_per_token": 2e-7, - "output_cost_per_token": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": false, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_vision": true, - "supports_web_search": false - }, - "openrouter/nvidia/nemotron-3.5-content-safety:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": false, "supports_reasoning": true, - "supports_response_schema": false, - "supports_vision": true, - "supports_web_search": false + "supports_system_messages": true, + "supports_native_structured_output": true, + "supports_response_schema": true, + "supports_vision": false }, - "openrouter/nvidia/nemotron-3.5-lightning": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 6e-8, - "litellm_provider": "openrouter", + "moonshot/kimi-k2.5": { + "cache_read_input_token_cost": 1e-7, + "input_cost_per_token": 6e-7, + "litellm_provider": "moonshot", "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 1.6e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000003, + "source": "https://platform.moonshot.ai/docs/guide/kimi-k2-5-quickstart", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_video_input": true, + "supports_vision": true }, - "openrouter/nvidia/nemotron-3.5-lightning:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "moonshot/kimi-k2.6": { + "cache_read_input_token_cost": 1.6e-7, + "input_cost_per_token": 9.5e-7, + "litellm_provider": "moonshot", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000004, + "source": "https://platform.kimi.ai/docs/pricing/chat-k26", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/openai/gpt-3.5-turbo": { - "input_cost_per_token": 5e-7, - "litellm_provider": "openrouter", - "max_tokens": 4096, - "mode": "chat", - "output_cost_per_token": 0.0000015, - "supports_tool_choice": true, - "max_input_tokens": 16385, - "max_output_tokens": 4096, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_video_input": true, + "supports_vision": true }, - "openrouter/openai/gpt-3.5-turbo-0613": { - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 4095, - "max_output_tokens": 3685, - "max_tokens": 3685, + "moonshot/kimi-k2.7-code": { + "cache_read_input_token_cost": 1.9e-7, + "input_cost_per_token": 9.5e-7, + "litellm_provider": "moonshot", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000004, + "source": "https://platform.kimi.ai/docs/pricing/chat-k27-code", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_prompt_caching": true, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_video_input": true, + "supports_vision": true }, - "openrouter/openai/gpt-3.5-turbo-16k": { + "moonshot/kimi-k3": { + "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, - "litellm_provider": "openrouter", - "max_tokens": 4096, + "litellm_provider": "moonshot", + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.000004, - "supports_tool_choice": true, - "max_input_tokens": 16385, - "max_output_tokens": 4096, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000015, + "reasoning_effort_levels": [ + "low", + "high", + "max" + ], + "source": "https://platform.kimi.ai/docs/pricing/chat-k3", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_video_input": true, + "supports_vision": true }, - "openrouter/openai/gpt-3.5-turbo-instruct": { - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 4095, - "max_output_tokens": 3685, - "max_tokens": 3685, + "moonshot/moonshot-v1-128k": { + "input_cost_per_token": 0.000002, + "litellm_provider": "moonshot", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": false, + "output_cost_per_token": 0.000005, + "source": "https://platform.moonshot.ai/docs/pricing", + "supports_function_calling": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/openai/gpt-3.5-turbo:batch": { - "input_cost_per_token": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 16385, - "max_output_tokens": 4096, - "max_tokens": 4096, + "moonshot/moonshot-v1-128k-vision-preview": { + "input_cost_per_token": 0.000002, + "litellm_provider": "moonshot", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 7.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000005, + "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": true + "supports_vision": true }, - "openrouter/openai/gpt-4": { - "input_cost_per_token": 0.00003, - "litellm_provider": "openrouter", - "max_tokens": 4096, + "moonshot/moonshot-v1-32k": { + "input_cost_per_token": 0.000001, + "litellm_provider": "moonshot", + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", - "output_cost_per_token": 0.00006, - "supports_tool_choice": true, - "max_input_tokens": 8191, - "max_output_tokens": 4096, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000003, + "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/openai/gpt-4-turbo": { - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, + "moonshot/moonshot-v1-32k-vision-preview": { + "input_cost_per_token": 0.000001, + "litellm_provider": "moonshot", + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000003, + "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": true, "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/openai/gpt-4-turbo-preview": { - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, + "moonshot/moonshot-v1-8k": { + "input_cost_per_token": 2e-7, + "litellm_provider": "moonshot", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "source": "https://openrouter.ai/openai/gpt-4-turbo-preview", + "output_cost_per_token": 0.000002, + "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, - "supports_tool_choice": true, "supports_response_schema": true, - "supports_vision": false + "supports_tool_choice": true }, - "openrouter/openai/gpt-4-turbo:batch": { - "input_cost_per_token": 0.000005, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, + "moonshot/moonshot-v1-8k-vision-preview": { + "input_cost_per_token": 2e-7, + "litellm_provider": "moonshot", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000002, + "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/openai/gpt-4.1": { - "cache_read_input_token_cost": 5e-7, + "moonshot/moonshot-v1-auto": { "input_cost_per_token": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1047576, - "max_output_tokens": 32768, - "max_tokens": 32768, + "litellm_provider": "moonshot", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.000008, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000005, + "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": true, + "supports_tool_choice": true + }, + "moonshotai.kimi-k2.5": { + "input_cost_per_token": 6e-7, + "input_cost_per_token_batches": 3e-7, + "input_cost_per_token_flex": 3e-7, + "input_cost_per_token_priority": 0.00000105, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 256000, + "max_output_tokens": 16000, + "max_tokens": 16000, + "mode": "chat", + "output_cost_per_token": 0.000003, + "output_cost_per_token_batches": 0.0000015, + "output_cost_per_token_flex": 0.0000015, + "output_cost_per_token_priority": 0.00000525, + "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_audio_input": false, + "supports_response_schema": true }, - "openrouter/openai/gpt-4.1-mini": { - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1047576, - "max_output_tokens": 32768, - "max_tokens": 32768, + "moonshotai.kimi-k3": { + "supports_regex_lookaround": false, + "cache_creation_input_token_cost": 0.000004125, + "cache_creation_input_token_cost_flex": 0.0000020625, + "cache_creation_input_token_cost_priority": 0.00000721875, + "cache_read_input_token_cost": 3.3e-7, + "cache_read_input_token_cost_flex": 1.65e-7, + "cache_read_input_token_cost_priority": 5.775e-7, + "input_cost_per_token": 0.0000033, + "input_cost_per_token_flex": 0.00000165, + "input_cost_per_token_priority": 0.000005775, + "litellm_provider": "bedrock_converse", + "token_count_framing": 64, + "token_count_tokenizer": "kimi_k3", + "max_input_tokens": 1000000, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.0000016, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000165, + "output_cost_per_token_flex": 0.00000825, + "output_cost_per_token_priority": 0.000028875, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, + "supports_prompt_cache_breakpoint": false, "supports_prompt_caching": true, - "supports_reasoning": false, + "supports_reasoning": true, "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/openai/gpt-4.1-mini:batch": { - "cache_read_input_token_cost": 5e-8, - "input_cost_per_token": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1047576, - "max_output_tokens": 32768, - "max_tokens": 32768, + "morph/morph-v3-fast": { + "input_cost_per_token": 8e-7, + "litellm_provider": "morph", + "max_input_tokens": 16000, + "max_output_tokens": 16000, + "max_tokens": 16000, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "supports_function_calling": false, + "supports_parallel_function_calling": false, + "supports_system_messages": true, + "supports_tool_choice": false, + "supports_vision": false + }, + "morph/morph-v3-large": { + "input_cost_per_token": 9e-7, + "litellm_provider": "morph", + "max_input_tokens": 16000, + "max_output_tokens": 16000, + "max_tokens": 16000, + "mode": "chat", + "output_cost_per_token": 0.0000019, + "supports_function_calling": false, + "supports_parallel_function_calling": false, + "supports_system_messages": true, + "supports_tool_choice": false, + "supports_vision": false + }, + "nebius/MiniMaxAI/MiniMax-M2.5": { + "max_tokens": 196608, + "max_input_tokens": 196608, + "max_output_tokens": 196608, + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000012, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 8e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_reasoning": true, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/MiniMaxAI%2FMiniMax-M2.5" }, - "openrouter/openai/gpt-4.1-nano": { - "cache_read_input_token_cost": 2.5e-8, - "input_cost_per_token": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1047576, - "max_output_tokens": 32768, - "max_tokens": 32768, + "nebius/MiniMaxAI/MiniMax-M3": { + "max_tokens": 1048576, + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000012, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 4e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_system_messages": true, - "supports_tool_choice": true, + "supports_reasoning": true, "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/MiniMaxAI%2FMiniMax-M3" }, - "openrouter/openai/gpt-4.1-nano:batch": { - "cache_read_input_token_cost": 1.25e-8, - "input_cost_per_token": 5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1047576, - "max_output_tokens": 32768, - "max_tokens": 32768, + "nebius/NousResearch/Hermes-3-Llama-3.1-405B": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 0.000001, + "output_cost_per_token": 0.000003, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-4.1:batch": { - "cache_read_input_token_cost": 2.5e-7, + "nebius/NousResearch/Hermes-4-405B": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 1047576, - "max_output_tokens": 32768, + "output_cost_per_token": 0.000003, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "supports_reasoning": true, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/NousResearch%2FHermes-4-405B" + }, + "nebius/NousResearch/Hermes-4-70B": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 1.3e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "supports_reasoning": true, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/NousResearch%2FHermes-4-70B" + }, + "nebius/Qwen/QwQ-32B": { "max_tokens": 32768, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 4.5e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_reasoning": true, + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-4o": { - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "nebius/Qwen/Qwen2-VL-72B-Instruct": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 1.3e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.00001, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_tool_choice": true, "supports_vision": true, - "cache_read_input_token_cost": 0.00000125, - "supports_prompt_caching": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-4o-2024-05-13": { - "input_cost_per_token": 0.000005, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, + "nebius/Qwen/Qwen2-VL-7B-Instruct": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 2e-8, + "output_cost_per_token": 6e-8, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-4o-2024-08-06": { - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, - "litellm_provider": "openrouter", + "nebius/Qwen/Qwen2.5-32B-Instruct": { + "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "max_output_tokens": 128000, + "input_cost_per_token": 6e-8, + "output_cost_per_token": 2e-7, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "supports_web_search": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-4o-2024-11-20": { - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, - "litellm_provider": "openrouter", + "nebius/Qwen/Qwen2.5-72B-Instruct": { + "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "max_output_tokens": 128000, + "input_cost_per_token": 1.3e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "source": "https://nebius.com/prices" + }, + "nebius/Qwen/Qwen2.5-Coder-7B": { + "max_tokens": 32768, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "input_cost_per_token": 1e-8, + "output_cost_per_token": 3e-8, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "source": "https://nebius.com/prices" + }, + "nebius/Qwen/Qwen2.5-VL-72B-Instruct": { + "max_tokens": 32000, + "max_input_tokens": 32000, + "max_output_tokens": 32000, + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 7.5e-7, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "supports_web_search": true, "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false + "source": "https://tokenfactory.nebius.com/models/catalog/image2text/Qwen%2FQwen2.5-VL-72B-Instruct" }, - "openrouter/openai/gpt-4o-mini": { - "input_cost_per_token": 1.5e-7, + "nebius/Qwen/Qwen3-14B": { + "max_tokens": 32768, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "input_cost_per_token": 8e-8, + "output_cost_per_token": 2.4e-7, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "source": "https://nebius.com/prices" + }, + "nebius/Qwen/Qwen3-235B-A22B": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 2e-7, "output_cost_per_token": 6e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "source": "https://nebius.com/prices" + }, + "nebius/Qwen/Qwen3-235B-A22B-Instruct-2507": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 2e-7, + "output_cost_per_token": 6e-7, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-235B-A22B-Instruct-2507" + }, + "nebius/Qwen/Qwen3-30B-A3B": { + "max_tokens": 32768, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "input_cost_per_token": 1e-7, + "output_cost_per_token": 3e-7, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 7.5e-8, - "supports_prompt_caching": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-4o-mini-2024-07-18": { - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 6e-7, - "cache_read_input_token_cost": 7.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "nebius/Qwen/Qwen3-30B-A3B-Instruct-2507": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1e-7, + "output_cost_per_token": 3e-7, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "supports_web_search": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-30B-A3B-Instruct-2507" }, - "openrouter/openai/gpt-4o-mini:batch": { - "cache_read_input_token_cost": 3.75e-8, - "input_cost_per_token": 7.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", + "nebius/Qwen/Qwen3-32B": { + "max_tokens": 40960, + "max_input_tokens": 40960, + "max_output_tokens": 40960, + "input_cost_per_token": 1e-7, "output_cost_per_token": 3e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "litellm_provider": "nebius", + "mode": "chat", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-32B" }, - "openrouter/openai/gpt-4o:batch": { - "cache_read_input_token_cost": 6.25e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "nebius/Qwen/Qwen3-4B": { + "max_tokens": 32768, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "input_cost_per_token": 8e-8, + "output_cost_per_token": 2.4e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5": { - "cache_read_input_token_cost": 1.25e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, + "nebius/Qwen/Qwen3-Next-80B-A3B-Thinking": { "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 0.0000012, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.00001, - "source": "https://openrouter.ai/api/v1/models", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-Next-80B-A3B-Thinking" }, - "openrouter/openai/gpt-5-chat": { - "cache_read_input_token_cost": 1.25e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "nebius/Qwen/Qwen3.5-397B-A17B": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000036, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.00001, - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], + "supports_function_calling": true, "supports_reasoning": true, - "supports_tool_choice": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3.5-397B-A17B" }, - "openrouter/openai/gpt-5-codex": { - "cache_read_input_token_cost": 1.25e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 272000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/Qwen/Qwen3.8-27B": { + "input_cost_per_token": 4.5e-7, + "litellm_provider": "nebius", + "max_input_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.00001, - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_reasoning": true, - "supports_tool_choice": true + "output_cost_per_token": 0.000003, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3.8-27B", + "supports_function_calling": true, + "supports_reasoning": true }, - "openrouter/openai/gpt-5-image": { - "cache_read_input_token_cost": 0.00000125, - "input_cost_per_token": 0.00001, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-R1": { "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 8e-7, + "output_cost_per_token": 0.0000024, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.00001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, + "supports_function_calling": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5-image-mini": { - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-R1-0528": { + "max_tokens": 164000, + "max_input_tokens": 164000, + "max_output_tokens": 164000, + "input_cost_per_token": 8e-7, + "output_cost_per_token": 0.0000024, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, + "supports_function_calling": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5-mini": { - "cache_read_input_token_cost": 2.5e-8, - "input_cost_per_token": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-R1-Distill-Llama-70B": { "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 7.5e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://openrouter.ai/api/v1/models", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5-mini:batch": { - "cache_read_input_token_cost": 1.25e-8, - "input_cost_per_token": 1.25e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-V3": { "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 5e-7, + "output_cost_per_token": 0.0000015, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5-nano": { - "cache_read_input_token_cost": 5e-9, - "input_cost_per_token": 5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-V3-0324": { "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 5e-7, + "output_cost_per_token": 0.0000015, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 4e-7, - "source": "https://openrouter.ai/api/v1/models", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5-nano:batch": { - "cache_read_input_token_cost": 2.5e-9, - "input_cost_per_token": 2.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-V4-Flash": { + "max_tokens": 1048576, + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 2.8e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Flash" }, - "openrouter/openai/gpt-5-pro": { - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.00012, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-V4-Flash-0731": { + "max_tokens": 1024000, + "max_input_tokens": 1024000, + "max_output_tokens": 1024000, + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 2.8e-7, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_tool_choice": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Flash-0731" }, - "openrouter/openai/gpt-5-pro:batch": { - "input_cost_per_token": 0.0000075, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-V4-Pro": { + "max_tokens": 1048576, + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "input_cost_per_token": 0.00000175, + "output_cost_per_token": 0.0000035, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.00006, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Pro" }, - "openrouter/openai/gpt-5.1": { - "input_cost_per_token": 0.00000125, - "output_cost_per_token": 0.00001, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-V4-Pro-0813": { + "input_cost_per_token": 0.00000132, + "litellm_provider": "nebius", + "max_input_tokens": 979000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.00000396, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Pro-0813", "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 1.25e-7, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_reasoning": true }, - "openrouter/openai/gpt-5.1-codex": { - "input_cost_per_token": 0.00000125, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 1.3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/deepseek-ai/DeepSeek-V4.1-Flash": { + "input_cost_per_token": 3e-7, + "litellm_provider": "nebius", + "max_input_tokens": 1048000, + "max_output_tokens": 384000, + "max_tokens": 384000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000012, + "source": "https://tokenfactory.nebius.com/endpoints?modals=endpoint-details&model-id=deepseek-ai/DeepSeek-V4.1-Flash", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/openai/gpt-5.1-codex-max": { - "cache_read_input_token_cost": 1.25e-7, - "input_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/google/gemma-3-27b-it": { + "max_tokens": 110000, + "max_input_tokens": 110000, + "max_output_tokens": 110000, + "input_cost_per_token": 1e-7, + "output_cost_per_token": 3e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.00001, - "source": "https://openrouter.ai/api/v1/models", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/google%2Fgemma-3-27b-it" }, - "openrouter/openai/gpt-5.1-codex-mini": { - "input_cost_per_token": 2.5e-7, - "output_cost_per_token": 0.000002, - "cache_read_input_token_cost": 3e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/meta-llama/Llama-3.3-70B-Instruct": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 1.3e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5.1:batch": { - "cache_read_input_token_cost": 6.25e-8, - "input_cost_per_token": 6.25e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, + "nebius/meta-llama/Llama-Guard-3-8B": { + "max_tokens": 128000, + "max_input_tokens": 128000, "max_output_tokens": 128000, + "input_cost_per_token": 2e-8, + "output_cost_per_token": 6e-8, + "litellm_provider": "nebius", + "mode": "chat", + "source": "https://nebius.com/prices" + }, + "nebius/meta-llama/Meta-Llama-3.1-405B-Instruct": { "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 0.000001, + "output_cost_per_token": 0.000003, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5.2": { - "input_cost_per_image": 0, - "cache_read_input_token_cost": 1.75e-7, - "input_cost_per_token": 0.00000175, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, + "nebius/meta-llama/Meta-Llama-3.1-70B-Instruct": { "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 1.3e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000014, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5.2-chat": { - "input_cost_per_image": 0, - "cache_read_input_token_cost": 1.75e-7, - "input_cost_per_token": 0.00000175, - "litellm_provider": "openrouter", + "nebius/meta-llama/Meta-Llama-3.1-8B-Instruct": { + "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 32000, - "max_tokens": 32000, + "max_output_tokens": 128000, + "input_cost_per_token": 2e-8, + "output_cost_per_token": 6e-8, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000014, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://nebius.com/prices" }, - "openrouter/openai/gpt-5.2-codex": { - "cache_read_input_token_cost": 1.75e-7, - "input_cost_per_token": 0.00000175, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, + "nebius/mistralai/Mistral-Nemo-Instruct-2407": { "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 4e-8, + "output_cost_per_token": 1.2e-7, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "source": "https://nebius.com/prices" + }, + "nebius/moonshotai/Kimi-K2.6": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 9.5e-7, + "output_cost_per_token": 0.000004, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000014, - "source": "https://openrouter.ai/api/v1/models", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/image2text/moonshotai%2FKimi-K2.6" }, - "openrouter/openai/gpt-5.2-pro": { - "input_cost_per_image": 0, - "input_cost_per_token": 0.000021, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/moonshotai/Kimi-K2.7-Code": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 9.5e-7, + "output_cost_per_token": 0.000004, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000168, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/moonshotai%2FKimi-K2.7-Code" }, - "openrouter/openai/gpt-5.2-pro:batch": { - "input_cost_per_token": 0.0000105, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/moonshotai/Kimi-K3": { + "max_tokens": 1024000, + "max_input_tokens": 1024000, + "max_output_tokens": 1024000, + "input_cost_per_token": 0.000003, + "output_cost_per_token": 0.000015, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000084, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/image2text/moonshotai%2FKimi-K3" }, - "openrouter/openai/gpt-5.2:batch": { - "cache_read_input_token_cost": 8.75e-8, - "input_cost_per_token": 8.75e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/nvidia/Cosmos3-Super-Reasoner": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1e-7, + "output_cost_per_token": 3e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000007, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/image2text/nvidia%2FCosmos3-Super-Reasoner" }, - "openrouter/openai/gpt-5.3-codex": { - "input_cost_per_token": 0.00000175, - "output_cost_per_token": 0.000014, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, + "nebius/nvidia/Llama-3.1-Nemotron-Ultra-253B-v1": { "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000018, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, - "supports_tool_choice": true, + "source": "https://nebius.com/prices" + }, + "nebius/nvidia/Llama-3.3-Nemotron-Super-49B-v1": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 1e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "source": "https://nebius.com/prices" + }, + "nebius/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000018, + "litellm_provider": "nebius", + "mode": "chat", "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 1.75e-7, - "supports_prompt_caching": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FLlama-3_1-Nemotron-Ultra-253B-v1" }, - "openrouter/openai/gpt-5.4": { - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000015, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 6e-8, + "output_cost_per_token": 2.4e-7, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, - "supports_tool_choice": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 2.5e-7, - "cache_read_input_token_cost_above_272k_tokens": 5e-7, - "input_cost_per_token_above_272k_tokens": 0.000005, - "output_cost_per_token_above_272k_tokens": 0.0000225, - "supports_prompt_caching": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNVIDIA-Nemotron-3-Nano-30B-A3B" }, - "openrouter/openai/gpt-5.4-image-2": { - "cache_read_input_token_cost": 0.000002, - "input_cost_per_token": 0.000008, - "litellm_provider": "openrouter", - "max_input_tokens": 272000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/nvidia/Nemotron-3-Nano-Omni": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 6e-8, + "output_cost_per_token": 2.4e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": true, - "supports_prompt_caching": true, + "supports_function_calling": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": false, "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3-Nano-Omni" }, - "openrouter/openai/gpt-5.4-mini": { - "input_cost_per_token": 7.5e-7, - "output_cost_per_token": 0.0000045, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/nvidia/Nemotron-3-Ultra-550b-a55b": { + "max_tokens": 1048576, + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "input_cost_per_token": 0.000001, + "output_cost_per_token": 0.000003, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, - "supports_tool_choice": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 7.5e-8, - "supports_prompt_caching": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3-Ultra-550b-a55b" }, - "openrouter/openai/gpt-5.4-mini:batch": { - "cache_read_input_token_cost": 3.75e-8, - "input_cost_per_token": 3.75e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/nvidia/Nemotron-3_5-Lightning": { + "max_tokens": 1048576, + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "input_cost_per_token": 6e-8, + "output_cost_per_token": 2.4e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.00000225, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3_5-Lightning" }, - "openrouter/openai/gpt-5.4-nano": { - "input_cost_per_token": 2e-7, - "output_cost_per_token": 0.00000125, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/nvidia/nemotron-3-super-120b-a12b": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 3e-7, + "output_cost_per_token": 9e-7, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", "supports_function_calling": true, - "supports_tool_choice": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 2e-8, - "supports_prompt_caching": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2Fnemotron-3-super-120b-a12b" }, - "openrouter/openai/gpt-5.4-nano:batch": { - "cache_read_input_token_cost": 1e-8, - "input_cost_per_token": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/openai/gpt-oss-120b": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 6e-7, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 6.25e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/openai%2Fgpt-oss-120b" + }, + "nebius/openbmb/MiniCPM-V-4_5": { + "max_tokens": 32000, + "max_input_tokens": 32000, + "max_output_tokens": 32000, + "input_cost_per_token": 6.58e-7, + "output_cost_per_token": 0.00000111, + "litellm_provider": "nebius", + "mode": "chat", "supports_vision": true, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/image2text/openbmb%2FMiniCPM-V-4_5" }, - "openrouter/openai/gpt-5.4-pro": { - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00018, - "input_cost_per_token_above_272k_tokens": 0.00006, - "output_cost_per_token_above_272k_tokens": 0.00027, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/reflection/beam": { + "input_cost_per_token": 2.5e-7, + "litellm_provider": "nebius", + "max_input_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.000001, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/reflection%2Fbeam", + "supports_function_calling": true, + "supports_reasoning": true + }, + "nebius/zai-org/GLM-5.1": { + "max_tokens": 202752, + "max_input_tokens": 202752, + "max_output_tokens": 202752, + "input_cost_per_token": 0.0000014, + "output_cost_per_token": 0.0000044, + "litellm_provider": "nebius", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_tool_choice": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_web_search": true + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.1" }, - "openrouter/openai/gpt-5.4-pro:batch": { - "input_cost_per_token": 0.000015, - "input_cost_per_token_above_272k_tokens": 0.00003, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "nebius/zai-org/GLM-5.2": { + "max_tokens": 1048576, + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "input_cost_per_token": 0.0000014, + "output_cost_per_token": 0.0000044, + "litellm_provider": "nebius", "mode": "chat", - "output_cost_per_token": 0.00009, - "output_cost_per_token_above_272k_tokens": 0.000135, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, "supports_reasoning": true, - "supports_response_schema": true, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.2" + }, + "nebius/zai-org/GLM-5.3": { + "input_cost_per_token": 0.0000014, + "litellm_provider": "nebius", + "max_input_tokens": 1024000, + "max_tokens": 1024000, + "mode": "chat", + "output_cost_per_token": 0.0000044, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.3", + "supports_function_calling": true, + "supports_reasoning": true + }, + "nebius/zai-org/GLM-5.3-Flash": { + "max_tokens": 1024000, + "max_input_tokens": 1024000, + "max_output_tokens": 1024000, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 5e-7, + "litellm_provider": "nebius", + "mode": "chat", + "supports_function_calling": true, + "supports_reasoning": true, + "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.3-Flash", + "supports_vision": true + }, + "novita/Sao10K/L3-8B-Stheno-v3.2": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 5e-8, + "output_cost_per_token": 5e-8, + "max_input_tokens": 8192, + "max_output_tokens": 32000, + "max_tokens": 32000, + "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true }, - "openrouter/openai/gpt-5.4:batch": { - "cache_read_input_token_cost": 1.25e-7, - "cache_read_input_token_cost_above_272k_tokens": 2.5e-7, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_272k_tokens": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/baichuan/baichuan-m2-32b": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.0000075, - "output_cost_per_token_above_272k_tokens": 0.00001125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 7e-8, + "output_cost_per_token": 7e-8, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, + "supports_system_messages": true + }, + "novita/baidu/cobuddy": { + "cache_read_input_token_cost": 7e-8, + "input_cost_per_token": 2.8e-7, + "litellm_provider": "novita", + "max_input_tokens": 131072, + "max_output_tokens": 65536, + "max_tokens": 65536, + "mode": "chat", + "output_cost_per_token": 0.00000113, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-5.5": { - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.00003, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/baidu/ernie-4.5-21B-a3b": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "input_cost_per_token": 7e-8, + "output_cost_per_token": 2.8e-7, + "max_input_tokens": 120000, + "max_output_tokens": 8000, + "max_tokens": 8000, "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_above_272k_tokens": 0.000001, - "input_cost_per_token_above_272k_tokens": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000045, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_system_messages": true }, - "openrouter/openai/gpt-5.5-pro": { - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00018, - "input_cost_per_token_above_272k_tokens": 0.00006, - "output_cost_per_token_above_272k_tokens": 0.00027, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/baidu/ernie-4.5-21B-a3b-thinking": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, + "input_cost_per_token": 7e-8, + "output_cost_per_token": 2.8e-7, + "max_input_tokens": 131072, + "max_output_tokens": 65536, + "max_tokens": 65536, + "supports_system_messages": true, + "supports_reasoning": true + }, + "novita/baidu/ernie-4.5-300b-a47b-paddle": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 2.8e-7, + "output_cost_per_token": 0.0000011, + "max_input_tokens": 123000, + "max_output_tokens": 12000, + "max_tokens": 12000, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/openai/gpt-5.5-pro:batch": { - "input_cost_per_token": 0.000015, - "input_cost_per_token_above_272k_tokens": 0.00003, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/baidu/ernie-4.5-vl-28b-a3b": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.00009, - "output_cost_per_token_above_272k_tokens": 0.000135, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 5.6e-7, + "max_input_tokens": 30000, + "max_output_tokens": 8000, + "max_tokens": 8000, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_reasoning": true }, - "openrouter/openai/gpt-5.5:batch": { - "cache_read_input_token_cost": 2.5e-7, - "cache_read_input_token_cost_above_272k_tokens": 5e-7, - "input_cost_per_token": 0.0000025, - "input_cost_per_token_above_272k_tokens": 0.000005, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/baidu/ernie-4.5-vl-28b-a3b-thinking": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000015, - "output_cost_per_token_above_272k_tokens": 0.0000225, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 3.9e-7, + "output_cost_per_token": 3.9e-7, + "max_input_tokens": 131072, + "max_output_tokens": 65536, + "max_tokens": 65536, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true, + "supports_reasoning": true }, - "openrouter/openai/gpt-5.6-luna": { - "cache_creation_input_token_cost": 2.5e-7, - "cache_creation_input_token_cost_above_272k_tokens": 5e-7, - "input_cost_per_token": 2e-7, - "output_cost_per_token": 0.0000012, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/baidu/ernie-4.5-vl-424b-a47b": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, + "input_cost_per_token": 4.2e-7, + "output_cost_per_token": 0.00000125, + "max_input_tokens": 123000, + "max_output_tokens": 16000, + "max_tokens": 16000, + "supports_vision": true, + "supports_system_messages": true, + "supports_reasoning": true + }, + "novita/deepseek/deepseek-ocr": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 3e-8, + "output_cost_per_token": 3e-8, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 2e-8, - "cache_read_input_token_cost_above_272k_tokens": 4e-8, - "input_cost_per_token_above_272k_tokens": 4e-7, - "output_cost_per_token_above_272k_tokens": 0.0000018, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/openai/gpt-5.6-luna-pro": { - "input_cost_per_token": 2e-7, - "output_cost_per_token": 0.0000012, - "cache_read_input_token_cost": 2e-8, - "cache_creation_input_token_cost": 2.5e-7, - "cache_creation_input_token_cost_above_272k_tokens": 5e-7, - "input_cost_per_token_above_272k_tokens": 4e-7, - "output_cost_per_token_above_272k_tokens": 0.0000018, - "cache_read_input_token_cost_above_272k_tokens": 4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-ocr-2": { + "input_cost_per_token": 3e-8, + "litellm_provider": "novita", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 3e-8, + "source": "https://novita.ai/pricing", + "supports_vision": true + }, + "novita/deepseek/deepseek-prover-v2-671b": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_web_search": true + "input_cost_per_token": 7e-7, + "output_cost_per_token": 0.0000025, + "max_input_tokens": 160000, + "max_output_tokens": 160000, + "max_tokens": 160000, + "supports_system_messages": true }, - "openrouter/openai/gpt-5.6-luna-pro:batch": { - "cache_read_input_token_cost": 1e-8, - "cache_read_input_token_cost_above_272k_tokens": 2e-8, - "input_cost_per_token": 1e-7, - "input_cost_per_token_above_272k_tokens": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-r1": { + "input_cost_per_token": 0.000004, + "litellm_provider": "novita", + "max_input_tokens": 64000, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", - "output_cost_per_token": 6e-7, - "output_cost_per_token_above_272k_tokens": 9e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000004, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-5.6-luna:batch": { - "cache_read_input_token_cost": 1e-8, - "cache_read_input_token_cost_above_272k_tokens": 2e-8, - "input_cost_per_token": 1e-7, - "input_cost_per_token_above_272k_tokens": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-r1-0528": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 6e-7, - "output_cost_per_token_above_272k_tokens": 9e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 7e-7, + "output_cost_per_token": 0.0000025, + "max_input_tokens": 163840, + "max_output_tokens": 32768, + "max_tokens": 32768, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true, + "cache_read_input_token_cost": 3.5e-7, + "input_cost_per_token_cache_hit": 3.5e-7, + "supports_reasoning": true }, - "openrouter/openai/gpt-5.6-sol": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_272k_tokens": 4e-7, - "default_reasoning_effort": "medium", - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-r1-0528-qwen3-8b": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000015, - "reasoning_effort_levels": [ - "none", - "low", - "medium", - "high", - "xhigh", - "max" - ], - "source": "https://openrouter.ai/api/v1/models", - "supported_modalities": [ - "text", - "image" - ], - "supported_output_modalities": [ - "text" - ], - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "input_cost_per_token": 6e-8, + "output_cost_per_token": 9e-8, + "max_input_tokens": 128000, + "max_output_tokens": 32000, + "max_tokens": 32000, + "supports_system_messages": true, + "supports_reasoning": true + }, + "novita/deepseek/deepseek-r1-distill-llama-70b": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 8e-7, + "output_cost_per_token": 8e-7, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true, + "supports_reasoning": true }, - "openrouter/openai/gpt-5.6-sol-pro": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_272k_tokens": 4e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-r1-distill-qwen-14b": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 1.5e-7, + "max_input_tokens": 32768, + "max_output_tokens": 16384, + "max_tokens": 16384, + "supports_tool_choice": true, + "supports_system_messages": true, "supports_response_schema": true, + "supports_reasoning": true + }, + "novita/deepseek/deepseek-r1-distill-qwen-32b": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 3e-7, + "output_cost_per_token": 3e-7, + "max_input_tokens": 64000, + "max_output_tokens": 32000, + "max_tokens": 32000, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true, + "supports_reasoning": true }, - "openrouter/openai/gpt-5.6-sol-pro:batch": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_272k_tokens": 0.0000025, - "cache_read_input_token_cost": 1e-7, - "cache_read_input_token_cost_above_272k_tokens": 2e-7, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_272k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-r1-turbo": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000005, - "output_cost_per_token_above_272k_tokens": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 7e-7, + "output_cost_per_token": 0.0000025, + "max_input_tokens": 64000, + "max_output_tokens": 16000, + "max_tokens": 16000, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_reasoning": true, + "supports_response_schema": true }, - "openrouter/openai/gpt-5.6-sol:batch": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_272k_tokens": 0.0000025, - "cache_read_input_token_cost": 1e-7, - "cache_read_input_token_cost_above_272k_tokens": 2e-7, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_272k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-r1/community": { + "input_cost_per_token": 0.000004, + "litellm_provider": "novita", + "max_input_tokens": 64000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", - "output_cost_per_token": 0.000005, - "output_cost_per_token_above_272k_tokens": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000004, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-5.6-terra": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000012, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v3-0324": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "input_cost_per_token": 2.7e-7, + "output_cost_per_token": 0.00000112, + "max_input_tokens": 163840, + "max_output_tokens": 65536, + "max_tokens": 65536, "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_272k_tokens": 4e-7, - "input_cost_per_token_above_272k_tokens": 0.000004, - "output_cost_per_token_above_272k_tokens": 0.000018, - "supports_prompt_caching": true, - "supports_web_search": true + "cache_read_input_token_cost": 1.35e-7, + "input_cost_per_token_cache_hit": 1.35e-7 }, - "openrouter/openai/gpt-5.6-terra-pro": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000012, - "cache_read_input_token_cost": 2e-7, - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "input_cost_per_token_above_272k_tokens": 0.000004, - "output_cost_per_token_above_272k_tokens": 0.000018, - "cache_read_input_token_cost_above_272k_tokens": 4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v3-turbo": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 4e-7, + "output_cost_per_token": 0.0000013, + "max_input_tokens": 64000, + "max_output_tokens": 16000, + "max_tokens": 16000, "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_system_messages": true }, - "openrouter/openai/gpt-5.6-terra-pro:batch": { - "cache_read_input_token_cost": 1e-7, - "cache_read_input_token_cost_above_272k_tokens": 2e-7, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_272k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v3.1": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000006, - "output_cost_per_token_above_272k_tokens": 0.000009, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 2.7e-7, + "output_cost_per_token": 0.000001, + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true, + "cache_read_input_token_cost": 1.35e-7, + "input_cost_per_token_cache_hit": 1.35e-7, + "supports_reasoning": true }, - "openrouter/openai/gpt-5.6-terra:batch": { - "cache_read_input_token_cost": 1e-7, - "cache_read_input_token_cost_above_272k_tokens": 2e-7, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_272k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v3.1-terminus": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000006, - "output_cost_per_token_above_272k_tokens": 0.000009, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 2.7e-7, + "output_cost_per_token": 0.000001, + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true, + "cache_read_input_token_cost": 1.35e-7, + "input_cost_per_token_cache_hit": 1.35e-7, + "supports_reasoning": true }, - "openrouter/openai/gpt-5:batch": { - "cache_read_input_token_cost": 6.25e-8, - "input_cost_per_token": 6.25e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v3.2": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 2.69e-7, + "output_cost_per_token": 4e-7, + "max_input_tokens": 163840, + "max_output_tokens": 65536, + "max_tokens": 65536, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true, + "cache_read_input_token_cost": 1.345e-7, + "input_cost_per_token_cache_hit": 1.345e-7, + "supports_reasoning": true }, - "openrouter/openai/gpt-6-astra": { - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00005, - "cache_read_input_token_cost": 0.000001, - "cache_creation_input_token_cost": 0.0000125, - "input_cost_per_token_above_272k_tokens": 0.00002, - "output_cost_per_token_above_272k_tokens": 0.000075, - "cache_read_input_token_cost_above_272k_tokens": 0.000002, - "cache_creation_input_token_cost_above_272k_tokens": 0.000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v3.2-exp": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 2.7e-7, + "output_cost_per_token": 4.1e-7, + "max_input_tokens": 163840, + "max_output_tokens": 65536, + "max_tokens": 65536, "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_reasoning": true }, - "openrouter/openai/gpt-6-astra-pro": { - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00005, - "cache_read_input_token_cost": 0.000001, - "cache_creation_input_token_cost": 0.0000125, - "cache_creation_input_token_cost_above_272k_tokens": 0.000025, - "input_cost_per_token_above_272k_tokens": 0.00002, - "output_cost_per_token_above_272k_tokens": 0.000075, - "cache_read_input_token_cost_above_272k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v3/community": { + "input_cost_per_token": 8.900000000000001e-7, + "litellm_provider": "novita", + "max_input_tokens": 64000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 8.900000000000001e-7, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-6-astra-pro:batch": { - "cache_creation_input_token_cost": 0.00000625, - "cache_creation_input_token_cost_above_272k_tokens": 0.0000125, - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_above_272k_tokens": 0.000001, - "input_cost_per_token": 0.000005, - "input_cost_per_token_above_272k_tokens": 0.00001, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v4-flash": { + "cache_read_input_token_cost": 2.8e-8, + "input_cost_per_token": 1.4e-7, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 393216, + "max_tokens": 393216, "mode": "chat", - "output_cost_per_token": 0.000025, - "output_cost_per_token_above_272k_tokens": 0.0000375, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 2.8e-7, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-6-astra:batch": { - "cache_creation_input_token_cost": 0.00000625, - "cache_creation_input_token_cost_above_272k_tokens": 0.0000125, - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_above_272k_tokens": 0.000001, - "input_cost_per_token": 0.000005, - "input_cost_per_token_above_272k_tokens": 0.00001, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v4-flash-0731": { + "cache_read_input_token_cost": 2.8e-8, + "input_cost_per_token": 4.4e-7, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 393216, + "max_tokens": 393216, "mode": "chat", - "output_cost_per_token": 0.000025, - "output_cost_per_token_above_272k_tokens": 0.0000375, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000132, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-6-luna": { - "cache_creation_input_token_cost": 1.25e-7, - "cache_creation_input_token_cost_above_272k_tokens": 2.5e-7, - "cache_read_input_token_cost": 1e-8, - "cache_read_input_token_cost_above_272k_tokens": 2e-8, - "input_cost_per_token": 1e-7, - "input_cost_per_token_above_272k_tokens": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v4-flash-vision-exp": { + "cache_read_input_token_cost": 2.8e-8, + "input_cost_per_token": 4.4e-7, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 393216, + "max_tokens": 393216, "mode": "chat", - "output_cost_per_token": 5e-7, - "output_cost_per_token_above_272k_tokens": 7.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000132, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/openai/gpt-6-luna-pro": { - "cache_creation_input_token_cost": 1.25e-7, - "cache_creation_input_token_cost_above_272k_tokens": 2.5e-7, - "cache_read_input_token_cost": 1e-8, - "cache_read_input_token_cost_above_272k_tokens": 2e-8, - "input_cost_per_token": 1e-7, - "input_cost_per_token_above_272k_tokens": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v4-pro": { + "cache_read_input_token_cost": 1.35e-7, + "input_cost_per_token": 0.0000016000000000000001, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 393216, + "max_tokens": 393216, "mode": "chat", - "output_cost_per_token": 5e-7, - "output_cost_per_token_above_272k_tokens": 7.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000032000000000000003, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-6-luna-pro:batch": { - "cache_creation_input_token_cost": 6.25e-8, - "cache_creation_input_token_cost_above_272k_tokens": 1.25e-7, - "cache_read_input_token_cost": 5e-9, - "cache_read_input_token_cost_above_272k_tokens": 1e-8, - "input_cost_per_token": 5e-8, - "input_cost_per_token_above_272k_tokens": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek-v4-pro-0813": { + "cache_read_input_token_cost": 1.3200000000000002e-7, + "input_cost_per_token": 0.00000132, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 393216, + "max_tokens": 393216, "mode": "chat", - "output_cost_per_token": 2.5e-7, - "output_cost_per_token_above_272k_tokens": 3.75e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000396, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-6-luna:batch": { - "cache_creation_input_token_cost": 6.25e-8, - "cache_creation_input_token_cost_above_272k_tokens": 1.25e-7, - "cache_read_input_token_cost": 5e-9, - "cache_read_input_token_cost_above_272k_tokens": 1e-8, - "input_cost_per_token": 5e-8, - "input_cost_per_token_above_272k_tokens": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/deepseek/deepseek_v3": { + "input_cost_per_token": 8.900000000000001e-7, + "litellm_provider": "novita", + "max_input_tokens": 64000, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", - "output_cost_per_token": 2.5e-7, - "output_cost_per_token_above_272k_tokens": 3.75e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 8.900000000000001e-7, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "novita/google/gemma-3-12b-it": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 5e-8, + "output_cost_per_token": 1e-7, + "max_input_tokens": 131072, + "max_output_tokens": 8192, + "max_tokens": 8192, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/openai/gpt-6-sol": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_272k_tokens": 4e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/google/gemma-3-27b-it": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 1.19e-7, + "output_cost_per_token": 2e-7, + "max_input_tokens": 98304, + "max_output_tokens": 16384, + "max_tokens": 16384, + "supports_vision": true, + "supports_system_messages": true + }, + "novita/google/gemma-4-26b-a4b-it": { + "input_cost_per_token": 1.3e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 4.0000000000000003e-7, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/openai/gpt-6-sol-pro": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_272k_tokens": 4e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/google/gemma-4-31b-it": { + "input_cost_per_token": 1.4e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 4.0000000000000003e-7, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/openai/gpt-6-sol-pro:batch": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_272k_tokens": 0.0000025, - "cache_read_input_token_cost": 1e-7, - "cache_read_input_token_cost_above_272k_tokens": 2e-7, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_272k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/gryphe/mythomax-l2-13b": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000005, - "output_cost_per_token_above_272k_tokens": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 9e-8, + "output_cost_per_token": 9e-8, + "max_input_tokens": 4096, + "max_output_tokens": 3200, + "max_tokens": 3200, + "supports_system_messages": true + }, + "novita/inclusionai/ling-3.0-flash": { + "cache_read_input_token_cost": 1.2e-8, + "input_cost_per_token": 6e-8, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 32768, + "max_tokens": 32768, + "mode": "chat", + "output_cost_per_token": 1.8e-7, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-6-sol:batch": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_272k_tokens": 0.0000025, - "cache_read_input_token_cost": 1e-7, - "cache_read_input_token_cost_above_272k_tokens": 2e-7, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_272k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/inclusionai/ling-3.0-flash-fast": { + "cache_read_input_token_cost": 1.2e-8, + "input_cost_per_token": 6e-8, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", - "output_cost_per_token": 0.000005, - "output_cost_per_token_above_272k_tokens": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 1.8e-7, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-6.1-sol": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 1e-7, - "cache_read_input_token_cost_above_272k_tokens": 2e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, + "novita/kwaipilot/kat-coder-pro": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000012, + "max_input_tokens": 256000, "max_output_tokens": 128000, "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, "supports_response_schema": true, + "cache_read_input_token_cost": 6e-8, + "input_cost_per_token_cache_hit": 6e-8 + }, + "novita/meta-llama/llama-3-70b-instruct": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 5.1e-7, + "output_cost_per_token": 7.4e-7, + "max_input_tokens": 8192, + "max_output_tokens": 8000, + "max_tokens": 8000, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/openai/gpt-6.1-sol-pro": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 1e-7, - "cache_read_input_token_cost_above_272k_tokens": 2e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/meta-llama/llama-3-8b-instruct": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, + "input_cost_per_token": 4e-8, + "output_cost_per_token": 4e-8, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, + "supports_system_messages": true, + "supports_response_schema": true + }, + "novita/meta-llama/llama-3.1-8b-instruct": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 2e-8, + "output_cost_per_token": 5e-8, + "max_input_tokens": 16384, + "max_output_tokens": 16384, + "max_tokens": 16384, + "supports_system_messages": true, + "supports_response_schema": true + }, + "novita/meta-llama/llama-3.2-1b-instruct": { + "input_cost_per_token": 2e-8, + "litellm_provider": "novita", + "max_input_tokens": 131000, + "max_output_tokens": 32000, + "max_tokens": 32000, + "mode": "chat", + "output_cost_per_token": 2e-8, + "source": "https://api.novita.ai/v3/openai/models", "supports_response_schema": true, + "supports_vision": false + }, + "novita/meta-llama/llama-3.2-3b-instruct": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 3e-8, + "output_cost_per_token": 5e-8, + "max_input_tokens": 32768, + "max_output_tokens": 32000, + "max_tokens": 32000, + "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true }, - "openrouter/openai/gpt-6.1-sol-pro:batch": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_272k_tokens": 0.0000025, - "cache_read_input_token_cost": 5e-8, - "cache_read_input_token_cost_above_272k_tokens": 1e-7, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_272k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/meta-llama/llama-3.3-70b-instruct": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000005, - "output_cost_per_token_above_272k_tokens": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 1.35e-7, + "output_cost_per_token": 4e-7, + "max_input_tokens": 12288, + "max_output_tokens": 12288, + "max_tokens": 12288, "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "novita/meta-llama/llama-4-maverick-17b-128e-instruct-fp8": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 2.7e-7, + "output_cost_per_token": 8.5e-7, + "max_input_tokens": 1048576, + "max_output_tokens": 8192, + "max_tokens": 8192, "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/openai/gpt-6.1-sol:batch": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_272k_tokens": 0.0000025, - "cache_read_input_token_cost": 5e-8, - "cache_read_input_token_cost_above_272k_tokens": 1e-7, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_272k_tokens": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/meta-llama/llama-4-scout-17b-16e-instruct": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000005, - "output_cost_per_token_above_272k_tokens": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 1.8e-7, + "output_cost_per_token": 5.9e-7, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, + "supports_vision": true, + "supports_system_messages": true + }, + "novita/microsoft/wizardlm-2-8x22b": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 6.2e-7, + "output_cost_per_token": 6.2e-7, + "max_input_tokens": 65535, + "max_output_tokens": 8000, + "max_tokens": 8000, + "supports_system_messages": true, + "supports_response_schema": true + }, + "novita/mindai/macaron-v1-tall": { + "cache_read_input_token_cost": 8e-8, + "input_cost_per_token": 4.5000000000000003e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 32768, + "max_tokens": 32768, + "mode": "chat", + "output_cost_per_token": 0.0000026, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openai/gpt-audio": { - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "input_cost_per_audio_token": 0.000032, - "output_cost_per_audio_token": 0.000064, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "novita/mindai/macaron-v1-venti": { + "cache_read_input_token_cost": 3e-7, + "input_cost_per_token": 0.0000015, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000045, + "source": "https://novita.ai/pricing", "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": false, - "supports_audio_input": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/openai/gpt-audio-mini": { - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000024, - "input_cost_per_audio_token": 6e-7, - "output_cost_per_audio_token": 0.0000024, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "novita/minimax/minimax-m2": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000012, + "max_input_tokens": 204800, + "max_output_tokens": 131072, + "max_tokens": 131072, "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": false, - "supports_audio_input": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_web_search": false + "supports_system_messages": true, + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token_cache_hit": 3e-8, + "supports_reasoning": true, + "supports_response_schema": true }, - "openrouter/openai/gpt-chat-latest": { - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.00003, - "cache_read_input_token_cost": 5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "novita/minimax/minimax-m2.1": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000012, + "max_input_tokens": 204800, + "max_output_tokens": 131072, + "max_tokens": 131072, "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": false, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_web_search": true + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token_cache_hit": 3e-8 }, - "openrouter/openai/gpt-oss-120b": { - "input_cost_per_token": 3.7e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "novita/minimax/minimax-m2.5": { + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "novita", + "max_input_tokens": 204800, + "max_output_tokens": 131100, + "max_tokens": 131100, "mode": "chat", - "output_cost_per_token": 1.7e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000012, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/openai/gpt-oss-120b:batch": { - "input_cost_per_token": 2.96e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "novita/minimax/minimax-m2.5-highspeed": { + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token": 6e-7, + "litellm_provider": "novita", + "max_input_tokens": 204800, + "max_output_tokens": 131100, + "max_tokens": 131100, "mode": "chat", - "output_cost_per_token": 1.36e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000024, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/openai/gpt-oss-20b": { - "cache_read_input_token_cost": 9e-9, - "input_cost_per_token": 1.8e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "novita/minimax/minimax-m2.7": { + "cache_read_input_token_cost": 6e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "novita", + "max_input_tokens": 204800, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 9e-8, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000012, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/openai/gpt-oss-20b:batch": { - "input_cost_per_token": 2.4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "novita/minimax/minimax-m2.7-highspeed": { + "cache_read_input_token_cost": 6e-8, + "input_cost_per_token": 6e-7, + "litellm_provider": "novita", + "max_input_tokens": 204800, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 1.12e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000024, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/openai/gpt-oss-safeguard-20b": { - "input_cost_per_token": 7.5e-8, - "output_cost_per_token": 3e-7, - "cache_read_input_token_cost": 3.75e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 65536, - "max_tokens": 65536, + "novita/minimax/minimax-m3": { + "cache_read_input_token_cost": 6e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "novita", + "max_input_tokens": 1000000, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000012, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/openai/o1": { - "cache_read_input_token_cost": 0.0000075, - "input_cost_per_token": 0.000015, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/minimaxai/minimax-m1-80k": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.00006, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 5.5e-7, + "output_cost_per_token": 0.0000022, + "max_input_tokens": 1000000, + "max_output_tokens": 40000, + "max_tokens": 40000, "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_reasoning": true, + "supports_response_schema": true }, - "openrouter/openai/o1-pro": { - "input_cost_per_token": 0.00015, - "output_cost_per_token": 0.0006, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/mistralai/mistral-nemo": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_tool_choice": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_web_search": true + "input_cost_per_token": 4e-8, + "output_cost_per_token": 1.7e-7, + "max_input_tokens": 60288, + "max_output_tokens": 16000, + "max_tokens": 16000, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/openai/o3": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/moonshotai/kimi-k2-0905": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "input_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000025, + "max_input_tokens": 262144, + "max_output_tokens": 100352, + "max_tokens": 100352, "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 5e-7, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/openai/o3-mini": { - "input_cost_per_token": 0.0000011, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/moonshotai/kimi-k2-instruct": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.0000044, + "input_cost_per_token": 5.7e-7, + "output_cost_per_token": 0.0000023, + "max_input_tokens": 131072, + "max_output_tokens": 100352, + "max_tokens": 100352, "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": false, - "cache_read_input_token_cost": 5.5e-7, - "supports_prompt_caching": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": true, - "supports_response_schema": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/openai/o3-mini-high": { - "input_cost_per_token": 0.0000011, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/moonshotai/kimi-k2-thinking": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.0000044, + "input_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000025, + "max_input_tokens": 262144, + "max_output_tokens": 100352, + "max_tokens": 100352, "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": false, - "cache_read_input_token_cost": 5.5e-7, - "supports_prompt_caching": true, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_pdf_input": true, + "supports_system_messages": true, "supports_response_schema": true, - "supports_web_search": true + "supports_reasoning": true, + "cache_read_input_token_cost": 1.5e-7, + "supports_prompt_caching": true }, - "openrouter/openai/o3-mini:batch": { - "cache_read_input_token_cost": 2.75e-7, - "input_cost_per_token": 5.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/moonshotai/kimi-k2.5": { + "cache_read_input_token_cost": 1.0000000000000001e-7, + "input_cost_per_token": 6e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.0000022, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000003, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": true + "supports_vision": true }, - "openrouter/openai/o3-pro": { - "input_cost_per_token": 0.00002, - "output_cost_per_token": 0.00008, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/moonshotai/kimi-k2.6": { + "cache_read_input_token_cost": 1.6e-7, + "input_cost_per_token": 8.000000000000001e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000034, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": false, - "supports_web_search": true + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/openai/o3:batch": { - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/moonshotai/kimi-k2.7-code": { + "cache_read_input_token_cost": 1.9e-7, + "input_cost_per_token": 9.499999999999999e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", "output_cost_per_token": 0.000004, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/openai/o4-mini": { - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/moonshotai/kimi-k3": { + "cache_read_input_token_cost": 3e-7, + "input_cost_per_token": 0.000003, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.000015, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 2.75e-7, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/openai/o4-mini-high": { - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 2.75e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/nousresearch/hermes-2-pro-llama-3-8b": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 1.4e-7, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/openai/o4-mini:batch": { - "cache_read_input_token_cost": 1.375e-7, - "input_cost_per_token": 5.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 100000, - "max_tokens": 100000, + "novita/nvidia/nemotron-3-nano-30b-a3b": { + "input_cost_per_token": 5.0000000000000004e-8, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", - "output_cost_per_token": 0.0000022, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 2.0000000000000002e-7, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/openrouter/auto": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 2000000, - "max_tokens": 2000000, + "novita/openai/gpt-oss-120b": { + "litellm_provider": "novita", "mode": "chat", + "input_cost_per_token": 5e-8, + "output_cost_per_token": 2.5e-7, + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, + "supports_vision": false, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": true, - "supports_audio_input": true, - "supports_video_input": true - }, - "openrouter/openrouter/bodybuilder": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_tokens": 128000, - "mode": "chat" + "supports_reasoning": true }, - "openrouter/openrouter/free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_tokens": 200000, + "novita/openai/gpt-oss-20b": { + "litellm_provider": "novita", "mode": "chat", - "supports_function_calling": true, + "input_cost_per_token": 4e-8, + "output_cost_per_token": 1.5e-7, + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, "supports_tool_choice": true, - "supports_reasoning": true, + "supports_vision": false, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": true + "supports_reasoning": true }, - "openrouter/perceptron/perceptron-mk1": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "max_tokens": 8192, + "novita/paddlepaddle/paddleocr-vl": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": false, + "input_cost_per_token": 2e-8, + "output_cost_per_token": 2e-8, + "max_input_tokens": 16384, + "max_output_tokens": 16384, + "max_tokens": 16384, "supports_vision": true, - "supports_web_search": false + "supports_system_messages": true }, - "openrouter/perceptron/perceptron-mk1.5": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 36864, + "novita/qwen/qwen-2.5-72b-instruct": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 3.8e-7, + "output_cost_per_token": 4e-7, + "max_input_tokens": 32000, + "max_output_tokens": 8192, + "max_tokens": 8192, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "novita/qwen/qwen-mt-plus": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 7.5e-7, + "max_input_tokens": 16384, "max_output_tokens": 8192, "max_tokens": 8192, + "supports_system_messages": true + }, + "novita/qwen/qwen2.5-7b-instruct": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.0000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "input_cost_per_token": 7e-8, + "output_cost_per_token": 7e-8, + "max_input_tokens": 32000, + "max_output_tokens": 8192, + "max_tokens": 8192, "supports_function_calling": true, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/perplexity/sonar": { - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 127072, - "max_output_tokens": 114364, - "max_tokens": 114364, + "novita/qwen/qwen2.5-vl-72b-instruct": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, + "input_cost_per_token": 8e-7, + "output_cost_per_token": 8e-7, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "max_tokens": 32768, "supports_vision": true, - "supports_web_search": true + "supports_system_messages": true }, - "openrouter/perplexity/sonar-deep-research": { - "input_cost_per_token": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 115200, - "max_tokens": 115200, + "novita/qwen/qwen3-235b-a22b-fp8": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000008, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "input_cost_per_token": 2e-7, + "output_cost_per_token": 8e-7, + "max_input_tokens": 40960, + "max_output_tokens": 20000, + "max_tokens": 20000, + "supports_system_messages": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": true + "supports_response_schema": true }, - "openrouter/perplexity/sonar-pro": { - "input_cost_per_token": 0.000003, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 8000, - "max_tokens": 8000, + "novita/qwen/qwen3-235b-a22b-instruct-2507": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": true + "input_cost_per_token": 9e-8, + "output_cost_per_token": 5.8e-7, + "max_input_tokens": 131072, + "max_output_tokens": 16384, + "max_tokens": 16384, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/perplexity/sonar-pro-search": { - "input_cost_per_token": 0.000003, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 8000, - "max_tokens": 8000, + "novita/qwen/qwen3-235b-a22b-thinking-2507": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.000003, + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_reasoning": true + }, + "novita/qwen/qwen3-30b-a3b-fp8": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 9e-8, + "output_cost_per_token": 4.5e-7, + "max_input_tokens": 40960, + "max_output_tokens": 20000, + "max_tokens": 20000, + "supports_system_messages": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": true + "supports_function_calling": true, + "supports_tool_choice": true }, - "openrouter/perplexity/sonar-reasoning-pro": { - "input_cost_per_token": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 115200, - "max_tokens": 115200, + "novita/qwen/qwen3-32b-fp8": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000008, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "input_cost_per_token": 1e-7, + "output_cost_per_token": 4.5e-7, + "max_input_tokens": 40960, + "max_output_tokens": 20000, + "max_tokens": 20000, + "supports_system_messages": true, + "supports_reasoning": true + }, + "novita/qwen/qwen3-4b-fp8": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 3e-8, + "output_cost_per_token": 3e-8, + "max_input_tokens": 128000, + "max_output_tokens": 8192, + "max_tokens": 8192, + "supports_system_messages": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": true + "supports_function_calling": true, + "supports_tool_choice": true }, - "openrouter/poolside/laguna-s-2.1": { - "input_cost_per_token": 9e-8, - "output_cost_per_token": 1.8e-7, - "cache_read_input_token_cost": 9e-9, - "deprecation_date": "2026-10-31", - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, + "novita/qwen/qwen3-8b-fp8": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 3.5e-8, + "output_cost_per_token": 1.38e-7, + "max_input_tokens": 128000, + "max_output_tokens": 20000, + "max_tokens": 20000, + "supports_system_messages": true, + "supports_reasoning": true + }, + "novita/qwen/qwen3-coder-30b-a3b-instruct": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 7e-8, + "output_cost_per_token": 2.7e-7, + "max_input_tokens": 160000, + "max_output_tokens": 32768, + "max_tokens": 32768, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_response_schema": false, - "supports_web_search": false + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/poolside/laguna-s-2.1:free": { - "deprecation_date": "2026-10-31", - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", + "novita/qwen/qwen3-coder-480b-a35b-instruct": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 3.8e-7, + "output_cost_per_token": 0.00000155, + "max_input_tokens": 262144, + "max_output_tokens": 65536, + "max_tokens": 65536, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "novita/qwen/qwen3-coder-next": { + "input_cost_per_token": 2.0000000000000002e-7, + "litellm_provider": "novita", "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000015, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_response_schema": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/poolside/laguna-xs-2.1": { - "input_cost_per_token": 6e-8, - "output_cost_per_token": 1.2e-7, - "cache_read_input_token_cost": 3e-8, - "deprecation_date": "2026-10-31", - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "novita/qwen/qwen3-max": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 0.00000211, + "output_cost_per_token": 0.00000845, + "max_input_tokens": 262144, + "max_output_tokens": 65536, + "max_tokens": 65536, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_response_schema": false, - "supports_web_search": false + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/poolside/laguna-xs-2.1:free": { - "deprecation_date": "2026-10-31", - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, + "novita/qwen/qwen3-next-80b-a3b-instruct": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 0.0000015, + "max_input_tokens": 131072, "max_output_tokens": 32768, "max_tokens": 32768, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_vision": false, - "supports_web_search": false + "supports_system_messages": true, + "supports_response_schema": true, + "supports_reasoning": true }, - "openrouter/prism-ml/ternary-bonsai-2-27b": { - "cache_read_input_token_cost": 3.75e-8, - "input_cost_per_token": 7.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, + "novita/qwen/qwen3-next-80b-a3b-thinking": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 0.0000015, + "max_input_tokens": 131072, "max_output_tokens": 32768, "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_system_messages": true, + "supports_response_schema": true, + "supports_reasoning": true }, - "openrouter/qwen/qwen-2.5-72b-instruct": { - "input_cost_per_token": 3.6e-7, - "output_cost_per_token": 4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, + "novita/qwen/qwen3-omni-30b-a3b-instruct": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 9.7e-7, + "max_input_tokens": 65536, "max_output_tokens": 16384, "max_tokens": 16384, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, + "supports_vision": true, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_audio_input": true, + "supports_audio_output": true }, - "openrouter/qwen/qwen-2.5-7b-instruct": { - "input_cost_per_token": 1e-7, - "output_cost_per_token": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 29491, - "max_tokens": 29491, + "novita/qwen/qwen3-omni-30b-a3b-thinking": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 9.7e-7, + "max_input_tokens": 65536, + "max_output_tokens": 16384, + "max_tokens": 16384, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, + "supports_vision": true, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_reasoning": true, + "supports_audio_input": true }, - "openrouter/qwen/qwen-2.5-coder-32b-instruct": { - "input_cost_per_token": 6.6e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 29491, - "max_tokens": 29491, + "novita/qwen/qwen3-vl-235b-a22b-instruct": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.000001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/qwen/qwen-plus": { - "input_cost_per_token": 2.6e-7, - "output_cost_per_token": 7.8e-7, - "cache_read_input_token_cost": 5.2e-8, - "cache_creation_input_token_cost": 3.25e-7, - "input_cost_per_token_above_256k_tokens": 7.8e-7, - "output_cost_per_token_above_256k_tokens": 0.00000234, - "cache_read_input_token_cost_above_256k_tokens": 1.56e-7, - "cache_creation_input_token_cost_above_256k_tokens": 9.75e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000015, + "max_input_tokens": 131072, "max_output_tokens": 32768, "max_tokens": 32768, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false + "supports_vision": true, + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/qwen/qwen-plus-2025-07-28": { - "cache_creation_input_token_cost": 3.25e-7, - "cache_read_input_token_cost": 5.2e-8, - "deprecation_date": "2026-10-09", - "input_cost_per_token": 2.6e-7, - "input_cost_per_token_above_256k_tokens": 7.8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, + "novita/qwen/qwen3-vl-235b-a22b-thinking": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 9.8e-7, + "output_cost_per_token": 0.00000395, + "max_input_tokens": 131072, "max_output_tokens": 32768, "max_tokens": 32768, + "supports_vision": true, + "supports_system_messages": true, + "supports_reasoning": true, + "supports_function_calling": true, + "supports_tool_choice": true + }, + "novita/qwen/qwen3-vl-30b-a3b-instruct": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 7.8e-7, - "output_cost_per_token_above_256k_tokens": 0.00000234, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 2e-7, + "output_cost_per_token": 7e-7, + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/qwen/qwen-vl-plus": { - "input_cost_per_token": 2.1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 8192, - "max_output_tokens": 2048, - "max_tokens": 2048, + "novita/qwen/qwen3-vl-30b-a3b-thinking": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 6.3e-7, + "input_cost_per_token": 2e-7, + "output_cost_per_token": 0.000001, + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, + "supports_function_calling": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/qwen/qwen2.5-vl-72b-instruct": { - "input_cost_per_token": 8e-7, - "output_cost_per_token": 0.000001, - "cache_read_input_token_cost": 4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 128000, - "max_output_tokens": 115200, - "max_tokens": 115200, + "novita/qwen/qwen3-vl-8b-instruct": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_tool_choice": false, - "supports_response_schema": true, + "input_cost_per_token": 8e-8, + "output_cost_per_token": 5e-7, + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, "supports_vision": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false + "supports_system_messages": true, + "supports_response_schema": true }, - "openrouter/qwen/qwen3-14b": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 1.2e-7, - "output_cost_per_token": 2.4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "novita/qwen/qwen3.5-122b-a10b": { + "input_cost_per_token": 4.0000000000000003e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000032000000000000003, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/qwen/qwen3-235b-a22b": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 4.55e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 8192, - "max_tokens": 8192, + "novita/qwen/qwen3.5-27b": { + "input_cost_per_token": 3e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 0.00000182, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000024, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/qwen/qwen3-235b-a22b-2507": { - "input_cost_per_token": 9e-8, - "litellm_provider": "openrouter", + "novita/qwen/qwen3.5-35b-a3b": { + "input_cost_per_token": 2.5e-7, + "litellm_provider": "novita", "max_input_tokens": 262144, - "max_output_tokens": 16384, - "max_tokens": 16384, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 5.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000002, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/qwen/qwen3-235b-a22b-thinking-2507": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 2.3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "novita/qwen/qwen3.5-397b-a17b": { + "input_cost_per_token": 6e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 0.0000023, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000036000000000000003, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/qwen/qwen3-30b-a3b": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 1.2e-7, - "output_cost_per_token": 5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "novita/qwen/qwen3.6-27b": { + "input_cost_per_token": 6e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000036000000000000003, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/qwen/qwen3-30b-a3b-instruct-2507": { - "input_cost_per_token": 4.815e-8, - "litellm_provider": "openrouter", + "novita/qwen/qwen3.6-35b-a3b": { + "input_cost_per_token": 2.48e-7, + "litellm_provider": "novita", "max_input_tokens": 262144, - "max_output_tokens": 32000, - "max_tokens": 32000, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 1.9305e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000014850000000000002, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/qwen/qwen3-30b-a3b-thinking-2507": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 81920, - "max_output_tokens": 32768, - "max_tokens": 32768, + "novita/qwen/qwen3.7-max": { + "cache_read_input_token_cost": 2.5e-7, + "input_cost_per_token": 0.00000125, + "litellm_provider": "novita", + "max_input_tokens": 1000000, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 0.0000024, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000375, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3-32b": { - "input_cost_per_token": 8e-8, - "output_cost_per_token": 2.8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "novita/qwen/qwen3.8-max": { + "cache_read_input_token_cost": 2.5e-7, + "input_cost_per_token": 0.000002, + "litellm_provider": "novita", + "max_input_tokens": 1000000, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000006, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/qwen/qwen3-8b": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 1.17e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, + "novita/sao10k/l3-70b-euryale-v2.1": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 0.00000148, + "output_cost_per_token": 0.00000148, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true + }, + "novita/sao10k/l3-8b-lunaris": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 5e-8, + "output_cost_per_token": 5e-8, + "max_input_tokens": 8192, "max_output_tokens": 8192, "max_tokens": 8192, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "novita/sao10k/l31-70b-euryale-v2.2": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 4.55e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 0.00000148, + "output_cost_per_token": 0.00000148, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_system_messages": true }, - "openrouter/qwen/qwen3-coder": { - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", + "novita/skywork/r1v4-lite": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 2e-7, + "output_cost_per_token": 6e-7, "max_input_tokens": 262144, "max_output_tokens": 65536, "max_tokens": 65536, - "mode": "chat", - "output_cost_per_token": 0.000001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_tool_choice": true, + "supports_vision": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "novita/stepfun/step-3.7-flash": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 2.0000000000000002e-7, + "litellm_provider": "novita", + "max_input_tokens": 262144, + "max_output_tokens": 256000, + "max_tokens": 256000, + "mode": "chat", + "output_cost_per_token": 0.00000115, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, - "supports_reasoning": false, + "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": true }, - "openrouter/qwen/qwen3-coder-30b-a3b-instruct": { - "input_cost_per_token": 7e-8, - "output_cost_per_token": 2.8e-7, - "litellm_provider": "openrouter", + "novita/tencent/hy3": { + "cache_read_input_token_cost": 3.5e-8, + "input_cost_per_token": 1.4e-7, + "litellm_provider": "novita", "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 5.8e-7, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_tool_choice": true, + "supports_prompt_caching": true, + "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": false }, - "openrouter/qwen/qwen3-coder-flash": { - "input_cost_per_token": 1.95e-7, - "input_cost_per_token_above_32k_tokens": 3.25e-7, - "cache_creation_input_token_cost_above_32k_tokens": 4.0625e-7, - "cache_read_input_token_cost_above_32k_tokens": 6.5e-8, - "output_cost_per_token_above_32k_tokens": 0.000001625, - "output_cost_per_token": 9.75e-7, - "cache_read_input_token_cost": 3.9e-8, - "cache_creation_input_token_cost": 2.4375e-7, - "input_cost_per_token_above_128k_tokens": 5.2e-7, - "output_cost_per_token_above_128k_tokens": 0.0000026, - "cache_read_input_token_cost_above_128k_tokens": 1.04e-7, - "cache_creation_input_token_cost_above_128k_tokens": 6.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "novita/thudm/glm-4-32b-0414": { + "input_cost_per_token": 5.5e-7, + "litellm_provider": "novita", + "max_input_tokens": 32000, + "max_output_tokens": 32000, + "max_tokens": 32000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000166, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": false }, - "openrouter/qwen/qwen3-coder-next": { - "input_cost_per_token": 1.2e-7, - "output_cost_per_token": 8e-7, - "cache_read_input_token_cost": 7e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "novita/xiaomimimo/mimo-v2-flash": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 1.1e-7, + "output_cost_per_token": 3.3e-7, + "max_input_tokens": 262144, + "max_output_tokens": 32000, + "max_tokens": 32000, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false + "cache_read_input_token_cost": 2.4e-8, + "input_cost_per_token_cache_hit": 2.4e-8, + "supports_reasoning": true }, - "openrouter/qwen/qwen3-coder-plus": { - "cache_creation_input_token_cost": 8.125e-7, - "cache_creation_input_token_cost_above_128k_tokens": 0.0000024375, - "cache_creation_input_token_cost_above_32k_tokens": 0.0000014625, - "cache_read_input_token_cost": 1.3e-7, - "cache_read_input_token_cost_above_128k_tokens": 3.9e-7, - "cache_read_input_token_cost_above_32k_tokens": 2.34e-7, - "deprecation_date": "2026-10-09", - "input_cost_per_token": 6.5e-7, - "input_cost_per_token_above_128k_tokens": 0.00000195, - "input_cost_per_token_above_32k_tokens": 0.00000117, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "novita/xiaomimimo/mimo-v2.5": { + "cache_read_input_token_cost": 3.4e-9, + "input_cost_per_token": 1.6800000000000002e-7, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.00000325, - "output_cost_per_token_above_128k_tokens": 0.00000975, - "output_cost_per_token_above_32k_tokens": 0.00000585, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 3.3600000000000004e-7, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, - "supports_reasoning": false, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/qwen/qwen3-max": { - "cache_creation_input_token_cost": 9.75e-7, - "cache_creation_input_token_cost_above_128k_tokens": 0.0000024375, - "cache_creation_input_token_cost_above_32k_tokens": 0.00000195, - "cache_read_input_token_cost": 1.56e-7, - "cache_read_input_token_cost_above_128k_tokens": 3.9e-7, - "cache_read_input_token_cost_above_32k_tokens": 3.12e-7, - "deprecation_date": "2026-10-09", - "input_cost_per_token": 7.8e-7, - "input_cost_per_token_above_128k_tokens": 0.00000195, - "input_cost_per_token_above_32k_tokens": 0.00000156, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, + "novita/xiaomimimo/mimo-v2.5-pro": { + "cache_read_input_token_cost": 4.3e-9, + "input_cost_per_token": 5.22e-7, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.0000039, - "output_cost_per_token_above_128k_tokens": 0.00000975, - "output_cost_per_token_above_32k_tokens": 0.0000078, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000001044, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, - "supports_reasoning": false, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3-max-thinking": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 7.8e-7, - "input_cost_per_token_above_128k_tokens": 0.00000195, - "input_cost_per_token_above_32k_tokens": 0.00000156, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, + "novita/zai-org/autoglm-phone-9b-multilingual": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 3.5e-8, + "output_cost_per_token": 1.38e-7, + "max_input_tokens": 65536, "max_output_tokens": 65536, "max_tokens": 65536, + "supports_vision": true, + "supports_system_messages": true + }, + "novita/zai-org/glm-4.5": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 0.0000039, - "output_cost_per_token_above_128k_tokens": 0.00000975, - "output_cost_per_token_above_32k_tokens": 0.0000078, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000022, + "max_input_tokens": 131072, + "max_output_tokens": 98304, + "max_tokens": 98304, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_system_messages": true, + "cache_read_input_token_cost": 1.1e-7, + "input_cost_per_token_cache_hit": 1.1e-7, + "supports_reasoning": true }, - "openrouter/qwen/qwen3-next-80b-a3b-instruct": { - "input_cost_per_token": 1e-7, - "output_cost_per_token": 0.0000011, - "cache_read_input_token_cost": 7e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "novita/zai-org/glm-4.5-air": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 1.3e-7, + "output_cost_per_token": 8.5e-7, + "max_input_tokens": 131072, + "max_output_tokens": 98304, + "max_tokens": 98304, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_web_search": false + "supports_system_messages": true, + "supports_reasoning": true, + "cache_read_input_token_cost": 2.5e-8, + "supports_prompt_caching": true }, - "openrouter/qwen/qwen3-next-80b-a3b-thinking": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 0.0000012, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "novita/zai-org/glm-4.5v": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000018, + "max_input_tokens": 65536, + "max_output_tokens": 16384, + "max_tokens": 16384, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, + "supports_vision": true, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "cache_read_input_token_cost": 1.1e-7, + "input_cost_per_token_cache_hit": 1.1e-7, + "supports_reasoning": true }, - "openrouter/qwen/qwen3-vl-235b-a22b-instruct": { - "input_cost_per_token": 2.1e-7, - "output_cost_per_token": 0.0000019, - "cache_read_input_token_cost": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "novita/zai-org/glm-4.6": { + "litellm_provider": "novita", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 5.5e-7, + "output_cost_per_token": 0.0000022, + "max_input_tokens": 204800, + "max_output_tokens": 131072, + "max_tokens": 131072, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_parallel_function_calling": true, "supports_tool_choice": true, + "supports_system_messages": true, "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_web_search": false + "cache_read_input_token_cost": 1.1e-7, + "input_cost_per_token_cache_hit": 1.1e-7, + "supports_reasoning": true }, - "openrouter/qwen/qwen3-vl-235b-a22b-thinking": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 4e-7, - "litellm_provider": "openrouter", + "novita/zai-org/glm-4.6v": { + "litellm_provider": "novita", + "mode": "chat", + "input_cost_per_token": 3e-7, + "output_cost_per_token": 9e-7, "max_input_tokens": 131072, "max_output_tokens": 32768, "max_tokens": 32768, - "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": false + "supports_system_messages": true, + "supports_response_schema": true, + "cache_read_input_token_cost": 5.5e-8, + "input_cost_per_token_cache_hit": 5.5e-8, + "supports_reasoning": true }, - "openrouter/qwen/qwen3-vl-30b-a3b-instruct": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 16384, - "max_tokens": 16384, + "novita/zai-org/glm-4.7": { + "litellm_provider": "novita", "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "input_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000022, + "max_input_tokens": 204800, + "max_output_tokens": 131072, + "max_tokens": 131072, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, + "supports_parallel_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_system_messages": true, + "supports_response_schema": true, + "cache_read_input_token_cost": 1.1e-7, + "input_cost_per_token_cache_hit": 1.1e-7, + "supports_reasoning": true }, - "openrouter/qwen/qwen3-vl-30b-a3b-thinking": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "novita/zai-org/glm-4.7-flash": { + "cache_read_input_token_cost": 1e-8, + "input_cost_per_token": 7e-8, + "litellm_provider": "novita", + "max_input_tokens": 200000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.0000024, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 4.0000000000000003e-7, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3-vl-32b-instruct": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 1.04e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "novita/zai-org/glm-4.7-h": { + "cache_read_input_token_cost": 1.1e-7, + "input_cost_per_token": 6e-7, + "litellm_provider": "novita", + "max_input_tokens": 204800, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 4.16e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000022, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_prompt_caching": true, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3-vl-8b-instruct": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 1.17e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "novita/zai-org/glm-5": { + "cache_read_input_token_cost": 2.0000000000000002e-7, + "input_cost_per_token": 0.000001, + "litellm_provider": "novita", + "max_input_tokens": 202800, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 4.55e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000032000000000000003, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_prompt_caching": true, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3-vl-8b-thinking": { - "deprecation_date": "2026-10-09", - "input_cost_per_token": 1.8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 32768, - "max_tokens": 32768, + "novita/zai-org/glm-5-turbo": { + "cache_read_input_token_cost": 2.4e-7, + "input_cost_per_token": 0.0000012, + "litellm_provider": "novita", + "max_input_tokens": 202800, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.0000021, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000004, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3.5-122b-a10b": { - "input_cost_per_token": 2.6e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "novita/zai-org/glm-5.1": { + "cache_read_input_token_cost": 2.6e-7, + "input_cost_per_token": 0.00000138, + "litellm_provider": "novita", + "max_input_tokens": 204800, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.00000208, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000044, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3.5-27b": { - "input_cost_per_token": 1.95e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, + "novita/zai-org/glm-5.2": { + "cache_read_input_token_cost": 2.6e-7, + "input_cost_per_token": 0.0000014, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.00000156, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000044, + "source": "https://novita.ai/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3.5-35b-a3b": { - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "novita/zai-org/glm-5.3": { + "cache_read_input_token_cost": 2.6e-7, + "input_cost_per_token": 0.0000014, + "litellm_provider": "novita", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.000001, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000044, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true, - "cache_read_input_token_cost": 5e-8, - "supports_audio_input": false, - "supports_pdf_input": false, "supports_prompt_caching": true, + "supports_reasoning": true, "supports_response_schema": true, - "supports_web_search": false + "supports_tool_choice": true, + "supports_vision": false }, - "openrouter/qwen/qwen3.5-397b-a17b": { - "cache_read_input_token_cost": 2.2e-7, - "input_cost_per_token": 4.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 81920, - "max_tokens": 81920, + "novita/zai-org/glm-5v-turbo": { + "cache_read_input_token_cost": 2.4e-7, + "input_cost_per_token": 0.0000012, + "litellm_provider": "novita", + "max_input_tokens": 204800, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.000003, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000004, + "source": "https://api.novita.ai/v3/openai/models", "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/qwen/qwen3.5-9b": { - "input_cost_per_token": 1e-7, + "nscale/Qwen/QwQ-32B": { + "input_cost_per_token": 1.8e-7, + "litellm_provider": "nscale", + "mode": "chat", + "output_cost_per_token": 2e-7, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nscale/Qwen/Qwen2.5-Coder-32B-Instruct": { + "input_cost_per_token": 6e-8, + "litellm_provider": "nscale", + "mode": "chat", + "output_cost_per_token": 2e-7, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nscale/Qwen/Qwen2.5-Coder-3B-Instruct": { + "input_cost_per_token": 1e-8, + "litellm_provider": "nscale", + "mode": "chat", + "output_cost_per_token": 3e-8, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nscale/Qwen/Qwen2.5-Coder-7B-Instruct": { + "input_cost_per_token": 1e-8, + "litellm_provider": "nscale", + "mode": "chat", + "output_cost_per_token": 3e-8, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-70B": { + "input_cost_per_token": 3.75e-7, + "litellm_provider": "nscale", + "metadata": { + "notes": "Pricing listed as $0.75/1M tokens total. Assumed 50/50 split for input/output." + }, + "mode": "chat", + "output_cost_per_token": 3.75e-7, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-8B": { + "input_cost_per_token": 2.5e-8, + "litellm_provider": "nscale", + "metadata": { + "notes": "Pricing listed as $0.05/1M tokens total. Assumed 50/50 split for input/output." + }, + "mode": "chat", + "output_cost_per_token": 2.5e-8, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": { + "input_cost_per_token": 9e-8, + "litellm_provider": "nscale", + "metadata": { + "notes": "Pricing listed as $0.18/1M tokens total. Assumed 50/50 split for input/output." + }, + "mode": "chat", + "output_cost_per_token": 9e-8, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": { + "input_cost_per_token": 7e-8, + "litellm_provider": "nscale", + "metadata": { + "notes": "Pricing listed as $0.14/1M tokens total. Assumed 50/50 split for input/output." + }, + "mode": "chat", + "output_cost_per_token": 7e-8, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": { + "input_cost_per_token": 1.5e-7, + "litellm_provider": "nscale", + "metadata": { + "notes": "Pricing listed as $0.30/1M tokens total. Assumed 50/50 split for input/output." + }, + "mode": "chat", "output_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 32768, - "max_tokens": 32768, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": { + "input_cost_per_token": 2e-7, + "litellm_provider": "nscale", + "metadata": { + "notes": "Pricing listed as $0.40/1M tokens total. Assumed 50/50 split for input/output." + }, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 2e-7, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, - "openrouter/qwen/qwen3.5-flash-02-23": { - "input_cost_per_token": 6.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "nscale/meta-llama/Llama-3.1-8B-Instruct": { + "input_cost_per_token": 3e-8, + "litellm_provider": "nscale", + "metadata": { + "notes": "Pricing listed as $0.06/1M tokens total. Assumed 50/50 split for input/output." + }, "mode": "chat", - "output_cost_per_token": 2.6e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 3e-8, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, - "openrouter/qwen/qwen3.5-plus-02-15": { - "input_cost_per_token": 2.6e-7, - "input_cost_per_token_above_256k_tokens": 3.25e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "nscale/meta-llama/Llama-3.3-70B-Instruct": { + "input_cost_per_token": 2e-7, + "litellm_provider": "nscale", + "metadata": { + "notes": "Pricing listed as $0.40/1M tokens total. Assumed 50/50 split for input/output." + }, "mode": "chat", - "output_cost_per_token": 0.00000156, - "output_cost_per_token_above_256k_tokens": 0.00000195, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 2e-7, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, - "openrouter/qwen/qwen3.5-plus-20260420": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 0.0000018, - "cache_creation_input_token_cost": 3.75e-7, - "input_cost_per_token_above_256k_tokens": 3.75e-7, - "output_cost_per_token_above_256k_tokens": 0.00000225, - "cache_creation_input_token_cost_above_256k_tokens": 4.6875e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "nscale/meta-llama/Llama-4-Scout-17B-16E-Instruct": { + "input_cost_per_token": 9e-8, + "litellm_provider": "nscale", "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 2.9e-7, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, - "openrouter/qwen/qwen3.6-27b": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, + "nscale/mistralai/mixtral-8x22b-instruct-v0.1": { + "input_cost_per_token": 6e-7, + "litellm_provider": "nscale", + "metadata": { + "notes": "Pricing listed as $1.20/1M tokens total. Assumed 50/50 split for input/output." + }, + "mode": "chat", + "output_cost_per_token": 6e-7, + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" + }, + "nvidia.nemotron-nano-12b-v2": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 2e-7, + "input_cost_per_token_batches": 1e-7, + "input_cost_per_token_flex": 1e-7, + "input_cost_per_token_priority": 3.5e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 6e-7, + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_flex": 3e-7, + "output_cost_per_token_priority": 0.00000105, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_system_messages": true, + "supports_vision": true }, - "openrouter/qwen/qwen3.6-35b-a3b": { - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 0.000001, - "cache_read_input_token_cost": 5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "nvidia.nemotron-nano-3-30b": { + "input_cost_per_token": 6e-8, + "input_cost_per_token_batches": 3e-8, + "input_cost_per_token_flex": 3e-8, + "input_cost_per_token_priority": 1.1e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 256000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 2.4e-7, + "output_cost_per_token_batches": 1.2e-7, + "output_cost_per_token_flex": 1.2e-7, + "output_cost_per_token_priority": 4.2e-7, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_reasoning": true, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", + "supports_audio_input": false, + "supports_native_structured_output": true, "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3.6-flash": { - "input_cost_per_token": 1.875e-7, - "output_cost_per_token": 0.000001125, - "cache_creation_input_token_cost": 2.34375e-7, - "input_cost_per_token_above_256k_tokens": 7.5e-7, - "output_cost_per_token_above_256k_tokens": 0.000003, - "cache_creation_input_token_cost_above_256k_tokens": 9.375e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "nvidia.nemotron-nano-9b-v2": { + "deprecation_date": "2027-03-30", + "input_cost_per_token": 6e-8, + "input_cost_per_token_batches": 3e-8, + "input_cost_per_token_flex": 3e-8, + "input_cost_per_token_priority": 1.1e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 8000, + "max_tokens": 8000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 2.3e-7, + "output_cost_per_token_batches": 1.2e-7, + "output_cost_per_token_flex": 1.2e-7, + "output_cost_per_token_priority": 4e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "supports_web_search": false + "supports_system_messages": true, + "supports_vision": false }, - "openrouter/qwen/qwen3.6-max-preview": { - "cache_creation_input_token_cost": 0.00000128375, - "cache_creation_input_token_cost_above_128k_tokens": 0.000001975, - "deprecation_date": "2026-10-09", - "input_cost_per_token": 0.000001027, - "input_cost_per_token_above_128k_tokens": 0.00000158, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, + "nvidia.nemotron-super-3-120b": { + "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7.5e-8, + "input_cost_per_token_flex": 7.5e-8, + "input_cost_per_token_priority": 2.625e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 256000, + "max_output_tokens": 32000, + "max_tokens": 32000, "mode": "chat", - "output_cost_per_token": 0.000006162, - "output_cost_per_token_above_128k_tokens": 0.00000948, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 6.5e-7, + "output_cost_per_token_batches": 3.25e-7, + "output_cost_per_token_flex": 3.25e-7, + "output_cost_per_token_priority": 0.0000011375, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, "supports_reasoning": true, "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3.6-plus": { - "cache_creation_input_token_cost": 4.0625e-7, - "input_cost_per_token_above_256k_tokens": 0.0000013, - "cache_creation_input_token_cost_above_256k_tokens": 0.000001625, - "output_cost_per_token_above_256k_tokens": 0.0000039, - "input_cost_per_token": 3.25e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "o1": { + "cache_read_input_token_cost": 0.0000075, + "deprecation_date": "2026-10-23", + "input_cost_per_token": 0.000015, + "input_cost_per_token_batches": 0.0000075, + "litellm_provider": "openai", + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "max_tokens": 100000, "mode": "chat", - "output_cost_per_token": 0.00000195, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00006, + "output_cost_per_token_batches": 0.00003, + "source": "https://developers.openai.com/api/docs/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_parallel_function_calling": false, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/qwen/qwen3.7-flash": { - "input_cost_per_token": 3e-8, - "input_cost_per_token_above_32k_tokens": 1e-7, - "cache_creation_input_token_cost_above_32k_tokens": 1.25e-7, - "cache_read_input_token_cost_above_32k_tokens": 2e-8, - "output_cost_per_token_above_32k_tokens": 4e-7, - "output_cost_per_token": 1.3e-7, - "cache_read_input_token_cost": 6e-9, - "cache_creation_input_token_cost": 3.8e-8, - "input_cost_per_token_above_256k_tokens": 2e-7, - "output_cost_per_token_above_256k_tokens": 8e-7, - "cache_read_input_token_cost_above_256k_tokens": 4e-8, - "cache_creation_input_token_cost_above_256k_tokens": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 65536, - "max_tokens": 65536, + "o1-2024-12-17": { + "cache_read_input_token_cost": 0.0000075, + "deprecation_date": "2026-10-23", + "input_cost_per_token": 0.000015, + "input_cost_per_token_batches": 0.0000075, + "litellm_provider": "openai", + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "max_tokens": 100000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00006, + "output_cost_per_token_batches": 0.00003, + "source": "https://developers.openai.com/api/docs/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, "supports_prompt_caching": true, - "supports_web_search": false - }, - "openrouter/qwen/qwen3.7-max": { - "input_cost_per_token": 0.000001475, - "output_cost_per_token": 0.000004425, - "cache_read_input_token_cost": 2.95e-7, - "cache_creation_input_token_cost": 0.00000184375, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_web_search": false - }, - "openrouter/qwen/qwen3.7-plus": { - "input_cost_per_token": 3.2e-7, - "input_cost_per_token_above_256k_tokens": 9.6e-7, - "cache_creation_input_token_cost_above_256k_tokens": 0.0000012, - "cache_read_input_token_cost_above_256k_tokens": 1.92e-7, - "output_cost_per_token_above_256k_tokens": 0.00000384, - "output_cost_per_token": 0.00000128, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "cache_read_input_token_cost": 6.4e-8, - "supports_prompt_caching": true, - "cache_creation_input_token_cost": 4e-7, - "supports_audio_input": false, - "supports_pdf_input": false, - "supports_web_search": false + "supports_vision": true }, - "openrouter/qwen/qwen3.8-2.4t-a95b": { + "o3": { + "cache_read_input_token_cost": 5e-7, + "cache_read_input_token_cost_flex": 2.5e-7, + "cache_read_input_token_cost_priority": 8.75e-7, "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, - "cache_read_input_token_cost": 2.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, + "input_cost_per_token_batches": 0.000001, + "input_cost_per_token_flex": 0.000001, + "input_cost_per_token_priority": 0.0000035, + "litellm_provider": "openai", + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "max_tokens": 100000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000008, + "output_cost_per_token_batches": 0.000004, + "output_cost_per_token_flex": 0.000004, + "output_cost_per_token_priority": 0.000014, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "source": "https://developers.openai.com/api/docs/pricing", + "supported_endpoints": [ + "/v1/responses", + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": false, + "supports_parallel_function_calling": false, + "supports_pdf_input": true, "supports_prompt_caching": true, - "supports_web_search": false - }, - "openrouter/qwen/qwen3.8-27b": { - "cache_creation_input_token_cost": 5.3125e-7, - "input_cost_per_token": 4.25e-7, - "output_cost_per_token": 0.00000255, - "cache_read_input_token_cost": 8.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_tool_choice": true, "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_web_search": true }, - "openrouter/qwen/qwen3.8-27b:free": { - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "o3-2025-04-16": { + "cache_read_input_token_cost": 5e-7, + "cache_read_input_token_cost_flex": 2.5e-7, + "cache_read_input_token_cost_priority": 8.75e-7, + "deprecation_date": "2026-12-11", + "input_cost_per_token": 0.000002, + "input_cost_per_token_batches": 0.000001, + "input_cost_per_token_flex": 0.000001, + "input_cost_per_token_priority": 0.0000035, + "litellm_provider": "openai", + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "max_tokens": 100000, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000008, + "output_cost_per_token_batches": 0.000004, + "output_cost_per_token_flex": 0.000004, + "output_cost_per_token_priority": 0.000014, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "source": "https://developers.openai.com/api/docs/pricing", + "supported_endpoints": [ + "/v1/responses", + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_parallel_function_calling": false, + "supports_pdf_input": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": false + "supports_web_search": true }, - "openrouter/qwen/qwen3.8-flash": { - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 4.7e-7, - "cache_read_input_token_cost": 1.6e-8, - "cache_creation_input_token_cost": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, + "o3-mini": { + "cache_read_input_token_cost": 5.5e-7, + "deprecation_date": "2026-10-23", + "input_cost_per_token": 0.0000011, + "input_cost_per_token_batches": 5.5e-7, + "litellm_provider": "openai", + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "max_tokens": 100000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000044, + "output_cost_per_token_batches": 0.0000022, + "source": "https://developers.openai.com/api/docs/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, + "supports_parallel_function_calling": false, "supports_prompt_caching": true, - "supports_web_search": false - }, - "openrouter/qwen/qwen3.8-max": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, - "cache_read_input_token_cost": 2.5e-7, - "cache_creation_input_token_cost": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "source": "https://openrouter.ai/qwen/qwen3.8-max", - "supports_function_calling": true, - "supports_tool_choice": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true - }, - "openrouter/qwen/qwen3.8-max-0902": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, - "cache_read_input_token_cost": 2.5e-7, - "cache_creation_input_token_cost": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/qwen/qwen3.8-max-prime": { - "input_cost_per_token": 0.000004, - "output_cost_per_token": 0.000012, - "cache_read_input_token_cost": 5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, + "o3-mini-2025-01-31": { + "cache_read_input_token_cost": 5.5e-7, + "deprecation_date": "2026-10-23", + "input_cost_per_token": 0.0000011, + "input_cost_per_token_batches": 5.5e-7, + "litellm_provider": "openai", + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "max_tokens": 100000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.0000044, + "output_cost_per_token_batches": 0.0000022, + "source": "https://developers.openai.com/api/docs/pricing", "supports_function_calling": true, - "supports_tool_choice": true, + "supports_parallel_function_calling": false, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": true, - "supports_video_input": true, - "supports_prompt_caching": true + "supports_tool_choice": true, + "supports_vision": false }, - "openrouter/qwen/qwen3.8-omni-flash": { - "cache_read_input_token_cost": 1.6e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, + "o4-mini": { + "cache_read_input_token_cost": 2.75e-7, + "cache_read_input_token_cost_flex": 1.38e-7, + "cache_read_input_token_cost_priority": 5e-7, + "deprecation_date": "2026-10-23", + "input_cost_per_token": 0.0000011, + "input_cost_per_token_batches": 5.5e-7, + "input_cost_per_token_flex": 5.5e-7, + "input_cost_per_token_priority": 0.000002, + "litellm_provider": "openai", + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "max_tokens": 100000, "mode": "chat", - "output_cost_per_token": 4.7e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 0.0000044, + "output_cost_per_token_batches": 0.0000022, + "output_cost_per_token_flex": 0.0000022, + "output_cost_per_token_priority": 0.000008, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "source": "https://developers.openai.com/api/docs/pricing", "supports_function_calling": true, - "supports_pdf_input": false, + "supports_parallel_function_calling": false, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": false + "supports_web_search": true }, - "openrouter/rekaai/reka-edge": { - "input_cost_per_token": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 16384, - "max_output_tokens": 14745, - "max_tokens": 14745, + "o4-mini-2025-04-16": { + "cache_read_input_token_cost": 2.75e-7, + "cache_read_input_token_cost_flex": 1.38e-7, + "cache_read_input_token_cost_priority": 5e-7, + "deprecation_date": "2026-10-23", + "input_cost_per_token": 0.0000011, + "input_cost_per_token_batches": 5.5e-7, + "input_cost_per_token_flex": 5.5e-7, + "input_cost_per_token_priority": 0.000002, + "litellm_provider": "openai", + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "max_tokens": 100000, "mode": "chat", - "output_cost_per_token": 1e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000044, + "output_cost_per_token_batches": 0.0000022, + "output_cost_per_token_flex": 0.0000022, + "output_cost_per_token_priority": 0.000008, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "source": "https://developers.openai.com/api/docs/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "supports_parallel_function_calling": false, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": false + "supports_web_search": true }, - "openrouter/rekaai/reka-flash-3": { - "input_cost_per_token": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 65536, - "max_output_tokens": 58982, - "max_tokens": 58982, + "oci/cohere.command-a-03-2025": { + "input_cost_per_token": 0.00000156, + "litellm_provider": "oci", + "max_input_tokens": 256000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 0.00000156, + "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", + "supports_function_calling": true, + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/relace/relace-apply-3": { - "input_cost_per_token": 8.5e-7, - "litellm_provider": "openrouter", + "oci/cohere.command-a-reasoning": { + "input_cost_per_token": 0.00000156, + "litellm_provider": "oci", "max_input_tokens": 256000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.00000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000156, + "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_native_streaming": true }, - "openrouter/relace/relace-search": { - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", + "oci/cohere.command-a-reasoning-08-2025": { + "input_cost_per_token": 0.00000156, + "litellm_provider": "oci", "max_input_tokens": 256000, - "max_output_tokens": 128000, - "max_tokens": 128000, - "mode": "chat", - "output_cost_per_token": 0.000003, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/sakana/fugu-max": { - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 0.000006, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000156, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_response_schema": false }, - "openrouter/sakana/fugu-ultra": { - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_above_272k_tokens": 0.000001, - "input_cost_per_token": 0.000005, - "input_cost_per_token_above_272k_tokens": 0.00001, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "oci/cohere.command-a-translate-08-2025": { + "input_cost_per_token": 9e-8, + "litellm_provider": "oci", + "max_input_tokens": 256000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 0.00003, - "output_cost_per_token_above_272k_tokens": 0.000045, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "output_cost_per_token": 9e-8, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": false, + "supports_response_schema": false }, - "openrouter/sakana/fugu-ultra-v2": { - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_above_272k_tokens": 0.000001, - "input_cost_per_token": 0.000005, - "input_cost_per_token_above_272k_tokens": 0.00001, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "oci/cohere.command-a-vision": { + "input_cost_per_token": 0.00000156, + "litellm_provider": "oci", + "max_input_tokens": 256000, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.00003, - "output_cost_per_token_above_272k_tokens": 0.000045, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000156, + "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_response_schema": false, + "supports_native_streaming": true, + "supports_vision": true }, - "openrouter/sakana/sakana-namazu": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 9.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 65536, - "max_tokens": 65536, + "oci/cohere.command-a-vision-07-2025": { + "input_cost_per_token": 0.00000156, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 0.000004, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000156, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_response_schema": false, + "supports_vision": true }, - "openrouter/sao10k/l3-lunaris-8b": { - "input_cost_per_token": 4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 8192, - "max_output_tokens": 7372, - "max_tokens": 7372, + "oci/cohere.command-latest": { + "input_cost_per_token": 0.00000156, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 5e-8, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 0.00000156, + "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", + "supports_function_calling": true, + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/sao10k/l3.1-euryale-70b": { - "input_cost_per_token": 8.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "oci/cohere.command-plus-latest": { + "input_cost_per_token": 0.00000156, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 8.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.00000156, + "source": "https://www.oracle.com/artificial-intelligence/enterprise-ai/cost-estimator/", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/sao10k/l3.3-euryale-70b": { - "input_cost_per_token": 6.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 16384, - "max_tokens": 16384, + "oci/cohere.command-r-08-2024": { + "input_cost_per_token": 1.5e-7, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 7.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 1.5e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_response_schema": false }, - "openrouter/stealth/space-bunny-alpha": { - "deprecation_date": "2026-10-05", - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 524288, - "max_tokens": 524288, + "oci/cohere.command-r-plus-08-2024": { + "input_cost_per_token": 0.00000156, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.00000156, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_video_input": true, - "supports_vision": true + "supports_response_schema": false }, - "openrouter/stepfun/step-3.5-flash": { - "input_cost_per_token": 1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, + "oci/google.gemini-2.5-flash": { + "input_cost_per_token": 1.5e-7, + "litellm_provider": "oci", + "max_input_tokens": 1048576, "max_output_tokens": 65536, "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 3e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 6e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_response_schema": true, + "supports_vision": true, + "supports_native_streaming": true, + "supports_image_size": false, "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_system_messages": true, + "supports_pdf_input": true, + "supports_audio_input": true, + "supports_video_input": true }, - "openrouter/stepfun/step-3.7-flash": { - "cache_read_input_token_cost": 4e-8, - "input_cost_per_token": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 230400, - "max_tokens": 230400, + "oci/google.gemini-2.5-flash-lite": { + "input_cost_per_token": 7.5e-8, + "litellm_provider": "oci", + "max_input_tokens": 1048576, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 0.00000115, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 3e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": false - }, - "openrouter/switchpoint/router": { - "input_cost_per_token": 8.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, - "mode": "chat", - "output_cost_per_token": 0.0000034, - "source": "https://openrouter.ai/switchpoint/router", - "supports_tool_choice": true + "supports_native_streaming": true, + "supports_image_size": false, + "supports_reasoning": false, + "supports_system_messages": true, + "supports_pdf_input": true, + "supports_audio_input": true, + "supports_video_input": true }, - "openrouter/tencent/hunyuan-a13b-instruct": { - "input_cost_per_token": 1.4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "oci/google.gemini-2.5-pro": { + "input_cost_per_token": 0.00000125, + "litellm_provider": "oci", + "max_input_tokens": 1048576, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 5.7e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, + "output_cost_per_token": 0.00001, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_vision": true, + "supports_native_streaming": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_pdf_input": true, + "supports_audio_input": true, + "supports_video_input": true }, - "openrouter/tencent/hy-mt2-1.8b": { - "input_cost_per_token": 4.4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 8192, - "max_output_tokens": 4096, - "max_tokens": 4096, + "oci/meta.llama-3.1-405b-instruct": { + "input_cost_per_token": 0.00001068, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 1.77e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, + "output_cost_per_token": 0.00001068, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "supports_native_streaming": true }, - "openrouter/tencent/hy-mt2-30b-a3b": { - "input_cost_per_token": 7.4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 8192, - "max_output_tokens": 4096, - "max_tokens": 4096, + "oci/meta.llama-3.1-70b-instruct": { + "input_cost_per_token": 7.2e-7, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 2.95e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 7.2e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/tencent/hy-mt2-7b": { - "input_cost_per_token": 7.4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 8192, - "max_output_tokens": 4096, - "max_tokens": 4096, + "oci/meta.llama-3.1-8b-instruct": { + "input_cost_per_token": 7.2e-7, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 2.95e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 7.2e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/tencent/hy3": { - "cache_read_input_token_cost": 3.3e-8, - "input_cost_per_token": 1.32e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 128000, - "max_tokens": 128000, + "oci/meta.llama-3.2-11b-vision-instruct": { + "input_cost_per_token": 0.000002, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "off_peak_pricing": { - "hours_utc": "16:00-00:00", - "input_cost_per_token": 8.25e-8, - "output_cost_per_token": 3.3e-7, - "cache_read_input_token_cost": 2.0625e-8 - }, - "output_cost_per_token": 5.28e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000002, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_response_schema": false, + "supports_vision": true }, - "openrouter/tencent/hy3-preview": { - "cache_read_input_token_cost": 6e-8, - "input_cost_per_token": 1.8e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 235929, - "max_tokens": 235929, + "oci/meta.llama-3.2-90b-vision-instruct": { + "input_cost_per_token": 0.000002, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000002, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": false, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_native_streaming": true, + "supports_vision": true }, - "openrouter/tencent/hy4-preview": { - "cache_read_input_token_cost": 4.2e-8, - "input_cost_per_token": 8.34e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 64000, - "max_tokens": 64000, + "oci/meta.llama-3.3-70b-instruct": { + "input_cost_per_token": 7.2e-7, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "off_peak_pricing": { - "input_cost_per_token": 7.506e-7, - "output_cost_per_token": 0.0000022509, - "cache_read_input_token_cost": 3.78e-8, - "hours_utc": "16:00-00:00" - }, - "output_cost_per_token": 0.000002501, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 7.2e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/thedrummer/cydonia-24b-v4.1": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "oci/meta.llama-3.3-70b-instruct-fp8-dynamic": { + "input_cost_per_token": 7.2e-7, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 4000, + "max_tokens": 4000, "mode": "chat", - "output_cost_per_token": 5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 7.2e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_response_schema": false }, - "openrouter/thedrummer/skyfall-36b-v2": { - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 5.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 29491, - "max_tokens": 29491, + "oci/meta.llama-4-maverick-17b-128e-instruct-fp8": { + "input_cost_per_token": 7.2e-7, + "litellm_provider": "oci", + "max_input_tokens": 1048576, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 8e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 7.2e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_response_schema": false, + "supports_native_streaming": true, + "supports_vision": true }, - "openrouter/thedrummer/unslopnemo-12b": { - "input_cost_per_token": 4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1024000, - "max_output_tokens": 819200, - "max_tokens": 819200, + "oci/meta.llama-4-scout-17b-16e-instruct": { + "input_cost_per_token": 7.2e-7, + "litellm_provider": "oci", + "max_input_tokens": 10485760, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 4e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 7.2e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/thinkingmachines/inkling": { - "cache_read_input_token_cost": 1.7e-7, - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 524288, - "max_output_tokens": 471859, - "max_tokens": 471859, + "oci/openai.gpt-5": { + "input_cost_per_token": 0.00000125, + "litellm_provider": "oci", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.00000405, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 0.00001, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, + "supports_native_streaming": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_response_schema": true, + "supports_vision": true }, - "openrouter/thinkingmachines/inkling-small": { - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 4.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 524288, - "max_output_tokens": 262144, - "max_tokens": 262144, + "oci/openai.gpt-5-mini": { + "input_cost_per_token": 2.5e-7, + "litellm_provider": "oci", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.0000012, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 0.000002, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, + "supports_native_streaming": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_response_schema": true, + "supports_vision": true }, - "openrouter/thinkingmachines/inkling-small:free": { - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 262144, - "max_tokens": 262144, + "oci/openai.gpt-5-nano": { + "input_cost_per_token": 5e-8, + "litellm_provider": "oci", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 4e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, + "supports_native_streaming": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false + "supports_response_schema": true, + "supports_vision": true }, - "openrouter/thinkingmachines/inkling:free": { - "input_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 262144, - "max_tokens": 262144, + "oci/openai.gpt-oss-120b": { + "input_cost_per_token": 1.5e-7, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "output_cost_per_token": 6e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": true, - "supports_web_search": false + "supports_native_streaming": true, + "supports_reasoning": true }, - "openrouter/typesafe/jev-router": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_tokens": 1000000, + "oci/openai.gpt-oss-20b": { + "input_cost_per_token": 7e-8, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "source": "https://openrouter.ai/typesafe/jev-router", + "output_cost_per_token": 3e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": true, - "supports_video_input": true + "supports_native_streaming": true, + "supports_reasoning": true }, - "openrouter/unbiased/pareto": { - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, + "oci/xai.grok-3": { + "input_cost_per_token": 0.000003, + "litellm_provider": "oci", + "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.0000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000015, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": false, "supports_response_schema": false, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_native_streaming": true }, - "openrouter/unbiased/pareto-26.10-preview": { - "input_cost_per_token": 8e-7, - "output_cost_per_token": 0.0000032, - "cache_read_input_token_cost": 3e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, + "oci/xai.grok-3-fast": { + "input_cost_per_token": 0.000005, + "litellm_provider": "oci", + "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.000025, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_tool_choice": true, - "supports_vision": true + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/undi95/remm-slerp-l2-13b": { - "input_cost_per_token": 3.5e-7, - "litellm_provider": "openrouter", - "max_tokens": 5529, + "oci/xai.grok-3-mini": { + "input_cost_per_token": 3e-7, + "litellm_provider": "oci", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 6.5e-7, - "supports_tool_choice": false, - "max_input_tokens": 6144, - "max_output_tokens": 5529, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 5e-7, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/upstage/solar-mini4": { - "cache_read_input_token_cost": 5e-9, - "input_cost_per_token": 5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 524288, + "oci/xai.grok-3-mini-fast": { + "input_cost_per_token": 6e-7, + "litellm_provider": "oci", + "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000004, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/upstage/solar-pro-3": { - "cache_read_input_token_cost": 1.5e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 117964, - "max_tokens": 117964, + "oci/xai.grok-4": { + "input_cost_per_token": 0.000003, + "litellm_provider": "oci", + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 6e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000015, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_response_schema": false, + "supports_native_streaming": true }, - "openrouter/upstage/solar-pro4": { - "cache_read_input_token_cost": 1.8e-8, - "input_cost_per_token": 9e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 524288, + "oci/xai.grok-4-fast": { + "input_cost_per_token": 0.000005, + "litellm_provider": "oci", + "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 3.6e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.000025, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_response_schema": false }, - "openrouter/writer/palmyra-x5": { - "input_cost_per_token": 6e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1040000, - "max_output_tokens": 8192, - "max_tokens": 8192, + "oci/xai.grok-4.1-fast": { + "input_cost_per_token": 0.000005, + "litellm_provider": "oci", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.000006, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_reasoning": false, - "supports_response_schema": false, - "supports_tool_choice": false, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 0.000025, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_response_schema": false }, - "openrouter/x-ai/grok-4": { + "oci/xai.grok-4.20": { "input_cost_per_token": 0.000003, - "litellm_provider": "openrouter", - "max_input_tokens": 256000, - "max_output_tokens": 256000, - "max_tokens": 256000, + "litellm_provider": "oci", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", "output_cost_per_token": 0.000015, - "source": "https://openrouter.ai/x-ai/grok-4", + "reasoning_effort_levels": [], + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_web_search": true + "supports_response_schema": false }, - "openrouter/x-ai/grok-4.20": { - "input_cost_per_token": 0.00000125, - "output_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 2000000, - "max_output_tokens": 1800000, - "max_tokens": 1800000, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, + "oci/xai.grok-4.20-0309-non-reasoning": { "cache_read_input_token_cost": 2e-7, "cache_read_input_token_cost_above_200k_tokens": 4e-7, + "input_cost_per_token": 0.00000125, "input_cost_per_token_above_200k_tokens": 0.0000025, + "litellm_provider": "oci", + "max_input_tokens": 1000000, + "max_output_tokens": 1000000, + "max_tokens": 1000000, + "mode": "chat", + "output_cost_per_token": 0.0000025, "output_cost_per_token_above_200k_tokens": 0.000005, - "supports_prompt_caching": true, - "supports_web_search": true + "reasoning_effort_levels": [], + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true }, - "openrouter/x-ai/grok-4.20-multi-agent": { - "input_cost_per_token": 0.00000125, - "output_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 2000000, - "max_output_tokens": 1800000, - "max_tokens": 1800000, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": false, - "supports_tool_choice": false, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, + "oci/xai.grok-4.20-0309-reasoning": { "cache_read_input_token_cost": 2e-7, "cache_read_input_token_cost_above_200k_tokens": 4e-7, + "input_cost_per_token": 0.00000125, "input_cost_per_token_above_200k_tokens": 0.0000025, + "litellm_provider": "oci", + "max_input_tokens": 1000000, + "max_output_tokens": 1000000, + "max_tokens": 1000000, + "mode": "chat", + "output_cost_per_token": 0.0000025, "output_cost_per_token_above_200k_tokens": 0.000005, - "supports_prompt_caching": true, - "supports_web_search": true + "reasoning_effort_levels": [], + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_reasoning": true }, - "openrouter/x-ai/grok-4.3": { - "input_cost_per_token": 0.00000125, - "output_cost_per_token": 0.0000025, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 900000, - "max_tokens": 900000, + "oci/xai.grok-4.20-multi-agent": { + "input_cost_per_token": 0.000003, + "litellm_provider": "oci", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.000015, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, + "supports_response_schema": false + }, + "oci/xai.grok-4.20-non-reasoning": { "cache_read_input_token_cost": 2e-7, "cache_read_input_token_cost_above_200k_tokens": 4e-7, + "input_cost_per_token": 0.00000125, "input_cost_per_token_above_200k_tokens": 0.0000025, + "litellm_provider": "oci", + "max_input_tokens": 1000000, + "max_output_tokens": 1000000, + "max_tokens": 1000000, + "mode": "chat", + "output_cost_per_token": 0.0000025, "output_cost_per_token_above_200k_tokens": 0.000005, - "supports_prompt_caching": true, - "supports_web_search": true + "reasoning_effort_levels": [], + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true }, - "openrouter/x-ai/grok-4.3:batch": { - "cache_read_input_token_cost": 1.6e-7, - "cache_read_input_token_cost_above_200k_tokens": 3.2e-7, - "input_cost_per_token": 0.000001, - "input_cost_per_token_above_200k_tokens": 0.000002, - "litellm_provider": "openrouter", + "oci/xai.grok-4.20-reasoning": { + "cache_read_input_token_cost": 2e-7, + "cache_read_input_token_cost_above_200k_tokens": 4e-7, + "input_cost_per_token": 0.00000125, + "input_cost_per_token_above_200k_tokens": 0.0000025, + "litellm_provider": "oci", "max_input_tokens": 1000000, - "max_output_tokens": 900000, - "max_tokens": 900000, + "max_output_tokens": 1000000, + "max_tokens": 1000000, "mode": "chat", - "output_cost_per_token": 0.000002, - "output_cost_per_token_above_200k_tokens": 0.000004, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 0.0000025, + "output_cost_per_token_above_200k_tokens": 0.000005, + "reasoning_effort_levels": [], + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_reasoning": true }, - "openrouter/x-ai/grok-4.5": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, - "litellm_provider": "openrouter", - "max_input_tokens": 500000, - "max_output_tokens": 450000, - "max_tokens": 450000, + "oci/xai.grok-code-fast-1": { + "input_cost_per_token": 0.000005, + "litellm_provider": "oci", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 0.000025, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 3e-7, - "cache_read_input_token_cost_above_200k_tokens": 6e-7, - "input_cost_per_token_above_200k_tokens": 0.000004, - "output_cost_per_token_above_200k_tokens": 0.000012, - "supports_prompt_caching": true, - "supports_web_search": true + "supports_response_schema": false }, - "openrouter/x-ai/grok-4.6": { - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, - "litellm_provider": "openrouter", - "max_input_tokens": 500000, - "max_output_tokens": 450000, - "max_tokens": 450000, + "ollama/codegeex4": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_above_200k_tokens": 0.000001, - "input_cost_per_token_above_200k_tokens": 0.000004, - "output_cost_per_token_above_200k_tokens": 0.000012, - "supports_prompt_caching": true, - "supports_web_search": true + "output_cost_per_token": 0, + "supports_function_calling": false }, - "openrouter/x-ai/grok-4.7": { - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_above_200k_tokens": 0.000001, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_200k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 500000, - "max_output_tokens": 450000, - "max_tokens": 450000, + "ollama/deepseek-coder-v2-instruct": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.000006, - "output_cost_per_token_above_200k_tokens": 0.000012, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/x-ai/grok-build-0.1": { - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000002, - "litellm_provider": "openrouter", - "max_input_tokens": 256000, - "max_output_tokens": 230400, - "max_tokens": 230400, + "ollama/deepseek-coder-v2-lite-instruct": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_pdf_input": true, - "supports_audio_input": false, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_200k_tokens": 4e-7, - "input_cost_per_token_above_200k_tokens": 0.000002, - "output_cost_per_token_above_200k_tokens": 0.000004, - "supports_prompt_caching": true, - "supports_web_search": true + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/xiaomi/mimo-v2-flash": { - "input_cost_per_token": 1e-7, - "output_cost_per_token": 3e-7, - "cache_creation_input_token_cost": 0, - "cache_read_input_token_cost": 1e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 262144, - "max_output_tokens": 16384, - "max_tokens": 16384, + "ollama/deepseek-v3.1:671b-cloud": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 163840, + "max_output_tokens": 163840, + "max_tokens": 163840, "mode": "chat", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_vision": false, - "supports_prompt_caching": true + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/xiaomi/mimo-v2.5": { - "input_cost_per_token": 1.4e-7, - "output_cost_per_token": 2.8e-7, - "cache_creation_input_token_cost": 0, - "cache_read_input_token_cost": 2.8e-9, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, + "ollama/gpt-oss:120b-cloud": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_vision": true, - "supports_audio_input": true, - "supports_pdf_input": false, - "supports_video_input": true, - "supports_response_schema": true, - "supports_prompt_caching": true, - "supports_web_search": false + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/xiaomi/mimo-v2.5-pro": { - "input_cost_per_token": 4.35e-7, - "output_cost_per_token": 8.7e-7, - "cache_creation_input_token_cost": 0, - "cache_read_input_token_cost": 3.6e-9, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, + "ollama/gpt-oss:20b-cloud": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_vision": false, - "supports_response_schema": true, - "supports_prompt_caching": true, - "supports_web_search": false + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/xiaomi/mimo-v2.6-flash": { - "cache_read_input_token_cost": 2.8e-9, - "input_cost_per_token": 1.4e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, + "ollama/internlm2_5-20b-chat": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 32768, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 2.8e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/xiaomi/mimo-v2.6-pro": { - "cache_read_input_token_cost": 3.6e-9, - "input_cost_per_token": 4.35e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, + "ollama/llama2": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 8.7e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 0 }, - "openrouter/xiaomi/mimo-v2.6-pro-ultraspeed": { - "cache_read_input_token_cost": 3.6e-8, - "input_cost_per_token": 0.00000435, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, + "ollama/llama2:13b": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "max_tokens": 4096, + "mode": "chat", + "output_cost_per_token": 0 + }, + "ollama/llama2:70b": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "max_tokens": 4096, + "mode": "chat", + "output_cost_per_token": 0 + }, + "ollama/llama2:7b": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "max_tokens": 4096, + "mode": "chat", + "output_cost_per_token": 0 + }, + "ollama/llama3": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 0 + }, + "ollama/llama3.1": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 0, + "supports_function_calling": true + }, + "ollama/llama3:70b": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.0000087, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "output_cost_per_token": 0 }, - "openrouter/z-ai/glm-4.5": { - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000022, - "cache_read_input_token_cost": 1.1e-7, - "deprecation_date": "2026-12-31", - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 98304, - "max_tokens": 98304, + "ollama/llama3:8b": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_web_search": false + "output_cost_per_token": 0 }, - "openrouter/z-ai/glm-4.5-air": { - "input_cost_per_token": 1.3e-7, - "output_cost_per_token": 8.5e-7, - "cache_read_input_token_cost": 2.5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, - "max_output_tokens": 98304, - "max_tokens": 98304, + "ollama/mistral-7B-Instruct-v0.1": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_response_schema": false, - "supports_web_search": false + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/z-ai/glm-4.5v": { - "input_cost_per_token": 6e-7, - "output_cost_per_token": 0.0000018, - "cache_read_input_token_cost": 1.1e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 65536, - "max_output_tokens": 16384, - "max_tokens": 16384, + "ollama/mistral-7B-Instruct-v0.2": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": false + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/z-ai/glm-4.6": { - "input_cost_per_token": 4.3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 204800, - "max_output_tokens": 16384, - "max_tokens": 16384, + "ollama/mistral-large-instruct-2407": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 65536, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.00000175, - "source": "https://openrouter.ai/api/v1/models", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true, - "cache_read_input_token_cost": 8e-8, - "supports_audio_input": false, - "supports_pdf_input": false, - "supports_response_schema": true, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/z-ai/glm-4.6:exacto": { - "input_cost_per_token": 4.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 202800, - "max_output_tokens": 131000, - "max_tokens": 131000, + "ollama/mixtral-8x22B-Instruct-v0.1": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 65536, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 0.0000019, - "source": "https://openrouter.ai/z-ai/glm-4.6:exacto", - "supports_function_calling": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_tool_choice": true + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/z-ai/glm-4.6v": { - "input_cost_per_token": 3e-7, - "output_cost_per_token": 9e-7, - "cache_read_input_token_cost": 5e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 131072, + "ollama/mixtral-8x7B-Instruct-v0.1": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 32768, "max_output_tokens": 32768, "max_tokens": 32768, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_web_search": false + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/z-ai/glm-4.7": { - "cache_creation_input_token_cost": 0, - "cache_read_input_token_cost": 1.1e-7, - "deprecation_date": "2026-12-31", - "input_cost_per_token": 6e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 204800, - "max_output_tokens": 131072, - "max_tokens": 131072, + "ollama/qwen3-coder:480b-cloud": { + "input_cost_per_token": 0, + "litellm_provider": "ollama", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.0000022, - "source": "https://openrouter.ai/api/v1/models", - "supports_assistant_prefill": true, - "supports_audio_input": false, - "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "output_cost_per_token": 0, + "supports_function_calling": true }, - "openrouter/z-ai/glm-4.7-flash": { - "input_cost_per_token": 6.05e-8, - "output_cost_per_token": 4e-7, - "cache_creation_input_token_cost": 0, + "openai.gpt-6-luna": { + "input_cost_per_token": 1e-7, + "input_cost_per_token_above_272k_tokens": 2e-7, + "cache_creation_input_token_cost": 1.25e-7, + "cache_creation_input_token_cost_above_272k_tokens": 2.5e-7, "cache_read_input_token_cost": 1e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 117964, - "max_tokens": 117964, + "cache_read_input_token_cost_above_272k_tokens": 2e-8, + "output_cost_per_token": 5e-7, + "output_cost_per_token_above_272k_tokens": 7.5e-7, + "litellm_provider": "bedrock_converse", + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, + "supports_max_reasoning_effort": true, + "supports_minimal_reasoning_effort": false, + "supports_none_reasoning_effort": true, "supports_tool_choice": true, + "supports_prompt_caching": true, "supports_reasoning": true, - "supports_vision": false, - "supports_prompt_caching": false, - "supports_response_schema": true, - "supports_web_search": false + "supports_xhigh_reasoning_effort": true, + "supports_vision": true, + "supports_sampling_params": false, + "source": "https://aws.amazon.com/bedrock/pricing/" }, - "openrouter/z-ai/glm-5": { - "cache_read_input_token_cost": 1.2e-7, - "input_cost_per_token": 6e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 204800, + "openai.gpt-6-sol": { + "input_cost_per_token": 0.000002, + "input_cost_per_token_above_272k_tokens": 0.000004, + "cache_creation_input_token_cost": 0.0000025, + "cache_creation_input_token_cost_above_272k_tokens": 0.000005, + "cache_read_input_token_cost": 2e-7, + "cache_read_input_token_cost_above_272k_tokens": 4e-7, + "output_cost_per_token": 0.00001, + "output_cost_per_token_above_272k_tokens": 0.000015, + "litellm_provider": "bedrock_converse", + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.00000192, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, - "supports_pdf_input": false, + "supports_max_reasoning_effort": true, + "supports_minimal_reasoning_effort": false, + "supports_none_reasoning_effort": true, + "supports_tool_choice": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_xhigh_reasoning_effort": true, + "supports_vision": true, + "supports_sampling_params": false, + "source": "https://aws.amazon.com/bedrock/pricing/" }, - "openrouter/z-ai/glm-5-turbo": { - "input_cost_per_token": 0.0000012, - "output_cost_per_token": 0.000004, - "cache_read_input_token_cost": 2.4e-7, - "deprecation_date": "2098-12-31", - "litellm_provider": "openrouter", - "max_input_tokens": 202752, - "max_output_tokens": 131072, - "max_tokens": 131072, + "openai.gpt-oss-120b-1:0": { + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, + "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7.5e-8, + "input_cost_per_token_flex": 7.5e-8, + "input_cost_per_token_priority": 2.625e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 6e-7, + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_flex": 3e-7, + "output_cost_per_token_priority": 0.00000105, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "supports_prompt_caching": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/z-ai/glm-5.1": { - "cache_creation_input_token_cost": 0, - "cache_read_input_token_cost": 1.794e-7, - "input_cost_per_token": 9.66e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 204800, - "max_output_tokens": 128000, - "max_tokens": 128000, + "openai.gpt-oss-20b-1:0": { + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, + "input_cost_per_token": 7e-8, + "input_cost_per_token_batches": 3.5e-8, + "input_cost_per_token_flex": 3.5e-8, + "input_cost_per_token_priority": 1.225e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", - "output_cost_per_token": 0.000003036, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "output_cost_per_token": 3e-7, + "output_cost_per_token_batches": 1.5e-7, + "output_cost_per_token_flex": 1.5e-7, + "output_cost_per_token_priority": 5.25e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/z-ai/glm-5.2": { - "cache_read_input_token_cost": 1.5e-7, - "input_cost_per_token": 1.52e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, + "openai.gpt-oss-safeguard-120b": { + "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7e-8, + "input_cost_per_token_flex": 7e-8, + "input_cost_per_token_priority": 2.6e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", - "output_cost_per_token": 0.000012, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 6e-7, + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_flex": 3e-7, + "output_cost_per_token_priority": 0.00000105, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false - }, - "openrouter/z-ai/glm-5.2:free": { - "input_cost_per_token": 0, - "output_cost_per_token": 0, - "litellm_provider": "openrouter", - "max_input_tokens": 32768, - "max_output_tokens": 29491, - "max_tokens": 29491, - "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, - "supports_function_calling": false, - "supports_pdf_input": false, - "supports_prompt_caching": false, - "supports_tool_choice": false, - "supports_reasoning": true, - "supports_response_schema": false, - "supports_vision": false, - "supports_web_search": false + "supports_system_messages": true, + "supports_vision": false }, - "openrouter/z-ai/glm-5.3": { - "cache_read_input_token_cost": 6.5e-8, + "openai.gpt-oss-safeguard-20b": { "input_cost_per_token": 7e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "input_cost_per_token_batches": 3e-8, + "input_cost_per_token_flex": 3e-8, + "input_cost_per_token_priority": 1.2e-7, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 128000, + "max_output_tokens": 16000, + "max_tokens": 16000, "mode": "chat", - "output_cost_per_token": 0.000007, - "source": "https://openrouter.ai/api/v1/models", + "output_cost_per_token": 2e-7, + "output_cost_per_token_batches": 1e-7, + "output_cost_per_token_flex": 1e-7, + "output_cost_per_token_priority": 3.5e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_system_messages": true, + "supports_vision": false }, - "openrouter/z-ai/glm-5.3-flash": { - "cache_read_input_token_cost": 3e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1310720, - "max_output_tokens": 131072, - "max_tokens": 131072, + "openai/container": { + "code_interpreter_cost_per_session": 0.03, + "litellm_provider": "openai", + "mode": "chat" + }, + "opencode-go/deepseek-v4-flash": { + "litellm_provider": "opencode-go", + "max_input_tokens": 1000000, + "max_output_tokens": 384000, + "max_tokens": 384000, "mode": "chat", - "output_cost_per_token": 5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/z-ai/glm-5.3-flash:batch": { - "cache_read_input_token_cost": 1.2e-8, - "input_cost_per_token": 6e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, + "opencode-go/deepseek-v4-pro": { + "litellm_provider": "opencode-go", + "max_input_tokens": 1000000, + "max_output_tokens": 384000, + "max_tokens": 384000, "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_assistant_prefill": true, "supports_function_calling": true, - "supports_pdf_input": false, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": false }, - "openrouter/z-ai/glm-5.3-flashx": { - "cache_read_input_token_cost": 9e-8, - "deprecation_date": "2098-12-31", - "input_cost_per_token": 3.7e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, + "opencode-go/glm-5": { + "litellm_provider": "opencode-go", + "max_input_tokens": 200000, + "max_output_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.00000125, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/z-ai/glm-5.3-prime": { - "cache_read_input_token_cost": 5.6e-7, - "input_cost_per_token": 0.0000028, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, - "max_output_tokens": 131072, - "max_tokens": 131072, + "opencode-go/glm-5.1": { + "litellm_provider": "opencode-go", + "max_input_tokens": 200000, + "max_output_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.0000088, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true }, - "openrouter/z-ai/glm-5.3:batch": { - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 4.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 131072, - "max_tokens": 131072, + "opencode-go/glm-5.2": { + "litellm_provider": "opencode-go", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.000002, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_reasoning": true }, - "openrouter/z-ai/glm-5v-turbo": { - "input_cost_per_token": 0.0000012, - "output_cost_per_token": 0.000004, - "cache_read_input_token_cost": 2.4e-7, - "deprecation_date": "2098-12-31", - "litellm_provider": "openrouter", - "max_input_tokens": 202752, - "max_output_tokens": 131072, - "max_tokens": 131072, + "opencode-go/glm-5.3": { + "litellm_provider": "opencode-go", + "max_input_tokens": 200000, + "max_output_tokens": 128000, "mode": "chat", - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_tool_choice": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_vision": true, "supports_prompt_caching": true, - "supports_web_search": false + "supports_reasoning": true, + "supports_tool_choice": true }, - "openrouter/~anthropic/claude-fable-latest": { - "cache_creation_input_token_cost": 0.0000125, - "cache_creation_input_token_cost_above_1hr": 0.00002, - "cache_read_input_token_cost": 2.5e-7, - "input_cost_per_token": 0.00001, - "litellm_provider": "openrouter", - "max_input_tokens": 1000000, + "opencode-go/gpt-5.6-luna": { + "litellm_provider": "opencode-go", + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.00005, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, + "supports_minimal_reasoning_effort": false, + "supports_native_streaming": true, + "supports_none_reasoning_effort": true, + "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "supports_web_search": true, + "supports_xhigh_reasoning_effort": true }, - "openrouter/~anthropic/claude-haiku-latest": { - "cache_creation_input_token_cost": 0.00000125, - "cache_creation_input_token_cost_above_1hr": 0.000002, - "cache_read_input_token_cost": 1e-7, - "input_cost_per_token": 0.000001, - "litellm_provider": "openrouter", - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "max_tokens": 64000, + "opencode-go/grok-4.5": { + "litellm_provider": "opencode-go", + "max_input_tokens": 500000, + "max_output_tokens": 500000, + "max_tokens": 500000, "mode": "chat", - "output_cost_per_token": 0.000005, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, @@ -53968,405 +45797,343 @@ "supports_vision": true, "supports_web_search": true }, - "openrouter/~anthropic/claude-opus-latest": { - "cache_creation_input_token_cost": 0.000005, - "cache_creation_input_token_cost_above_1hr": 0.000008, - "cache_read_input_token_cost": 2e-7, - "input_cost_per_token": 0.000004, - "litellm_provider": "openrouter", + "opencode-go/hy3": { + "litellm_provider": "opencode-go", "max_input_tokens": 1000000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.00002, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/~anthropic/claude-sonnet-latest": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_1hr": 0.000004, - "cache_read_input_token_cost": 2e-7, - "input_cost_per_token": 0.000002, - "litellm_provider": "openrouter", + "opencode-go/hy3-preview": { + "litellm_provider": "opencode-go", "max_input_tokens": 1000000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "output_cost_per_token": 0.00001, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/~deepseek/deepseek-flash-latest": { - "cache_read_input_token_cost": 6e-9, - "input_cost_per_token": 3e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "opencode-go/kimi-k2.5": { + "litellm_provider": "opencode-go", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 0.0000012, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_video_input": true, + "supports_vision": true }, - "openrouter/~deepseek/deepseek-pro-latest": { - "cache_read_input_token_cost": 4.4e-8, - "input_cost_per_token": 0.00000132, - "litellm_provider": "openrouter", - "max_input_tokens": 1048576, - "max_output_tokens": 384000, - "max_tokens": 384000, + "opencode-go/kimi-k2.6": { + "litellm_provider": "opencode-go", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "off_peak_pricing": { - "windows": [ - { - "weekdays": [ - "saturday", - "sunday" - ], - "hours_utc": "00:00-00:00" - }, - { - "weekdays": [ - "monday", - "tuesday", - "wednesday", - "thursday", - "friday" - ], - "hours_utc": "00:00-01:00" - }, - { - "weekdays": [ - "monday", - "tuesday", - "wednesday", - "thursday", - "friday" - ], - "hours_utc": "04:00-06:00" - }, - { - "weekdays": [ - "monday", - "tuesday", - "wednesday", - "thursday", - "friday" - ], - "hours_utc": "10:00-00:00" - } - ], - "input_cost_per_token": 6.6e-7, - "output_cost_per_token": 0.00000198, - "cache_read_input_token_cost": 2.2e-8 - }, - "output_cost_per_token": 0.00000396, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_video_input": true, + "supports_vision": true }, - "openrouter/~deepseek/deepseek-v4-flash-latest": { - "cache_read_input_token_cost": 1.6e-8, - "input_cost_per_token": 4e-8, - "litellm_provider": "openrouter", - "max_input_tokens": 1310720, - "max_output_tokens": 943718, - "max_tokens": 943718, + "opencode-go/kimi-k2.7-code": { + "litellm_provider": "opencode-go", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, "mode": "chat", - "output_cost_per_token": 6.4e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_reasoning": true + }, + "opencode-go/kimi-k3": { + "litellm_provider": "opencode-go", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": false, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_video_input": true, + "supports_vision": true }, - "openrouter/~google/gemini-flash-latest": { - "cache_creation_input_token_cost": 4.16666666666667e-8, - "cache_read_input_audio_token_cost": 7.5e-8, - "cache_read_input_token_cost": 7.5e-8, - "input_cost_per_audio_token": 7.5e-7, - "input_cost_per_token": 7.5e-7, - "litellm_provider": "openrouter", + "opencode-go/mimo-v2-omni": { + "litellm_provider": "opencode-go", "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.00000375, - "source": "https://openrouter.ai/api/v1/models", + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_video_input": true, + "supports_vision": true }, - "openrouter/~google/gemini-pro-latest": { - "cache_creation_input_token_cost": 3.75e-7, - "cache_read_input_audio_token_cost": 2e-7, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_200k_tokens": 4e-7, - "input_cost_per_audio_token": 0.000002, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_200k_tokens": 0.000004, - "litellm_provider": "openrouter", + "opencode-go/mimo-v2-pro": { + "litellm_provider": "opencode-go", "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", - "output_cost_per_token": 0.000012, - "output_cost_per_token_above_200k_tokens": 0.000018, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": true, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/~moonshotai/kimi-latest": { - "cache_read_input_token_cost": 3e-7, - "input_cost_per_token": 0.000003, - "litellm_provider": "openrouter", + "opencode-go/mimo-v2.5": { + "litellm_provider": "opencode-go", "max_input_tokens": 1048576, - "max_output_tokens": 943718, - "max_tokens": 943718, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_audio_input": true, "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_video_input": true, + "supports_vision": true }, - "openrouter/~openai/gpt-astra-latest": { - "cache_creation_input_token_cost": 0.0000125, - "cache_creation_input_token_cost_above_272k_tokens": 0.000025, - "cache_read_input_token_cost": 0.000001, - "cache_read_input_token_cost_above_272k_tokens": 0.000002, - "input_cost_per_token": 0.00001, - "input_cost_per_token_above_272k_tokens": 0.00002, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "opencode-go/mimo-v2.5-pro": { + "litellm_provider": "opencode-go", + "max_input_tokens": 1048576, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", - "output_cost_per_token": 0.00005, - "output_cost_per_token_above_272k_tokens": 0.000075, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/~openai/gpt-luna-latest": { - "cache_creation_input_token_cost": 1.25e-7, - "cache_creation_input_token_cost_above_272k_tokens": 2.5e-7, - "cache_read_input_token_cost": 1e-8, - "cache_read_input_token_cost_above_272k_tokens": 2e-8, - "input_cost_per_token": 1e-7, - "input_cost_per_token_above_272k_tokens": 2e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "opencode-go/minimax-m2.5": { + "litellm_provider": "opencode-go", + "max_input_tokens": 196608, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 5e-7, - "output_cost_per_token_above_272k_tokens": 7.5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_computer_use": false, "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": false }, - "openrouter/~openai/gpt-mini-latest": { - "cache_read_input_token_cost": 7.5e-8, - "input_cost_per_token": 7.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 400000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "opencode-go/minimax-m2.7": { + "litellm_provider": "opencode-go", + "max_input_tokens": 1000192, + "max_output_tokens": 1000192, + "max_tokens": 1000192, "mode": "chat", - "output_cost_per_token": 0.0000045, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_reasoning": false + }, + "opencode-go/minimax-m3": { + "litellm_provider": "opencode-go", + "max_input_tokens": 512000, + "max_output_tokens": 512000, + "max_tokens": 512000, + "mode": "chat", + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/~openai/gpt-sol-latest": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_272k_tokens": 4e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "opencode-go/qwen3.5-plus": { + "litellm_provider": "opencode-go", + "max_input_tokens": 991808, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_272k_tokens": 0.000015, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/~openai/gpt-terra-latest": { - "cache_creation_input_token_cost": 0.0000025, - "cache_creation_input_token_cost_above_272k_tokens": 0.000005, - "cache_read_input_token_cost": 2e-7, - "cache_read_input_token_cost_above_272k_tokens": 4e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_272k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 1050000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "opencode-go/qwen3.6-plus": { + "litellm_provider": "opencode-go", + "max_input_tokens": 1000000, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 0.000012, - "output_cost_per_token_above_272k_tokens": 0.000018, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, "supports_reasoning": true, - "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_vision": true }, - "openrouter/~x-ai/grok-latest": { - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_above_200k_tokens": 0.000001, - "input_cost_per_token": 0.000002, - "input_cost_per_token_above_200k_tokens": 0.000004, - "litellm_provider": "openrouter", - "max_input_tokens": 500000, - "max_output_tokens": 450000, - "max_tokens": 450000, + "opencode-go/qwen3.7-max": { + "litellm_provider": "opencode-go", + "max_input_tokens": 991808, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 0.000006, - "output_cost_per_token_above_200k_tokens": 0.000012, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": true + "supports_tool_choice": true }, - "openrouter/~z-ai/glm-flash-latest": { - "cache_read_input_token_cost": 5e-8, - "input_cost_per_token": 1.5e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1310720, - "max_output_tokens": 943718, - "max_tokens": 943718, + "opencode-go/qwen3.7-plus": { + "litellm_provider": "opencode-go", + "max_input_tokens": 991808, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 5e-7, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_web_search": false + "supports_vision": true }, - "openrouter/~z-ai/glm-latest": { - "cache_read_input_token_cost": 1.2142e-7, - "input_cost_per_token": 6.538e-7, - "litellm_provider": "openrouter", - "max_input_tokens": 1310720, - "max_output_tokens": 131072, - "max_tokens": 131072, + "opencode-go/qwen3.8-max": { + "litellm_provider": "opencode-go", + "max_input_tokens": 991808, + "max_output_tokens": 65536, + "max_tokens": 65536, "mode": "chat", - "output_cost_per_token": 0.0000020548, - "source": "https://openrouter.ai/api/v1/models", - "supports_audio_input": false, + "source": "https://opencode.ai", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, - "supports_pdf_input": false, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": false, - "supports_web_search": false + "supports_tool_choice": true + }, + "openinfer/@oi/GLM-5.3-Flash": { + "max_tokens": 393216, + "max_input_tokens": 393216, + "max_output_tokens": 393216, + "input_cost_per_token": 1.95e-7, + "output_cost_per_token": 0.00000156, + "litellm_provider": "openinfer", + "mode": "chat", + "supports_function_calling": true, + "supports_reasoning": true, + "source": "https://api.openinfer.ai/v1/models" + }, + "openinfer/@oi/Qwen3.5-27B": { + "max_tokens": 32768, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "input_cost_per_token": 1.95e-7, + "output_cost_per_token": 0.00000156, + "litellm_provider": "openinfer", + "mode": "chat", + "supports_function_calling": true, + "supports_reasoning": true, + "source": "https://api.openinfer.ai/v1/models" + }, + "openinfer/@oi/Qwen3.5-9B": { + "max_tokens": 100000, + "max_input_tokens": 100000, + "max_output_tokens": 100000, + "input_cost_per_token": 0.000005, + "output_cost_per_token": 1.5e-7, + "litellm_provider": "openinfer", + "mode": "chat", + "supports_function_calling": true, + "supports_reasoning": true, + "source": "https://api.openinfer.ai/v1/models" }, "ovhcloud/DeepSeek-R1-Distill-Llama-70B": { "input_cost_per_token": 6.7e-7, @@ -55025,12 +46792,14 @@ }, "qwen.qwen3-235b-a22b-2507-v1:0": { "input_cost_per_token": 2.2e-7, + "input_cost_per_token_flex": 1.1e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 262144, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 8.8e-7, + "output_cost_per_token_flex": 4.4e-7, "supports_function_calling": true, "supports_reasoning": true, "supports_tool_choice": true, @@ -55041,13 +46810,19 @@ }, "qwen.qwen3-32b-v1:0": { "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7.5e-8, + "input_cost_per_token_flex": 7.5e-8, + "input_cost_per_token_priority": 2.625e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 32768, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 6e-7, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-32b.html", + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_flex": 3e-7, + "output_cost_per_token_priority": 0.00000105, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_tool_choice": true, @@ -55056,13 +46831,17 @@ "qwen.qwen3-coder-30b-a3b-v1:0": { "deprecation_date": "2027-03-30", "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7.5e-8, + "input_cost_per_token_priority": 2.625e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 6e-7, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-30b-a3b-instruct.html", + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_priority": 0.00000105, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_tool_choice": true, @@ -55072,7 +46851,7 @@ "deprecation_date": "2027-03-30", "input_cost_per_token": 4.5e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 131072, + "max_input_tokens": 262144, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", @@ -55085,16 +46864,22 @@ }, "qwen.qwen3-coder-next": { "input_cost_per_token": 5e-7, + "input_cost_per_token_batches": 2.5e-7, + "input_cost_per_token_flex": 2.5e-7, + "input_cost_per_token_priority": 8.75e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 256000, "max_output_tokens": 16000, "max_tokens": 16000, "mode": "chat", "output_cost_per_token": 0.0000012, + "output_cost_per_token_batches": 6e-7, + "output_cost_per_token_flex": 6e-7, + "output_cost_per_token_priority": 0.0000021, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false @@ -55117,13 +46902,17 @@ }, "qwen.qwen3-vl-235b-a22b": { "input_cost_per_token": 5.3e-7, + "input_cost_per_token_batches": 2.6e-7, + "input_cost_per_token_priority": 9.3e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 256000, "max_output_tokens": 8000, "max_tokens": 8000, "mode": "chat", "output_cost_per_token": 0.00000266, - "source": "https://aws.amazon.com/bedrock/pricing/", + "output_cost_per_token_batches": 0.00000133, + "output_cost_per_token_priority": 0.00000466, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_function_calling": true, "supports_system_messages": true, @@ -57540,14 +49329,14 @@ "max_tokens": 1048576, "max_input_tokens": 1048576, "max_output_tokens": 1048576, - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 6e-7, + "input_cost_per_token": 9e-8, + "output_cost_per_token": 3e-7, "cache_read_input_token_cost": 6e-9, - "input_cost_per_token_balanced": 1.2e-7, - "output_cost_per_token_balanced": 4.8e-7, + "input_cost_per_token_balanced": 7e-8, + "output_cost_per_token_balanced": 2.2e-7, "cache_read_input_token_cost_balanced": 5e-9, - "input_cost_per_token_flex": 8e-8, - "output_cost_per_token_flex": 3e-7, + "input_cost_per_token_flex": 5e-8, + "output_cost_per_token_flex": 1.5e-7, "cache_read_input_token_cost_flex": 4e-9, "litellm_provider": "sail", "mode": "chat", @@ -57846,6 +49635,52 @@ "output_cost_per_token": 0, "supports_reasoning": true }, + "scaledown/classify": { + "input_cost_per_token": 5e-8, + "output_cost_per_token": 0, + "litellm_provider": "scaledown", + "mode": "chat", + "supports_system_messages": false, + "supports_vision": false, + "supports_function_calling": false + }, + "scaledown/compress": { + "input_cost_per_token": 5e-8, + "output_cost_per_token": 0, + "litellm_provider": "scaledown", + "mode": "chat", + "supports_system_messages": true, + "supports_function_calling": false, + "supports_vision": false + }, + "scaledown/decisions": { + "input_cost_per_token": 5e-8, + "output_cost_per_token": 0, + "litellm_provider": "scaledown", + "mode": "chat", + "supports_system_messages": false, + "supports_vision": false, + "supports_function_calling": false + }, + "scaledown/extract": { + "input_cost_per_token": 5e-8, + "output_cost_per_token": 0, + "litellm_provider": "scaledown", + "mode": "chat", + "supports_response_schema": false, + "supports_system_messages": false, + "supports_vision": false, + "supports_function_calling": false + }, + "scaledown/summarize": { + "input_cost_per_token": 5e-8, + "output_cost_per_token": 0, + "litellm_provider": "scaledown", + "mode": "chat", + "supports_system_messages": true, + "supports_vision": false, + "supports_function_calling": false + }, "scaleway/deepseek-v4-flash-0731": { "cache_read_input_token_cost": 8e-8, "input_cost_per_token": 4e-7, @@ -57987,6 +49822,21 @@ "supports_vision": true, "supports_reasoning": true }, + "scx-ai/DeepSeek-V4.1-flash": { + "cache_read_input_token_cost": 1e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "scx-ai", + "max_input_tokens": 1000000, + "max_output_tokens": 375000, + "max_tokens": 375000, + "mode": "chat", + "output_cost_per_token": 6e-7, + "source": "https://scx.ai/models/deepseek-v4-1-flash", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_vision": true + }, "scx-ai/GLM-5.2": { "cache_read_input_token_cost": 2.2e-7, "input_cost_per_token": 6.1e-7, @@ -58004,6 +49854,38 @@ "supports_tool_choice": true, "supports_vision": false }, + "scx-ai/GLM-5.3": { + "cache_read_input_token_cost": 1.8e-7, + "input_cost_per_token": 9.8e-7, + "litellm_provider": "scx-ai", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00000308, + "source": "https://scx.ai/models/glm-5-3", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": false + }, + "scx-ai/Kimi-K3": { + "cache_read_input_token_cost": 2.8e-7, + "input_cost_per_token": 0.00000283, + "litellm_provider": "scx-ai", + "max_input_tokens": 1000000, + "max_output_tokens": 1000000, + "max_tokens": 1000000, + "mode": "chat", + "output_cost_per_token": 0.00001413, + "source": "https://scx.ai/models/kimi-k3", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": false + }, "scx-ai/Qwen3.8-Max": { "cache_read_input_token_cost": 2.1e-7, "input_cost_per_token": 0.00000165, @@ -58402,7 +50284,7 @@ "max_tokens": 384000, "mode": "chat", "output_cost_per_token": 2.8e-7, - "source": "https://www.tencentcloud.com/products/tokenhub", + "source": "https://www.tencentcloud.com/document/product/1300/78937", "supported_endpoints": [ "/v1/chat/completions" ], @@ -58419,16 +50301,71 @@ }, "tencent/deepseek-v4-pro": { "cache_creation_input_token_cost": 0, - "cache_read_input_token_cost": 3.625e-9, - "input_cost_per_token": 4.35e-7, - "input_cost_per_token_cache_hit": 3.625e-9, + "cache_read_input_token_cost": 1.45e-7, + "input_cost_per_token": 0.00000174, + "input_cost_per_token_cache_hit": 1.45e-7, "litellm_provider": "tencent", "max_input_tokens": 1000000, "max_output_tokens": 384000, "max_tokens": 384000, "mode": "chat", - "output_cost_per_token": 8.7e-7, - "source": "https://www.tencentcloud.com/products/tokenhub", + "output_cost_per_token": 0.00000348, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_vision": false + }, + "tencent/deepseek-v4-pro-0813": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 4.4e-8, + "input_cost_per_token": 0.00000132, + "input_cost_per_token_cache_hit": 4.4e-8, + "litellm_provider": "tencent", + "max_input_tokens": 1000000, + "max_output_tokens": 384000, + "max_tokens": 384000, + "mode": "chat", + "off_peak_pricing": { + "cache_read_input_token_cost": 2.2e-8, + "input_cost_per_token": 6.6e-7, + "output_cost_per_token": 0.00000198, + "weekday_timezone": "Asia/Shanghai", + "windows": [ + { + "hours_utc": [ + "00:00-01:00", + "04:00-06:00", + "10:00-00:00" + ], + "weekdays": [ + 1, + 2, + 3, + 4, + 5 + ] + }, + { + "hours_utc": "00:00-00:00", + "weekdays": [ + 6, + 7 + ] + } + ] + }, + "output_cost_per_token": 0.00000396, + "source": "https://www.tencentcloud.com/document/product/1300/78937", "supported_endpoints": [ "/v1/chat/completions" ], @@ -58443,6 +50380,255 @@ "supports_reasoning": true, "supports_vision": false }, + "tencent/deepseek-v4.1-flash": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 6e-9, + "input_cost_per_token": 3e-7, + "input_cost_per_token_cache_hit": 6e-9, + "litellm_provider": "tencent", + "max_input_tokens": 1000000, + "max_output_tokens": 384000, + "max_tokens": 384000, + "mode": "chat", + "off_peak_pricing": { + "cache_read_input_token_cost": 3e-9, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 6e-7, + "weekday_timezone": "Asia/Shanghai", + "windows": [ + { + "hours_utc": [ + "00:00-01:00", + "04:00-06:00", + "10:00-00:00" + ], + "weekdays": [ + 1, + 2, + 3, + 4, + 5 + ] + }, + { + "hours_utc": "00:00-00:00", + "weekdays": [ + 6, + 7 + ] + } + ] + }, + "output_cost_per_token": 0.0000012, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_vision": true + }, + "tencent/glm-5.2": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 2.6e-7, + "input_cost_per_token": 0.0000014, + "input_cost_per_token_cache_hit": 2.6e-7, + "litellm_provider": "tencent", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.0000044, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "tencent/glm-5.3": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 2.6e-7, + "input_cost_per_token": 0.0000014, + "input_cost_per_token_cache_hit": 2.6e-7, + "litellm_provider": "tencent", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.0000044, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "tencent/glm-5.3-flash": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token": 1.5e-7, + "input_cost_per_token_cache_hit": 3e-8, + "litellm_provider": "tencent", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 5e-7, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_video_input": true, + "supports_vision": true + }, + "tencent/glm-5.3-flashx": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 7.5e-8, + "input_cost_per_token": 3.7e-7, + "input_cost_per_token_cache_hit": 7.5e-8, + "litellm_provider": "tencent", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00000125, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_video_input": true, + "supports_vision": true + }, + "tencent/kimi-k2.7-code": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 1.9e-7, + "input_cost_per_token": 9.5e-7, + "input_cost_per_token_cache_hit": 1.9e-7, + "litellm_provider": "tencent", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, + "mode": "chat", + "output_cost_per_token": 0.000004, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "tencent/kimi-k2.7-code-highspeed": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 3.8e-7, + "input_cost_per_token": 0.0000019, + "input_cost_per_token_cache_hit": 3.8e-7, + "litellm_provider": "tencent", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, + "mode": "chat", + "output_cost_per_token": 0.000008, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "tencent/kimi-k2.8-preview": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 2.5e-7, + "input_cost_per_token": 0.000001, + "input_cost_per_token_cache_hit": 2.5e-7, + "litellm_provider": "tencent", + "max_input_tokens": 1000000, + "max_output_tokens": 1000000, + "max_tokens": 1000000, + "mode": "chat", + "output_cost_per_token": 0.000004, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "tencent/kimi-k3": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 3e-7, + "input_cost_per_token": 0.000003, + "input_cost_per_token_cache_hit": 3e-7, + "litellm_provider": "tencent", + "max_input_tokens": 1000000, + "max_output_tokens": 1000000, + "max_tokens": 1000000, + "mode": "chat", + "output_cost_per_token": 0.000015, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, "tencent/minimax-m3": { "cache_creation_input_token_cost": 0, "cache_read_input_token_cost": 6e-8, @@ -58452,7 +50638,7 @@ "max_input_tokens": 1000000, "mode": "chat", "output_cost_per_token": 0.0000012, - "source": "https://www.tencentcloud.com/products/tokenhub", + "source": "https://www.tencentcloud.com/document/product/1300/78937", "supported_endpoints": [ "/v1/chat/completions" ], @@ -58907,7 +51093,7 @@ "max_tokens": 1000000, "mode": "chat", "output_cost_per_token": 0.0000045, - "source": "https://api.together.ai/v1/models", + "source": "https://www.together.ai/pricing", "supports_prompt_caching": true }, "together_ai/Qwen/Qwen3.7-Plus": { @@ -59023,7 +51209,7 @@ }, "together_ai/deepseek-ai/DeepSeek-V4-Flash-0731": { "cache_read_input_token_cost": 3e-8, - "deprecation_date": "2026-09-29", + "deprecation_date": "2026-10-22", "input_cost_per_token": 1.4e-7, "litellm_provider": "together_ai", "max_input_tokens": 1048576, @@ -59039,7 +51225,6 @@ }, "together_ai/deepseek-ai/DeepSeek-V4-Pro-0813": { "cache_read_input_token_cost": 1.3e-7, - "deprecation_date": "2026-09-29", "input_cost_per_token": 0.00000132, "litellm_provider": "together_ai", "max_input_tokens": 1048576, @@ -59135,6 +51320,7 @@ "max_tokens": 131072 }, "together_ai/meta-llama/Llama-3.3-70B-Instruct-Turbo": { + "deprecation_date": "2026-10-22", "input_cost_per_token": 0.00000104, "litellm_provider": "together_ai", "max_input_tokens": 131072, @@ -59208,6 +51394,19 @@ "source": "https://api.together.ai/v1/models", "supports_prompt_caching": true }, + "together_ai/meta-models/Muse-Spark-Open-816B-A42B": { + "cache_read_input_token_cost": 1.5e-7, + "input_cost_per_token": 0.00000125, + "litellm_provider": "together_ai", + "max_input_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.00000425, + "source": "https://docs.together.ai/docs/serverless-models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true + }, "together_ai/mistralai/Ministral-3-14B-Instruct-2512": { "input_cost_per_token": 2e-7, "output_cost_per_token": 2e-7, @@ -59299,13 +51498,13 @@ "source": "https://api.together.ai/v1/models" }, "together_ai/moonshotai/Kimi-K3": { - "cache_read_input_token_cost": 2.7e-7, - "input_cost_per_token": 0.0000027, + "cache_read_input_token_cost": 3e-7, + "input_cost_per_token": 0.000003, "litellm_provider": "together_ai", "max_input_tokens": 1048576, "max_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 0.0000135, + "output_cost_per_token": 0.000015, "reasoning_effort_levels": [ "low", "high", @@ -59330,6 +51529,23 @@ "output_cost_per_token": 8.8e-7, "source": "https://api.together.ai/v1/models" }, + "together_ai/nvidia/NVIDIA-Nemotron-3.5-Super-VL": { + "cache_read_input_token_cost": 2e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "together_ai", + "max_input_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 5e-7, + "source": "https://api.together.ai/v1/models", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "together_ai/nvidia/NVIDIA-Nemotron-Nano-9B-v2": { "deprecation_date": "2026-02-25", "input_cost_per_token": 6e-8, @@ -59374,6 +51590,19 @@ "output_cost_per_token": 2e-7, "source": "https://api.together.ai/v1/models" }, + "together_ai/reflection/Beam-1.0-501B-A23B": { + "cache_read_input_token_cost": 4e-8, + "input_cost_per_token": 4e-7, + "litellm_provider": "together_ai", + "max_input_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://api.together.ai/v1/models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true + }, "together_ai/thinkingmachines/Inkling": { "cache_read_input_token_cost": 1.7e-7, "input_cost_per_token": 0.000001, @@ -59390,7 +51619,6 @@ "supports_tool_choice": true }, "together_ai/together/Tev1-4B-experimental": { - "cache_read_input_token_cost": 4.2e-8, "input_cost_per_token": 4.2e-8, "litellm_provider": "together_ai", "max_input_tokens": 32768, @@ -59521,6 +51749,17 @@ "mode": "chat", "supports_video_input": true }, + "twelvelabs.pegasus-1-5-v1:0": { + "input_cost_per_video_per_second": 0.00049, + "litellm_provider": "bedrock", + "max_input_tokens": 261120, + "max_output_tokens": 98304, + "max_tokens": 98304, + "mode": "chat", + "output_cost_per_token": 0.0000075, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-twelvelabs-pegasus-v1-5.html", + "supports_video_input": true + }, "us-gov.anthropic.claude-fable-5-1": { "cache_creation_input_token_cost": 0.000015, "cache_creation_input_token_cost_above_1hr": 0.000024, @@ -59585,6 +51824,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -59619,6 +51859,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -59752,6 +51993,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -59788,6 +52030,7 @@ "supports_max_reasoning_effort": true, "supports_mid_conversation_system": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": false, "supports_output_config": true, "supports_parallel_tool_use_config": true, "supports_pdf_input": true, @@ -59800,6 +52043,7 @@ "supports_xhigh_reasoning_effort": true }, "us-gov.nvidia.nemotron-nano-12b-v2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 2.4e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, @@ -59807,6 +52051,7 @@ "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 7.2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-nvidia-nvidia-nemotron-nano-12b-v2-vl-bf16.html", "supports_system_messages": true, "supports_vision": true, "supports_audio_input": false, @@ -59831,6 +52076,7 @@ "supports_vision": false }, "us-gov.nvidia.nemotron-nano-9b-v2": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 7.2e-8, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, @@ -59838,6 +52084,7 @@ "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 2.76e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-nvidia-nvidia-nemotron-nano-9b-v2.html", "supports_system_messages": true, "supports_audio_input": false, "supports_function_calling": true, @@ -59865,14 +52112,16 @@ "input_cost_per_token": 1.8e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 7.2e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-120b.html", "supported_endpoints": [ "/v1/chat/completions" ], "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, @@ -59882,14 +52131,16 @@ "input_cost_per_token": 8.4e-8, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, - "max_output_tokens": 128000, - "max_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 3.6e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-20b.html", "supported_endpoints": [ "/v1/chat/completions" ], "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_inline_reasoning": true, "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, @@ -59918,13 +52169,22 @@ }, "us.amazon.nova-2-lite-v1:0": { "cache_read_input_token_cost": 8.25e-8, + "cache_read_input_token_cost_flex": 4.125e-8, + "cache_read_input_token_cost_priority": 1.44375e-7, "input_cost_per_token": 3.3e-7, + "input_cost_per_token_batches": 1.65e-7, + "input_cost_per_token_flex": 1.65e-7, + "input_cost_per_token_priority": 5.775e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 1000000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", "output_cost_per_token": 0.00000275, + "output_cost_per_token_batches": 0.000001375, + "output_cost_per_token_flex": 0.000001375, + "output_cost_per_token_priority": 0.0000048125, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -59939,13 +52199,17 @@ "input_cost_per_token": 0.000001375, "input_cost_per_image_token": 0.000001375, "input_cost_per_audio_token": 0.000001375, + "input_cost_per_token_batches": 6.875e-7, + "input_cost_per_token_priority": 0.00000240625, "litellm_provider": "bedrock_converse", "max_input_tokens": 1000000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", "output_cost_per_token": 0.000011, - "source": "https://aws.amazon.com/nova/pricing/", + "output_cost_per_token_batches": 0.0000055, + "output_cost_per_token_priority": 0.00001925, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -59958,12 +52222,15 @@ "us.amazon.nova-lite-v1:0": { "cache_read_input_token_cost": 1.5e-8, "input_cost_per_token": 6e-8, + "input_cost_per_token_batches": 3e-8, "litellm_provider": "bedrock_converse", "max_input_tokens": 300000, "max_output_tokens": 10000, "max_tokens": 10000, "mode": "chat", "output_cost_per_token": 2.4e-7, + "output_cost_per_token_batches": 1.2e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -59974,12 +52241,15 @@ "us.amazon.nova-micro-v1:0": { "cache_read_input_token_cost": 8.75e-9, "input_cost_per_token": 3.5e-8, + "input_cost_per_token_batches": 1.75e-8, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, "max_output_tokens": 10000, "max_tokens": 10000, "mode": "chat", "output_cost_per_token": 1.4e-7, + "output_cost_per_token_batches": 7e-8, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_prompt_caching": true, "supports_response_schema": true, @@ -59987,13 +52257,22 @@ }, "us.amazon.nova-pro-v1:0": { "cache_read_input_token_cost": 2e-7, + "cache_read_input_token_cost_flex": 1e-7, + "cache_read_input_token_cost_priority": 3.5e-7, "input_cost_per_token": 8e-7, + "input_cost_per_token_batches": 4e-7, + "input_cost_per_token_flex": 4e-7, + "input_cost_per_token_priority": 0.0000014, "litellm_provider": "bedrock_converse", "max_input_tokens": 300000, "max_output_tokens": 10000, "max_tokens": 10000, "mode": "chat", "output_cost_per_token": 0.0000032, + "output_cost_per_token_batches": 0.0000016, + "output_cost_per_token_flex": 0.0000016, + "output_cost_per_token_priority": 0.0000056, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, @@ -60119,7 +52398,7 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_native_structured_output": true, + "supports_native_structured_output": false, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -60206,6 +52485,10 @@ "cache_read_input_token_cost": 1.1e-8, "input_cost_per_token": 1.1e-7, "litellm_provider": "bedrock_converse", + "token_count_model": { + "model": "us.anthropic.claude-opus-5-5", + "region": "us-east-1" + }, "supports_tool_search": true, "max_input_tokens": 1000000, "max_output_tokens": 128000, @@ -60231,6 +52514,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -60519,6 +52803,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -60614,7 +52899,7 @@ "cache_creation_input_token_cost_above_200k_tokens": 0.0000075, "cache_read_input_token_cost_above_200k_tokens": 6e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "max_input_tokens": 200000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", @@ -60647,6 +52932,7 @@ "cache_creation_input_token_cost_above_200k_tokens": 0.00000825, "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.0000132, "cache_read_input_token_cost_above_200k_tokens": 6.6e-7, + "deprecation_date": "2027-04-08", "litellm_provider": "bedrock_converse", "supports_tool_search": true, "max_input_tokens": 200000, @@ -60694,7 +52980,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -60742,6 +53028,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -60756,6 +53043,10 @@ "cache_read_input_token_cost": 1.1e-7, "input_cost_per_token": 0.0000022, "litellm_provider": "bedrock_converse", + "token_count_model": { + "model": "us.anthropic.claude-opus-5-5", + "region": "us-east-1" + }, "supports_tool_search": true, "max_input_tokens": 1000000, "max_output_tokens": 128000, @@ -60781,6 +53072,7 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_native_structured_output": false, + "supports_bedrock_native_structured_output": true, "supports_max_reasoning_effort": true, "supports_output_config": true, "bedrock_output_config_effort_ceiling": "xhigh", @@ -60806,12 +53098,19 @@ }, "us.deepseek.v3.2": { "input_cost_per_token": 6.2e-7, + "input_cost_per_token_batches": 3.1e-7, + "input_cost_per_token_flex": 3.1e-7, + "input_cost_per_token_priority": 0.000001085, "litellm_provider": "bedrock_converse", "max_input_tokens": 163840, "max_output_tokens": 163840, "max_tokens": 163840, "mode": "chat", "output_cost_per_token": 0.00000185, + "output_cost_per_token_batches": 9.25e-7, + "output_cost_per_token_flex": 9.25e-7, + "output_cost_per_token_priority": 0.0000032375, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -60821,6 +53120,7 @@ "supports_vision": false }, "us.meta.llama3-1-405b-instruct-v1:0": { + "deprecation_date": "2026-07-07", "input_cost_per_token": 0.0000024, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -60833,29 +53133,37 @@ "supports_tool_choice": false }, "us.meta.llama3-1-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 7.2e-7, + "input_cost_per_token_batches": 3.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, - "max_output_tokens": 2048, - "max_tokens": 2048, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 7.2e-7, - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", + "output_cost_per_token_batches": 3.6e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-1-8b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 2.2e-7, + "input_cost_per_token_batches": 1.1e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, - "max_output_tokens": 2048, - "max_tokens": 2048, + "max_output_tokens": 4096, + "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 2.2e-7, + "output_cost_per_token_batches": 1.1e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-2-11b-instruct-v1:0": { + "deprecation_date": "2026-07-07", "input_cost_per_token": 1.6e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -60869,6 +53177,7 @@ "supports_vision": true }, "us.meta.llama3-2-1b-instruct-v1:0": { + "deprecation_date": "2026-07-07", "input_cost_per_token": 1e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -60876,10 +53185,12 @@ "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 1e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-2-1b-instruct.md", "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-2-3b-instruct-v1:0": { + "deprecation_date": "2026-07-07", "input_cost_per_token": 1.5e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -60887,10 +53198,12 @@ "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 1.5e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-2-3b-instruct.md", "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-2-90b-instruct-v1:0": { + "deprecation_date": "2026-07-07", "input_cost_per_token": 7.2e-7, "litellm_provider": "bedrock", "max_input_tokens": 128000, @@ -60904,26 +53217,32 @@ "supports_vision": true }, "us.meta.llama3-3-70b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 7.2e-7, + "input_cost_per_token_batches": 3.6e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 7.2e-7, + "output_cost_per_token_batches": 3.6e-7, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama4-maverick-17b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 2.4e-7, "input_cost_per_token_batches": 1.2e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, + "max_input_tokens": 1000000, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 9.7e-7, "output_cost_per_token_batches": 4.85e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-4-maverick-17b-instruct.md", "supported_modalities": [ "text", "image" @@ -60936,15 +53255,17 @@ "supports_tool_choice": false }, "us.meta.llama4-scout-17b-instruct-v1:0": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 1.7e-7, "input_cost_per_token_batches": 8.5e-8, "litellm_provider": "bedrock_converse", - "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, + "max_input_tokens": 10000000, + "max_output_tokens": 8192, + "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 6.6e-7, "output_cost_per_token_batches": 3.3e-7, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-4-scout-17b-instruct.md", "supported_modalities": [ "text", "image" @@ -60960,11 +53281,11 @@ "input_cost_per_token": 0.000002, "litellm_provider": "bedrock_converse", "max_input_tokens": 128000, - "max_output_tokens": 4096, - "max_tokens": 4096, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", "output_cost_per_token": 0.000006, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-pixtral-large.md", "supports_function_calling": true, "supports_tool_choice": false }, @@ -60980,6 +53301,8 @@ "input_cost_per_token_flex": 0.00000165, "input_cost_per_token_priority": 0.000005775, "litellm_provider": "bedrock_converse", + "token_count_framing": 64, + "token_count_tokenizer": "kimi_k3", "max_input_tokens": 1000000, "max_output_tokens": 131072, "max_tokens": 131072, @@ -61006,7 +53329,8 @@ "output_cost_per_token": 0.0000165, "output_cost_per_token_above_272k_tokens": 0.00002475, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", @@ -61037,7 +53361,8 @@ "output_cost_per_token": 0.000033, "output_cost_per_token_above_272k_tokens": 0.0000495, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", @@ -61071,11 +53396,12 @@ "output_cost_per_token": 0.00000132, "output_cost_per_token_above_272k_tokens": 0.00000198, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-luna.html", "supported_modalities": [ "text", "image" @@ -61107,11 +53433,12 @@ "output_cost_per_token": 0.000022, "output_cost_per_token_above_272k_tokens": 0.000033, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-sol.html", "supported_modalities": [ "text", "image" @@ -61143,11 +53470,12 @@ "output_cost_per_token": 0.0000132, "output_cost_per_token_above_272k_tokens": 0.0000198, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-terra.html", "supported_modalities": [ "text", "image" @@ -61185,6 +53513,7 @@ "output_cost_per_token": 0.000055, "output_cost_per_token_above_272k_tokens": 0.0000825, "litellm_provider": "bedrock_converse", + "supports_bedrock_converse_tool_result_images": false, "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, @@ -61225,6 +53554,7 @@ "output_cost_per_token": 5.5e-7, "output_cost_per_token_above_272k_tokens": 8.25e-7, "litellm_provider": "bedrock_converse", + "supports_bedrock_converse_tool_result_images": false, "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, @@ -61239,7 +53569,7 @@ "supports_function_calling": true, "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, - "supports_none_reasoning_effort": false, + "supports_none_reasoning_effort": true, "supports_tool_choice": true, "supports_prompt_caching": true, "supports_reasoning": true, @@ -61263,6 +53593,7 @@ "output_cost_per_token": 0.000011, "output_cost_per_token_above_272k_tokens": 0.0000165, "litellm_provider": "bedrock_converse", + "supports_bedrock_converse_tool_result_images": false, "max_input_tokens": 1050000, "max_output_tokens": 128000, "max_tokens": 128000, @@ -61277,7 +53608,7 @@ "supports_function_calling": true, "supports_max_reasoning_effort": true, "supports_minimal_reasoning_effort": false, - "supports_none_reasoning_effort": false, + "supports_none_reasoning_effort": true, "supports_tool_choice": true, "supports_prompt_caching": true, "supports_reasoning": true, @@ -61293,17 +53624,26 @@ "us.openai.gpt-6.1-sol": { "cache_creation_input_token_cost": 0.00000275, "cache_creation_input_token_cost_above_272k_tokens": 0.0000055, + "cache_creation_input_token_cost_above_272k_tokens_ultrafast": 0.000033, + "cache_creation_input_token_cost_ultrafast": 0.0000165, "cache_read_input_token_cost": 1.1e-7, "cache_read_input_token_cost_above_272k_tokens": 2.2e-7, + "cache_read_input_token_cost_above_272k_tokens_ultrafast": 0.00000132, + "cache_read_input_token_cost_ultrafast": 6.6e-7, "input_cost_per_token": 0.0000022, "input_cost_per_token_above_272k_tokens": 0.0000044, + "input_cost_per_token_above_272k_tokens_ultrafast": 0.0000264, + "input_cost_per_token_ultrafast": 0.0000132, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1050000, + "supports_bedrock_converse_tool_result_images": false, + "max_input_tokens": 1000000, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", "output_cost_per_token": 0.000011, "output_cost_per_token_above_272k_tokens": 0.0000165, + "output_cost_per_token_above_272k_tokens_ultrafast": 0.000099, + "output_cost_per_token_ultrafast": 0.000066, "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-sol.html", "supported_endpoints": [ "/v1/chat/completions", @@ -61335,6 +53675,17 @@ "mode": "chat", "supports_video_input": true }, + "us.twelvelabs.pegasus-1-5-v1:0": { + "input_cost_per_video_per_second": 0.00049, + "litellm_provider": "bedrock", + "max_input_tokens": 261120, + "max_output_tokens": 98304, + "max_tokens": 98304, + "mode": "chat", + "output_cost_per_token": 0.0000075, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-twelvelabs-pegasus-v1-5.html", + "supports_video_input": true + }, "us.writer.palmyra-x4-v1:0": { "input_cost_per_token": 0.0000025, "litellm_provider": "bedrock_converse", @@ -61350,12 +53701,12 @@ "us.writer.palmyra-x5-v1:0": { "input_cost_per_token": 6e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "max_input_tokens": 128000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.000006, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-writer-palmyra-x5.md", "supports_function_calling": true, "supports_pdf_input": true }, @@ -61389,6 +53740,11 @@ }, "us.xai.grok-4.7": { "supports_regex_lookaround": false, + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_bedrock_runtime_chat_completions_tools_with_reasoning": true, + "supports_bedrock_runtime_chat_completions_response_format": true, "cache_read_input_token_cost": 5.5e-7, "cache_read_input_token_cost_flex": 2.75e-7, "cache_read_input_token_cost_priority": 9.625e-7, @@ -61396,14 +53752,14 @@ "input_cost_per_token_flex": 0.0000011, "input_cost_per_token_priority": 0.00000385, "litellm_provider": "bedrock_converse", - "max_input_tokens": 500000, - "max_output_tokens": 500000, - "max_tokens": 500000, + "max_input_tokens": 524288, + "max_output_tokens": 524288, + "max_tokens": 524288, "mode": "chat", "output_cost_per_token": 0.0000066, "output_cost_per_token_flex": 0.0000033, "output_cost_per_token_priority": 0.00001155, - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-xai-grok-4-7.html", "supports_function_calling": true, "supports_prompt_caching": false, "supports_reasoning": true, @@ -61829,7 +54185,7 @@ "max_tokens": 64000, "mode": "chat", "output_cost_per_token": 0.000025, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_prompt_caching": true, @@ -63656,7 +56012,7 @@ "mode": "chat", "output_cost_per_token": 0.000015, "output_cost_per_token_batches": 0.0000075, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -63693,7 +56049,7 @@ "mode": "chat", "output_cost_per_token": 0.000015, "output_cost_per_token_batches": 0.0000075, - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, "supports_pdf_input": true, @@ -63955,7 +56311,7 @@ "supported_regions": [ "us-central1" ], - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, @@ -63999,12 +56355,33 @@ "supported_regions": [ "global" ], - "supports_assistant_prefill": true, + "supports_assistant_prefill": false, "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, + "vertex_ai/deepseek-ai/deepseek-v4.1-flash-maas": { + "cache_read_input_token_cost": 3e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "vertex_ai-deepseek_models", + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", + "supported_regions": [ + "global" + ], + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "vertex_ai/gemini-2.0-flash": { "deprecation_date": "2026-06-01", "input_cost_per_audio_token": 0.000001, @@ -64084,15 +56461,29 @@ "vertex_ai/gemini-3-flash-preview": { "cache_read_input_audio_token_cost": 1e-7, "cache_read_input_token_cost": 5e-8, - "cache_read_input_token_cost_flex": 5e-8, - "input_cost_per_token": 5e-7, + "cache_read_input_token_cost_priority": 9e-8, + "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_audio_token": 0.000001, + "input_cost_per_audio_token_batches": 5e-7, + "input_cost_per_audio_token_priority": 0.0000018, + "input_cost_per_token": 5e-7, + "input_cost_per_token_batches": 2.5e-7, + "input_cost_per_token_flex": 2.5e-7, + "input_cost_per_token_priority": 9e-7, "litellm_provider": "vertex_ai", "max_input_tokens": 1048576, "max_output_tokens": 65535, "max_tokens": 65535, "mode": "chat", "output_cost_per_token": 0.000003, + "output_cost_per_token_batches": 0.0000015, + "output_cost_per_token_flex": 0.0000015, + "output_cost_per_token_priority": 0.0000054, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", @@ -64110,6 +56501,7 @@ ], "supports_audio_input": true, "supports_function_calling": true, + "supports_native_streaming": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, @@ -64119,23 +56511,7 @@ "supports_video_input": true, "supports_vision": true, "supports_web_search": true, - "supports_native_streaming": true, - "input_cost_per_token_priority": 9e-7, - "input_cost_per_audio_token_priority": 0.0000018, - "output_cost_per_token_priority": 0.0000054, - "cache_read_input_token_cost_priority": 9e-8, - "search_context_cost_per_query": { - "search_context_size_low": 0.014, - "search_context_size_medium": 0.014, - "search_context_size_high": 0.014 - }, - "web_search_billing_unit": "per_query", - "google_maps_grounding_cost_per_query": 0.014, - "input_cost_per_audio_token_batches": 5e-7, - "input_cost_per_token_batches": 2.5e-7, - "input_cost_per_token_flex": 2.5e-7, - "output_cost_per_token_batches": 0.0000015, - "output_cost_per_token_flex": 0.0000015 + "web_search_billing_unit": "per_query" }, "vertex_ai/gemini-3-pro-image": { "cache_read_input_token_cost": 2e-7, @@ -64478,22 +56854,36 @@ "google_maps_grounding_cost_per_query": 0.014 }, "vertex_ai/gemini-3.1-pro-preview": { - "prompt_cache_min_tokens": 4096, + "cache_creation_input_token_cost_above_200k_tokens": 2.5e-7, "cache_read_input_token_cost": 2e-7, "cache_read_input_token_cost_above_200k_tokens": 4e-7, - "cache_creation_input_token_cost_above_200k_tokens": 2.5e-7, + "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-7, + "cache_read_input_token_cost_priority": 3.6e-7, + "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token": 0.000002, "input_cost_per_token_above_200k_tokens": 0.000004, + "input_cost_per_token_above_200k_tokens_priority": 0.0000072, "input_cost_per_token_batches": 0.000001, + "input_cost_per_token_flex": 0.000001, + "input_cost_per_token_priority": 0.0000036, "litellm_provider": "vertex_ai", "max_input_tokens": 1048576, "max_output_tokens": 65536, "max_tokens": 65536, "mode": "chat", + "output_cost_per_image": 0.00012, "output_cost_per_token": 0.000012, "output_cost_per_token_above_200k_tokens": 0.000018, + "output_cost_per_token_above_200k_tokens_priority": 0.0000324, "output_cost_per_token_batches": 0.000006, - "output_cost_per_image": 0.00012, + "output_cost_per_token_flex": 0.000006, + "output_cost_per_token_priority": 0.0000216, + "prompt_cache_min_tokens": 4096, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", @@ -64511,33 +56901,18 @@ ], "supports_audio_input": true, "supports_function_calling": true, + "supports_native_streaming": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, + "supports_url_context": true, "supports_video_input": true, "supports_vision": true, "supports_web_search": true, - "supports_url_context": true, - "supports_native_streaming": true, - "input_cost_per_token_priority": 0.0000036, - "input_cost_per_token_above_200k_tokens_priority": 0.0000072, - "output_cost_per_token_priority": 0.0000216, - "output_cost_per_token_above_200k_tokens_priority": 0.0000324, - "cache_read_input_token_cost_priority": 3.6e-7, - "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-7, - "cache_read_input_token_cost_flex": 2e-7, - "search_context_cost_per_query": { - "search_context_size_low": 0.014, - "search_context_size_medium": 0.014, - "search_context_size_high": 0.014 - }, - "web_search_billing_unit": "per_query", - "google_maps_grounding_cost_per_query": 0.014, - "input_cost_per_token_flex": 0.000001, - "output_cost_per_token_flex": 0.000006 + "web_search_billing_unit": "per_query" }, "vertex_ai/gemini-3.1-pro-preview-customtools": { "prompt_cache_min_tokens": 4096, @@ -64828,7 +57203,67 @@ "input_cost_per_token_priority": 0.00000135, "output_cost_per_token_priority": 0.00000675, "cache_read_input_token_cost_priority": 1.35e-7, - "deprecation_date": "2027-01-28", + "deprecation_date": "2027-01-28", + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "web_search_billing_unit": "per_query", + "google_maps_grounding_cost_per_query": 0.014 + }, + "vertex_ai/gemini-3.8-flash": { + "prompt_cache_min_tokens": 4096, + "deprecation_date": "2027-09-02", + "cache_read_input_token_cost": 7.5e-8, + "cache_read_input_token_cost_batches": 3.75e-8, + "cache_read_input_token_cost_flex": 3.75e-8, + "input_cost_per_token": 7.5e-7, + "input_cost_per_token_batches": 3.75e-7, + "input_cost_per_token_flex": 3.75e-7, + "litellm_provider": "vertex_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 65536, + "max_tokens": 65536, + "mode": "chat", + "output_cost_per_reasoning_token": 0.00000375, + "output_cost_per_token": 0.00000375, + "output_cost_per_token_batches": 0.000001875, + "output_cost_per_token_flex": 0.000001875, + "regional_endpoint_uplift_multiplier": 1.1, + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_minimal_reasoning_effort": false, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "supports_native_streaming": true, + "input_cost_per_token_priority": 0.00000135, + "output_cost_per_token_priority": 0.00000675, + "cache_read_input_token_cost_priority": 1.35e-7, "search_context_cost_per_query": { "search_context_size_low": 0.014, "search_context_size_medium": 0.014, @@ -64837,24 +57272,86 @@ "web_search_billing_unit": "per_query", "google_maps_grounding_cost_per_query": 0.014 }, - "vertex_ai/gemini-3.8-flash": { - "prompt_cache_min_tokens": 4096, - "cache_read_input_token_cost": 7.5e-8, - "cache_read_input_token_cost_batches": 3.75e-8, - "cache_read_input_token_cost_flex": 3.75e-8, - "input_cost_per_token": 7.5e-7, - "input_cost_per_token_batches": 3.75e-7, - "input_cost_per_token_flex": 3.75e-7, + "vertex_ai/gemini-3.8-flash-cyber": { + "cache_read_input_token_cost": 1.5e-7, + "cache_read_input_token_cost_batches": 7.5e-8, + "cache_read_input_token_cost_flex": 7.5e-8, + "cache_read_input_token_cost_priority": 2.7e-7, + "input_cost_per_token": 0.0000015, + "input_cost_per_token_batches": 7.5e-7, + "input_cost_per_token_flex": 7.5e-7, + "input_cost_per_token_priority": 0.0000027, "litellm_provider": "vertex_ai", "max_input_tokens": 1048576, "max_output_tokens": 65536, "max_tokens": 65536, "mode": "chat", - "output_cost_per_reasoning_token": 0.00000375, - "output_cost_per_token": 0.00000375, - "output_cost_per_token_batches": 0.000001875, - "output_cost_per_token_flex": 0.000001875, + "output_cost_per_reasoning_token": 0.0000075, + "output_cost_per_token": 0.0000075, + "output_cost_per_token_batches": 0.00000375, + "output_cost_per_token_flex": 0.00000375, + "output_cost_per_token_priority": 0.0000135, + "regional_endpoint_uplift_multiplier": 1.1, + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_function_calling": false, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_minimal_reasoning_effort": false, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": false, + "supports_url_context": false, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": false, + "supports_native_streaming": true + }, + "vertex_ai/gemini-4-argon": { + "deprecation_date": "2027-10-13", + "cache_read_input_token_cost": 1e-7, + "cache_read_input_token_cost_batches": 5e-8, + "cache_read_input_token_cost_flex": 5e-8, + "cache_read_input_token_cost_priority": 1.8e-7, + "google_maps_grounding_cost_per_query": 0.014, + "input_cost_per_audio_token": 0.000002, + "input_cost_per_audio_token_batches": 0.000001, + "input_cost_per_audio_token_priority": 0.0000036, + "input_cost_per_token": 0.000002, + "input_cost_per_token_batches": 0.000001, + "input_cost_per_token_flex": 0.000001, + "input_cost_per_token_priority": 0.0000036, + "litellm_provider": "vertex_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "max_tokens": 1048576, + "mode": "chat", + "output_cost_per_reasoning_token": 0.00001, + "output_cost_per_token": 0.00001, + "output_cost_per_token_batches": 0.000005, + "output_cost_per_token_flex": 0.000005, + "output_cost_per_token_priority": 0.000018, + "prompt_cache_min_tokens": 6144, "regional_endpoint_uplift_multiplier": 1.1, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", @@ -64872,11 +57369,12 @@ ], "supports_audio_input": true, "supports_function_calling": true, + "supports_minimal_reasoning_effort": false, + "supports_native_streaming": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_minimal_reasoning_effort": false, "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, @@ -64884,42 +57382,43 @@ "supports_video_input": true, "supports_vision": true, "supports_web_search": true, - "supports_native_streaming": true, - "input_cost_per_token_priority": 0.00000135, - "output_cost_per_token_priority": 0.00000675, - "cache_read_input_token_cost_priority": 1.35e-7, - "search_context_cost_per_query": { - "search_context_size_low": 0.014, - "search_context_size_medium": 0.014, - "search_context_size_high": 0.014 - }, - "web_search_billing_unit": "per_query", - "google_maps_grounding_cost_per_query": 0.014 + "web_search_billing_unit": "per_query" }, - "vertex_ai/gemini-3.8-flash-cyber": { - "cache_read_input_token_cost": 1.5e-7, - "cache_read_input_token_cost_batches": 7.5e-8, - "cache_read_input_token_cost_flex": 7.5e-8, - "cache_read_input_token_cost_priority": 2.7e-7, - "input_cost_per_token": 0.0000015, - "input_cost_per_token_batches": 7.5e-7, - "input_cost_per_token_flex": 7.5e-7, - "input_cost_per_token_priority": 0.0000027, + "vertex_ai/gemini-4-argon-cyber-permissive": { + "cache_read_input_token_cost": 2e-7, + "cache_read_input_token_cost_batches": 1e-7, + "cache_read_input_token_cost_flex": 1e-7, + "cache_read_input_token_cost_priority": 3.6e-7, + "google_maps_grounding_cost_per_query": 0.014, + "input_cost_per_audio_token": 0.000004, + "input_cost_per_audio_token_batches": 0.000002, + "input_cost_per_audio_token_priority": 0.0000072, + "input_cost_per_token": 0.000004, + "input_cost_per_token_batches": 0.000002, + "input_cost_per_token_flex": 0.000002, + "input_cost_per_token_priority": 0.0000072, "litellm_provider": "vertex_ai", "max_input_tokens": 1048576, - "max_output_tokens": 65536, - "max_tokens": 65536, + "max_output_tokens": 1048576, + "max_tokens": 1048576, "mode": "chat", - "output_cost_per_reasoning_token": 0.0000075, - "output_cost_per_token": 0.0000075, - "output_cost_per_token_batches": 0.00000375, - "output_cost_per_token_flex": 0.00000375, - "output_cost_per_token_priority": 0.0000135, + "output_cost_per_reasoning_token": 0.00002, + "output_cost_per_token": 0.00002, + "output_cost_per_token_batches": 0.00001, + "output_cost_per_token_flex": 0.00001, + "output_cost_per_token_priority": 0.000036, + "prompt_cache_min_tokens": 6144, "regional_endpoint_uplift_multiplier": 1.1, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", - "/v1/completions" + "/v1/completions", + "/v1/batch" ], "supported_modalities": [ "text", @@ -64931,27 +57430,31 @@ "text" ], "supports_audio_input": true, - "supports_function_calling": false, + "supports_function_calling": true, + "supports_minimal_reasoning_effort": false, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_minimal_reasoning_effort": false, "supports_response_schema": true, "supports_system_messages": true, - "supports_tool_choice": false, - "supports_url_context": false, + "supports_tool_choice": true, + "supports_url_context": true, "supports_video_input": true, "supports_vision": true, - "supports_web_search": false, - "supports_native_streaming": true + "supports_web_search": true, + "web_search_billing_unit": "per_query" }, "vertex_ai/gemini-nano-banana-2.1": { "cache_read_input_token_cost": 1.5e-7, "cache_read_input_token_cost_batches": 7.5e-8, + "cache_read_input_token_cost_flex": 7.5e-8, "cache_read_input_token_cost_priority": 2.7e-7, "input_cost_per_image": 0.00168, "input_cost_per_token": 0.0000015, "input_cost_per_token_batches": 7.5e-7, + "input_cost_per_token_flex": 7.5e-7, "input_cost_per_token_priority": 0.0000027, "litellm_provider": "vertex_ai-language-models", "max_input_tokens": 131072, @@ -64962,6 +57465,7 @@ "output_cost_per_image_token": 0.00003, "output_cost_per_token": 0.0000075, "output_cost_per_token_batches": 0.00000375, + "output_cost_per_token_flex": 0.00000375, "output_cost_per_token_priority": 0.0000135, "search_context_cost_per_query": { "search_context_size_low": 0.014, @@ -65030,6 +57534,56 @@ "supports_video_input": true, "supports_vision": true }, + "vertex_ai/gemini-robotics-er-2-preview": { + "cache_read_input_token_cost": 1e-7, + "cache_read_input_token_cost_batches": 5e-8, + "input_cost_per_audio_token": 0.000001, + "input_cost_per_token": 0.000001, + "input_cost_per_token_batches": 5e-7, + "litellm_provider": "vertex_ai", + "max_input_tokens": 131072, + "max_output_tokens": 65536, + "max_tokens": 65536, + "mode": "chat", + "output_cost_per_reasoning_token": 0.000005, + "output_cost_per_token": 0.000005, + "output_cost_per_token_batches": 0.0000025, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "source": "https://ai.google.dev/gemini-api/docs/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_audio_output": false, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "web_search_billing_unit": "per_query" + }, "vertex_ai/gemma-4-26b-a4b-it": { "cache_read_input_token_cost": 1.5e-8, "input_cost_per_token": 1.5e-7, @@ -66291,6 +58845,22 @@ "supports_vision": false, "source": "https://wandb.ai/site/pricing/tokens/" }, + "wandb/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B": { + "cache_read_input_token_cost": 2e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "wandb", + "max_input_tokens": 1049000, + "max_tokens": 943718, + "mode": "chat", + "output_cost_per_token": 7.5e-7, + "source": "https://openrouter.ai/api/v1/models/nvidia/nemotron-3.5-super-vl/endpoints", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "wandb/openai/gpt-oss-120b": { "supports_reasoning": true, "max_tokens": 131072, @@ -66723,13 +59293,19 @@ }, "writer.palmyra-vision-7b": { "input_cost_per_token": 1.5e-7, + "input_cost_per_token_batches": 7.5e-8, + "input_cost_per_token_flex": 7.5e-8, + "input_cost_per_token_priority": 2.625e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 4096, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", "output_cost_per_token": 6e-7, - "source": "https://aws.amazon.com/bedrock/pricing/", + "output_cost_per_token_batches": 3e-7, + "output_cost_per_token_flex": 3e-7, + "output_cost_per_token_priority": 0.00000105, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_vision": true }, "writer.palmyra-x4-v1:0": { @@ -66747,38 +59323,15 @@ "writer.palmyra-x5-v1:0": { "input_cost_per_token": 6e-7, "litellm_provider": "bedrock_converse", - "max_input_tokens": 1000000, + "max_input_tokens": 128000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", "output_cost_per_token": 0.000006, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-writer-palmyra-x5.md", "supports_function_calling": true, "supports_pdf_input": true }, - "xai.grok-4.7": { - "supports_regex_lookaround": false, - "cache_read_input_token_cost": 5e-7, - "cache_read_input_token_cost_flex": 2.5e-7, - "cache_read_input_token_cost_priority": 8.75e-7, - "input_cost_per_token": 0.000002, - "input_cost_per_token_flex": 0.000001, - "input_cost_per_token_priority": 0.0000035, - "litellm_provider": "bedrock_converse", - "max_input_tokens": 500000, - "max_output_tokens": 500000, - "max_tokens": 500000, - "mode": "chat", - "output_cost_per_token": 0.000006, - "output_cost_per_token_flex": 0.000003, - "output_cost_per_token_priority": 0.0000105, - "source": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/bedrock/USD/current/bedrock.json", - "supports_function_calling": true, - "supports_prompt_caching": false, - "supports_reasoning": true, - "supports_tool_choice": true, - "supports_vision": true - }, "xai/grok-2": { "input_cost_per_token": 0.000002, "litellm_provider": "xai", @@ -67771,53 +60324,71 @@ "supports_vision": true }, "zai.glm-4.7": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 6e-7, + "input_cost_per_token_batches": 3e-7, + "input_cost_per_token_flex": 3e-7, + "input_cost_per_token_priority": 0.00000105, "litellm_provider": "bedrock_converse", "max_input_tokens": 203000, - "max_output_tokens": 4000, - "max_tokens": 4000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 0.0000022, + "output_cost_per_token_batches": 0.0000011, + "output_cost_per_token_flex": 0.0000011, + "output_cost_per_token_priority": 0.00000385, "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, "supports_tool_choice": true, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false }, "zai.glm-4.7-flash": { + "deprecation_date": "2027-03-30", "input_cost_per_token": 7e-8, + "input_cost_per_token_batches": 3.5e-8, + "input_cost_per_token_flex": 3.5e-8, + "input_cost_per_token_priority": 1.225e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 203000, - "max_output_tokens": 4000, - "max_tokens": 4000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 4e-7, + "output_cost_per_token_batches": 2e-7, + "output_cost_per_token_flex": 2e-7, + "output_cost_per_token_priority": 7e-7, "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, "supports_tool_choice": true, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false }, "zai.glm-5": { "input_cost_per_token": 0.000001, + "input_cost_per_token_batches": 5e-7, + "input_cost_per_token_priority": 0.00000175, "output_cost_per_token": 0.0000032, "litellm_provider": "bedrock_converse", "max_input_tokens": 200000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", + "output_cost_per_token_batches": 0.0000016, + "output_cost_per_token_priority": 0.0000056, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, "supports_system_messages": true, "supports_tool_choice": true, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json", "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false diff --git a/cecli/tools/_yield.py b/cecli/tools/_yield.py index feaf905962a..4d2341fadf6 100644 --- a/cecli/tools/_yield.py +++ b/cecli/tools/_yield.py @@ -231,6 +231,22 @@ async def execute(cls, coder, **kwargs): except Exception: logger.warning("Failed to reap finished sub-agents", exc_info=True) + # Creator agents may have changed extensions on disk; reload them so + # the new skills, hooks, workflows, tools, commands, and servers are + # available to the rest of the session immediately. + if coder.edit_format == "subagent": + try: + agent_service = AgentService.get_instance(coder) + if agent_service.get_agent_name(coder) == "creator": + from cecli.helpers.creator.utils import ( + reload_creator_extensions, + ) + + for message in await reload_creator_extensions(coder): + coder.io.tool_output(message) + except Exception as e: + logger.warning("Failed to reload creator extensions: %s", e) + coder.agent_finished = True # If this is a sub-agent, capture the summary for the parent diff --git a/cecli/tools/command.py b/cecli/tools/command.py index c51cad1b376..1eea71819c4 100644 --- a/cecli/tools/command.py +++ b/cecli/tools/command.py @@ -18,7 +18,7 @@ from cecli.helpers.background_commands import BackgroundCommandManager from cecli.run_cmd import run_cmd, run_cmd_subprocess from cecli.tools.utils.base_tool import BaseTool -from cecli.tools.utils.helpers import ToolError +from cecli.tools.utils.helpers import ToolError, is_creator_coder from cecli.tools.utils.output import color_markers, tool_footer, tool_header from cecli.tools.utils.responses import ToolResponse from cecli.tools.validations import ToolValidations @@ -214,7 +214,7 @@ async def execute( return response # Check for implicit background (trailing & on Linux) - if ".cecli/agents" in command: + if ".cecli/agents" in command and not is_creator_coder(coder): response.append_error( "Do not attempt to access internal files with " "standard cli tools. Please use the tools you have been provided." diff --git a/cecli/tools/grep.py b/cecli/tools/grep.py index d3d0233eb18..9082b4e4e4f 100644 --- a/cecli/tools/grep.py +++ b/cecli/tools/grep.py @@ -9,7 +9,7 @@ from cecli.helpers.hashline import strip_hashline from cecli.run_cmd import run_cmd_subprocess from cecli.tools.utils.base_tool import BaseTool -from cecli.tools.utils.helpers import ToolError +from cecli.tools.utils.helpers import ToolError, is_creator_coder from cecli.tools.utils.output import color_markers, tool_footer, tool_header from cecli.tools.utils.responses import ToolResponse from cecli.tools.validations import ToolValidations @@ -80,9 +80,11 @@ def _should_use_regex(pattern, requested): return bool(requested) or "|" in pattern -def _build_exclude_args(tool_name, cmd_args): +def _build_exclude_args(tool_name, cmd_args, exclude_cecli=True): """Add exclusion arguments for common build/artifact dirs and history files.""" for exclude_dir in DEFAULT_EXCLUDE_DIRS: + if exclude_dir == ".cecli" and not exclude_cecli: + continue if tool_name == "rg": cmd_args.extend(["-g", f"!{exclude_dir}"]) elif tool_name == "ag": @@ -253,7 +255,7 @@ def _extract_file_entries(abs_path, content): return entries -def _build_powershell_exclude_regex(exclude_dirs=None, exclude_files=None): +def _build_powershell_exclude_regex(exclude_dirs=None, exclude_files=None, exclude_cecli=True): """Build a PowerShell -notmatch regex that skips default build/artifact dirs. Also skips default history files (matched by basename suffix). @@ -264,6 +266,8 @@ def _build_powershell_exclude_regex(exclude_dirs=None, exclude_files=None): exclude_files = DEFAULT_EXCLUDE_FILES fragments = [] for entry in exclude_dirs: + if entry == ".cecli" and not exclude_cecli: + continue # Escape the literal text, then turn * globs into path-segment wildcards. frag = re.escape(entry).replace(r"\*", r"[^\\/]*") fragments.append(r"(?:\\|/)(?:" + frag + r")(?:\\|/|$)") @@ -284,6 +288,7 @@ def _build_powershell_search_script( context_before, context_after, count_only, + exclude_cecli=True, ): """Build a PowerShell pipeline that mimics rg/ag/grep via Select-String.""" @@ -296,7 +301,7 @@ def _ps_quote(value): if file_pattern != "*": parts.extend(["-Filter", _ps_quote(file_pattern)]) - exclude_regex = _build_powershell_exclude_regex() + exclude_regex = _build_powershell_exclude_regex(exclude_cecli=exclude_cecli) if exclude_regex: parts.extend( [ @@ -695,7 +700,12 @@ def execute( continue try: - search_dir_path = Path(repo.root) / directory + creator = is_creator_coder(coder) + expanded_directory = Path(directory).expanduser() + if creator and expanded_directory.is_absolute(): + search_dir_path = expanded_directory + else: + search_dir_path = Path(repo.root) / expanded_directory if tool_name == "powershell": # PowerShell (Select-String) backend: build a pipeline script and @@ -710,6 +720,7 @@ def execute( context_before=context_before, context_after=context_after, count_only=False, + exclude_cecli=not creator, ), tool_path, ) @@ -749,7 +760,7 @@ def execute( # Exclusions exclude_args = [] - _build_exclude_args(tool_name, exclude_args) + _build_exclude_args(tool_name, exclude_args, exclude_cecli=not creator) # --- PASS1: Build count command (fast, no context) --- count_cmd_parts = [tool_path] diff --git a/cecli/tools/ls.py b/cecli/tools/ls.py index 08fb1d3786e..a5ffa23c80d 100644 --- a/cecli/tools/ls.py +++ b/cecli/tools/ls.py @@ -1,7 +1,7 @@ import os from cecli.tools.utils.base_tool import BaseTool -from cecli.tools.utils.helpers import ToolError +from cecli.tools.utils.helpers import ToolError, is_creator_coder, is_path_within_root from cecli.tools.utils.output import color_markers, tool_footer, tool_header from cecli.tools.utils.responses import ToolResponse from cecli.tools.validations import ToolValidations @@ -47,10 +47,17 @@ def execute(cls, coder, path=None, **kwargs): try: # Create an absolute path from the provided relative path - abs_path = os.path.abspath(os.path.join(coder.root, dir_path)) + external_paths_allowed = is_creator_coder(coder) + expanded_path = os.path.expanduser(dir_path) + if os.path.isabs(expanded_path): + abs_path = os.path.abspath(expanded_path) + else: + abs_path = os.path.abspath(os.path.join(coder.root, expanded_path)) # Security check: ensure the resolved path is within the project root - if not abs_path.startswith(os.path.abspath(coder.root)): + # unless this is a creator agent, which may inspect config and + # extension directories anywhere on disk. + if not external_paths_allowed and not is_path_within_root(abs_path, coder.root): coder.io.tool_error( f"Error: Path '{dir_path}' attempts to access files outside the project root." ) @@ -70,7 +77,7 @@ def execute(cls, coder, path=None, **kwargs): try: with os.scandir(abs_path) as entries: for entry in entries: - if not entry.name.startswith("."): + if external_paths_allowed or not entry.name.startswith("."): contents.append(safe_relpath(entry.path, coder.root)) except OSError as e: coder.io.tool_error(f"Error listing directory '{dir_path}': {e}") diff --git a/cecli/tools/resource_manager.py b/cecli/tools/resource_manager.py index 3377d19eb65..84fbd68e2c6 100644 --- a/cecli/tools/resource_manager.py +++ b/cecli/tools/resource_manager.py @@ -447,9 +447,8 @@ def _view(cls, coder, file_path): def _create(cls, coder, file_path): """Create a new file on the file system and make it editable in the coder's context.""" try: - abs_path = coder.abs_root_path(file_path) + abs_path = coder.abs_root_path(os.path.expanduser(file_path)) - # Check if file already exists if os.path.exists(abs_path): coder.io.tool_output(f"⚠ File '{file_path}' already exists", type="tool-result") return f"File already exists: {file_path}" @@ -488,7 +487,7 @@ def _resolve_file_path(cls, coder, file_path): filename = parts[1] rel_path = coder.local_agent_folder(f"{command_key}/{filename}") return coder.abs_root_path(rel_path) - return coder.abs_root_path(file_path) + return coder.abs_root_path(os.path.expanduser(file_path)) @classmethod def _load_skill(cls, coder, skill_name): diff --git a/cecli/tools/utils/helpers.py b/cecli/tools/utils/helpers.py index c9b6e082178..efca476cdfe 100644 --- a/cecli/tools/utils/helpers.py +++ b/cecli/tools/utils/helpers.py @@ -29,9 +29,44 @@ def is_provided(value, *, treat_zero_as_missing=False): return True +def is_creator_coder(coder) -> bool: + """Return True for the creator sub-agent or agents allowed external paths. + + The creator frequently edits files outside the project root (global config, + user skills, hooks) and inside ``.cecli`` directories, so path guards must + not confine it to the project. + """ + agent_config = getattr(coder, "agent_config", {}) or {} + if agent_config.get("creator_mode") or agent_config.get("allow_external_paths"): + return True + + try: + from cecli.helpers.agents.service import AgentService + + return AgentService.get_instance(coder).get_agent_name(coder) == "creator" + except Exception: + return False + + +def is_path_within_root(path, root) -> bool: + """Return True when *path* resolves inside *root*. + + Uses ``commonpath`` so sibling paths sharing a string prefix (for example + ``/repo-other`` and ``/repo``) are not treated as contained. + """ + try: + common = os.path.commonpath( + [os.path.realpath(os.path.abspath(path)), os.path.realpath(os.path.abspath(root))] + ) + return common == os.path.realpath(os.path.abspath(root)) + except (ValueError, OSError): + return False + + def resolve_paths(coder, file_path): """Resolves absolute and relative paths for a given file path.""" try: + file_path = os.path.expanduser(file_path) abs_path = coder.abs_root_path(file_path) rel_path = coder.get_rel_fname(abs_path) return abs_path, rel_path diff --git a/cecli/tui/app.py b/cecli/tui/app.py index 79ea0175e71..5e3e29ca9e6 100644 --- a/cecli/tui/app.py +++ b/cecli/tui/app.py @@ -15,7 +15,9 @@ from rich.style import Style from textual import events from textual.app import App, ComposeResult +from textual.containers import Horizontal, Vertical from textual.theme import Theme +from textual.widgets import Static from cecli import __version__ from cecli.decoding import safe_open @@ -29,6 +31,7 @@ from .widgets import ( CompletionBar, FileList, + FloatingPanels, InputArea, InputContainer, KeyHints, @@ -211,6 +214,13 @@ def __init__(self, coder_worker, output_queue, input_queue, args): show=True, ) + self.bind( + self._encode_keys(self.get_keys_for("panels")), + "toggle_floating_panels", + description="Toggle Panels", + show=False, + ) + self.register_theme(BASE_THEME) self.theme = "cecli" @@ -294,6 +304,7 @@ def _get_config(self): "history": "alt+shift+h", "voice": "ctrl+r", "paste": "alt+v", + "panels": "alt+p", "focus": "ctrl+f", "cancel": "ctrl+c", "clear": "ctrl+l", @@ -365,13 +376,21 @@ def compose(self) -> ComposeResult: # Git info loaded in on_mount to avoid blocking startup yield OutputContainer(id="output") yield StatusBar(id="status-bar") - yield InputContainer( - InputArea(history_file=history_file, id="input"), - FileList(id="file-list", classes="empty"), - id="input-container", - coder_mode=coder_mode, - ) - yield KeyHints(id="key-hints") + with Horizontal(id="input-row"): + with Vertical(id="input-column"): + yield InputContainer( + InputArea(history_file=history_file, id="input"), + FileList(id="file-list", classes="empty"), + id="input-container", + coder_mode=coder_mode, + ) + yield KeyHints(id="key-hints") + with Vertical(id="extensions-column"): + yield FloatingPanels(id="floating-panels") + yield Static( + f"{self.get_keys_for('panels')} to toggle", + id="extensions-hint", + ) yield MainFooter( model_name=model_name, project_name=project_name, @@ -928,6 +947,12 @@ def on_input_area_submit(self, message: InputArea.Submit): self._handle_spawn_agent_command(user_input, stripped) return + # Intercept /creator to spawn the creator sub-agent immediately, even + # while the primary coder is generating. + if stripped == "/creator" or stripped.startswith("/creator "): + self._handle_creator_command(user_input, stripped) + return + # Intercept /workspace to open an already-registered workspace # sub-agent immediately, mirroring /spawn-agent. if stripped == "/workspace" or stripped.startswith("/workspace "): @@ -1160,6 +1185,29 @@ async def _run_spawn(): self.worker.loop.call_soon_threadsafe(lambda: self.worker.loop.create_task(_run_spawn())) + def _handle_creator_command(self, user_input: str, stripped: str) -> None: + """Dispatch /creator immediately without a full generation cycle.""" + from cecli.commands.creator import CreatorCommand + + parts = stripped.split(maxsplit=1) + creator_args = parts[1].strip() if len(parts) > 1 else "" + + input_area = self.query_one("#input", InputArea) + input_area.value = "" + + input_area.save_to_history(user_input) + self.add_user_message(user_input) + + coder = self.worker.coder + if self.worker.loop is None: + self.show_error("Worker loop not available. Cannot spawn creator sub-agent.") + return + + async def _run_creator(): + await CreatorCommand.execute(coder.io, coder, creator_args) + + self.worker.loop.call_soon_threadsafe(lambda: self.worker.loop.create_task(_run_creator())) + def _handle_workspace_command(self, user_input: str, stripped: str) -> None: """Dispatch /workspace immediately without a full generation cycle. @@ -1273,6 +1321,13 @@ def action_quit(self): # Delay exit to allow status bar to render self.set_timer(0.3, self._do_quit) + def action_toggle_floating_panels(self): + """Show or hide the floating status panels.""" + try: + self.query_one(FloatingPanels).toggle_visible() + except Exception: + pass + def action_noop(self): pass diff --git a/cecli/tui/styles.tcss b/cecli/tui/styles.tcss index 72c689bc724..b69ed096671 100644 --- a/cecli/tui/styles.tcss +++ b/cecli/tui/styles.tcss @@ -23,6 +23,29 @@ Screen { border: none; } +/* Input row: input container plus the floating status panels */ +#input-row { + height: auto; + width: 100%; + padding: 0; + margin: 0; + align-vertical: top; +} + +#input-column { + width: 1fr; + height: auto; + margin: 0; + padding: 0; +} + +#extensions-column { + width: 43; + height: auto; + margin: 0; + padding: 0; +} + /* Output area */ #output, .subagent-output { height: 1fr; @@ -36,7 +59,8 @@ Screen { /* Input area - floating card style */ #input-container { height: auto; - max-height: 33%; + width: 1fr; + max-height: 33vh; min-height: 0; padding: 0 0 0 1; margin: 0 1 0 1; @@ -100,13 +124,22 @@ TextArea > .text-area--selection { display: none; } -/* Key hints below input */ +/* Key hints sit inside the same column as their box */ #key-hints { color: $secondary; height: 1; - width: 100%; - padding: 0 2 0 2; - margin: 0 0 1 0; + width: 1fr; + padding: 0 1 0 1; + margin: 0 1 1 1; +} + +#extensions-hint { + width: 1fr; + height: 1; + padding: 0 1 0 1; + margin: 0 1 1 0; + color: $secondary; + text-align: right; } #key-hints > .key-hints-left { @@ -128,4 +161,3 @@ TextArea > .text-area--selection { color: $accent; padding: 0 1; } - diff --git a/cecli/tui/widgets/__init__.py b/cecli/tui/widgets/__init__.py index 8e8c2db6288..09d2e795bb3 100644 --- a/cecli/tui/widgets/__init__.py +++ b/cecli/tui/widgets/__init__.py @@ -2,6 +2,7 @@ from .completion_bar import CompletionBar from .file_list import FileList +from .floating_panels import FloatingPanels from .footer import MainFooter from .input_area import InputArea from .input_container import InputContainer @@ -19,5 +20,6 @@ "OutputContainer", "StatusBar", "FileList", + "FloatingPanels", "SubAgentPills", ] diff --git a/cecli/tui/widgets/floating_panels.py b/cecli/tui/widgets/floating_panels.py new file mode 100644 index 00000000000..5720a9f36f1 --- /dev/null +++ b/cecli/tui/widgets/floating_panels.py @@ -0,0 +1,395 @@ +"""Extensions box showing sub-agents, skills, and MCP servers.""" + +from textual.containers import Horizontal, Vertical, VerticalScroll +from textual.message import Message +from textual.reactive import reactive +from textual.widgets import Static + + +class StatusItemClicked(Message): + """Posted when a status item is clicked.""" + + def __init__(self, category: str, name: str, active: bool): + super().__init__() + self.category = category + self.name = name + self.active = active + + +class StatusItem(Static): + """A clickable item in the extensions box.""" + + def __init__(self, display_name, item_name, category, active, **kwargs): + classes = "status-item active" if active else "status-item inactive" + super().__init__( + f"{'●' if active else '○'} {display_name}", + markup=False, + classes=classes, + **kwargs, + ) + self.item_name = item_name + self.item_category = category + self.item_active = active + + def on_click(self, event): + self.post_message(StatusItemClicked(self.item_category, self.item_name, self.item_active)) + event.stop() + + +class StatusColumn(Vertical): + """A reactive vertical column of clickable status items.""" + + items = reactive([], recompose=True) + + def __init__(self, category: str, **kwargs): + super().__init__(**kwargs) + self.category = category + + def compose(self): + if not self.items: + yield Static("(none)", classes="status-empty") + return + + for display_name, name, active in self.items: + yield StatusItem(display_name, name, self.category, active) + + +class FloatingPanels(Vertical): + """Single box showing sub-agents, skills, and MCP servers. + + Every item is clickable: sub-agents spawn, skills load/unload, and MCP + servers load/unload. The three categories sit in equal-width columns + beneath labels embedded in the top border, and share one scroll area. + """ + + MAX_NAME_WIDTH = 10 + + DEFAULT_CSS = """ + FloatingPanels { + height: auto; + width: 42; + border: round $accent 50%; + border-title-align: center; + border-title-color: $accent 50%; + border-title-style: bold; + padding: 0 1; + background: $surface; + } + + FloatingPanels > .status-body { + height: 100%; + width: 100%; + } + + FloatingPanels .status-scroll { + width: 100%; + height: 1fr; + overflow-x: hidden; + overflow-y: auto; + scrollbar-size: 0 0; + scrollbar-visibility: hidden; + } + + FloatingPanels .status-items { + width: 100%; + height: auto; + } + + FloatingPanels .status-column { + width: 1fr; + height: auto; + } + + FloatingPanels .status-spacer { + width: 1; + height: 1; + } + + FloatingPanels .status-item { + width: 100%; + height: 1; + color: $accent; + } + + FloatingPanels .status-item.inactive { + color: white; + text-style: dim; + } + + FloatingPanels .status-empty { + width: auto; + height: 1; + color: $secondary; + } + """ + + def __init__(self, **kwargs): + super().__init__(**kwargs) + self.border_title = "agents ───── skills ───── servers" + self._synced_height = None + self._synced_column_height = None + self._command_running = False + + def compose(self): + with Vertical(classes="status-body"): + with VerticalScroll(classes="status-scroll", id="status-scroll"): + with Horizontal(classes="status-items"): + yield StatusColumn("subagents", id="subagents-column", classes="status-column") + yield Static("", classes="status-spacer") + yield StatusColumn("skills", id="skills-column", classes="status-column") + yield Static("", classes="status-spacer") + yield StatusColumn("servers", id="servers-column", classes="status-column") + + def on_mount(self): + """Start periodic height sync and data refreshes.""" + self.call_after_refresh(self._sync_height) + self.set_interval(0.25, self._sync_height) + self.set_interval(1.0, self.refresh_data) + self.refresh_data() + + def on_status_item_clicked(self, event: StatusItemClicked): + """Route an item click to the matching coder command.""" + self._run_coder_command(event.category, event.name, event.active) + self._scroll_to_top() + self.app.call_after_refresh(self._focus_input) + + def _sync_height(self): + """Match the box height to the input container's current height.""" + try: + height = self.app.query_one("#input-container").region.height + except Exception: + return + + if height <= 0: + return + + if height != self._synced_height: + self._synced_height = height + self.styles.height = height + + try: + hints = self.app.query_one("#key-hints") + column = self.app.query_one("#input-column") + hints_height = hints.region.height + hints_height += int(hints.styles.margin.top) + hints_height += int(hints.styles.margin.bottom) + column_height = height + hints_height + except Exception: + return + + if column_height != self._synced_column_height: + self._synced_column_height = column_height + column.styles.height = column_height + + def toggle_visible(self): + """Hide the whole extensions column, or show it again.""" + target = self.parent + if getattr(target, "id", None) != "extensions-column": + target = self + + target.display = not target.display + if target.display: + self.refresh_data() + + def refresh_data(self): + """Refresh the extensions columns from the active coder's state.""" + coder = self._get_active_coder() + if coder is None: + self._set_column("#subagents-column", []) + self._set_column("#skills-column", []) + self._set_column("#servers-column", []) + return + + self._set_column("#subagents-column", self._collect_agents()) + self._set_column("#skills-column", self._collect_skills(coder)) + self._set_column("#servers-column", self._collect_mcp(coder)) + + def _focus_input(self): + """Return keyboard focus to the prompt input.""" + try: + self.app.query_one("#input").focus() + except Exception: + pass + + def _scroll_to_top(self): + """Return the shared list to the top after an item click.""" + try: + self.query_one("#status-scroll", VerticalScroll).scroll_to(y=0, animate=False) + except Exception: + pass + + def _set_column(self, selector, entries): + """Replace one column's clickable items, skipping unchanged data.""" + try: + column = self.query_one(selector, StatusColumn) + except Exception: + return + + ordered = sorted(entries, key=lambda item: (not item[1], item[0].lower())) + items = [(self._format_name(name), name, active) for name, active in ordered] + signature = tuple(items) + if signature == getattr(column, "_last_signature", None): + return + + column._last_signature = signature + column.items = items + + def _run_coder_command(self, category, name, active): + """Schedule the click's coder command on the worker loop.""" + if self._command_running: + return + + coder = self._get_primary_coder() if category == "subagents" else self._get_active_coder() + if coder is None: + return + + command_cls = self._command_class(category, active) + if command_cls is None: + return + + self._command_running = True + + async def runner(): + try: + await command_cls.execute(coder.io, coder, name) + finally: + self._command_running = False + try: + self.app.call_from_thread(self.refresh_data) + except Exception: + pass + + loop = getattr(getattr(self.app, "worker", None), "loop", None) + if loop is not None: + loop.call_soon_threadsafe(lambda: loop.create_task(runner())) + return + + try: + self.run_worker(runner()) + except Exception: + self._command_running = False + + def _command_class(self, category, active): + """Return the command class for an item click.""" + if category == "subagents": + from cecli.commands.spawn_agent import SpawnAgentCommand + + return SpawnAgentCommand + + if category == "skills": + if active: + from cecli.commands.remove_skill import RemoveSkillCommand + + return RemoveSkillCommand + + from cecli.commands.load_skill import LoadSkillCommand + + return LoadSkillCommand + + if category == "servers": + if active: + from cecli.commands.remove_mcp import RemoveMcpCommand + + return RemoveMcpCommand + + from cecli.commands.load_mcp import LoadMcpCommand + + return LoadMcpCommand + + return None + + def _get_primary_coder(self): + """Return the primary coder from the TUI worker, if available.""" + try: + return self.app.worker.coder + except Exception: + return None + + def _get_active_coder(self): + """Return the foreground coder, falling back to the primary coder.""" + primary = self._get_primary_coder() + if primary is None: + return None + + try: + from cecli.helpers.agents.service import AgentService + + return AgentService.get_instance(primary).foreground_coder or primary + except Exception: + return primary + + def _collect_agents(self): + """Collect agent types, marking the foreground agent's type active.""" + primary = self._get_primary_coder() + if primary is None: + return [] + + try: + from cecli.helpers.agents.service import AgentService + + service = AgentService.get_instance(primary) + registered = set(AgentService.get_registry().keys()) + live = {info.name for info in service.sub_agents.values() if info is not None} + active_name = None + foreground_uuid = getattr(service, "foreground_uuid", None) + if foreground_uuid: + info = service.sub_agents.get(str(foreground_uuid)) + if info is not None: + active_name = info.name + except Exception: + return [] + + return [(name, name == active_name) for name in registered | live] + + def _collect_skills(self, coder): + """Collect discoverable skills and mark the loaded ones active.""" + manager = getattr(coder, "skills_manager", None) + if manager is None: + return [] + + loaded = set(getattr(manager, "_loaded_skills", set()) or set()) + try: + available = {skill.name for skill in manager.find_skills()} + except Exception: + available = set() + + return [(name, name in loaded) for name in available | loaded] + + def _collect_mcp(self, coder): + """Collect MCP servers and mark connected, unfiltered ones active.""" + manager = getattr(coder, "mcp_manager", None) + if manager is None: + return [] + + registered = getattr(coder, "registered_servers", None) or {} + included = set(registered.get("included", set()) or set()) + excluded = set(registered.get("excluded", set()) or set()) + + try: + servers = manager.servers + except Exception: + servers = [] + + entries = [] + for server in servers: + name = getattr(server, "name", "") + if not name: + continue + + active = bool(getattr(server, "is_connected", False)) + if included and name not in included: + active = False + if name in excluded: + active = False + + entries.append((name, active)) + + return entries + + @staticmethod + def _format_name(name: str) -> str: + """Shorten a name to keep rows compact.""" + name = str(name) + if len(name) > FloatingPanels.MAX_NAME_WIDTH: + return name[: FloatingPanels.MAX_NAME_WIDTH - 1] + "…" + return name diff --git a/cecli/tui/widgets/input_area.py b/cecli/tui/widgets/input_area.py index eecfd66c3e5..52b92c33266 100644 --- a/cecli/tui/widgets/input_area.py +++ b/cecli/tui/widgets/input_area.py @@ -252,6 +252,12 @@ def on_key(self, event) -> None: event.prevent_default() self.app.action_paste_clipboard() + if self.app.is_key_for("panels", event.key): + event.stop() + event.prevent_default() + self.app.action_toggle_floating_panels() + return + if self.app.is_key_for("history", event.key): event.stop() event.prevent_default() diff --git a/cecli/website/docs/config/workflows.md b/cecli/website/docs/config/workflows.md index 5b4ccabdb16..b516c4c3f87 100644 --- a/cecli/website/docs/config/workflows.md +++ b/cecli/website/docs/config/workflows.md @@ -9,7 +9,7 @@ A **workflow** is a named, optionally scheduled sequence of prompts that run una ## Where to define them -Preferred: one YAML file per workflow in `~/.cecli/workflows/`: +Preferred: one YAML file per workflow in `~/.cecli/workflows/` (with `~/.agents/workflows/` also discovered): ``` ~/.cecli/workflows/ @@ -25,6 +25,7 @@ Also supported: a repo-local `.cecli/workflows/` or `.agents/workflows/` directo 4. **Standard local directory** — `.agents/workflows/*.yml` / `*.yaml` in the git root (or working directory), loaded in sorted order. 5. **Merged config** — the `workflows:` key in `.cecli.conf.yml`. 6. **Global files** — `~/.cecli/workflows/*.yml` / `*.yaml`, loaded in sorted order. +7. **Standard global directory** — `~/.agents/workflows/*.yml` / `*.yaml`, loaded in sorted order. On a duplicate `name`, the higher-precedence definition wins and the loser is skipped with a warning. A source that fails to parse is skipped with a warning as well — a malformed workflows file never crashes startup. @@ -147,7 +148,7 @@ tasks: ## Compatibility -- Workflows are additive: nothing changes until you create a `workflows:` key or a file in `~/.cecli/workflows/`. +- Workflows are additive: nothing changes until you create a `workflows:` key or a file in `~/.cecli/workflows/` or `~/.agents/workflows/`. - `wf:` agents show up in `/agent-tree` alongside `ws:` agents; a workflow can itself delegate sub-agents (nested delegation enabled). - Existing `/queue` semantics are unchanged; workflows reuse the same queue. diff --git a/requirements.txt b/requirements.txt index 0aaac2092d5..9df58356241 100644 --- a/requirements.txt +++ b/requirements.txt @@ -251,6 +251,7 @@ ngram==4.0.3 numpy==2.3.5 # via # -c requirements/common-constraints.txt + # -r requirements/requirements.in # chromadb # moonshine-voice # onnxruntime @@ -263,6 +264,7 @@ oauthlib==3.3.1 onnxruntime==1.29.0 # via # -c requirements/common-constraints.txt + # -r requirements/requirements.in # chromadb opentelemetry-api==1.44.0 # via diff --git a/requirements/common-constraints.txt b/requirements/common-constraints.txt index d36c2fc0a31..0fa6520cbdb 100644 --- a/requirements/common-constraints.txt +++ b/requirements/common-constraints.txt @@ -206,6 +206,7 @@ nodeenv==1.9.1 numpy==2.3.5 # via # -r requirements/requirements-help.in + # -r requirements/requirements.in # chromadb # contourpy # matplotlib @@ -219,7 +220,9 @@ oauthlib==3.3.1 objgraph==3.6.2 # via -r requirements/requirements-dev.in onnxruntime==1.29.0 - # via chromadb + # via + # -r requirements/requirements.in + # chromadb opentelemetry-api==1.44.0 # via # chromadb diff --git a/requirements/requirements.in b/requirements/requirements.in index e4a1ba5650b..bd251b356bb 100644 --- a/requirements/requirements.in +++ b/requirements/requirements.in @@ -34,6 +34,10 @@ pypandoc>=1.15 # storage/caching chromadb>=1.5.9 +# chromadb's default ONNX embedder needs these directly; keep them in the base +# install so /help and creator docs work without the optional [help] extra. +numpy>=1.26.4 +onnxruntime>=1.14.1 diskcache>=5.6.3 # general string parsing diff --git a/tests/basic/test_diff_helpers.py b/tests/basic/test_diff_helpers.py new file mode 100644 index 00000000000..b319346ce05 --- /dev/null +++ b/tests/basic/test_diff_helpers.py @@ -0,0 +1,89 @@ +from cecli.helpers.diff.diff_helpers import unified_diff_compacted + + +def _added_lines(diff): + return [line[1:] for line in diff if line.startswith("+") and not line.startswith("+++")] + + +def _removed_lines(diff): + return [line[1:] for line in diff if line.startswith("-") and not line.startswith("---")] + + +def test_compacts_duplicate_insertion_block(): + old = ["def first():", " return 1", "", "", "def second():", " return 1"] + new = [ + "def first():", + " return 1", + "", + "", + "def middle():", + " return 1", + "", + "", + "def second():", + " return 1", + ] + + diff = unified_diff_compacted(old, new, "a", "b", context_lines=3) + added = _added_lines(diff) + first_added = next(line for line in added if line.strip()) + + assert "def middle():" in first_added + assert not any("def first():" in line for line in added) + assert not any("def second():" in line for line in added) + + +def test_compacts_duplicate_deletion_block(): + old = [ + "def first():", + " return 1", + "", + "", + "def middle():", + " return 1", + "", + "", + "def second():", + " return 1", + ] + new = ["def first():", " return 1", "", "", "def second():", " return 1"] + + diff = unified_diff_compacted(old, new, "a", "b", context_lines=3) + removed = _removed_lines(diff) + first_removed = next(line for line in removed if line.strip()) + + assert "def middle():" in first_removed + assert not any("def first():" in line for line in removed) + assert not any("def second():" in line for line in removed) + + +def test_unchanged_input_returns_empty_diff(): + lines = ["alpha", "beta"] + + assert unified_diff_compacted(lines, list(lines)) == [] + + +def test_replacement_diff_has_correct_ranges(): + diff = unified_diff_compacted(["a", "b", "c"], ["a", "B", "c"], "old", "new") + + assert diff == [ + "--- old", + "+++ new", + "@@ -1,3 +1,3 @@", + " a", + "-b", + "+B", + " c", + ] + + +def test_compacts_insertion_at_start(): + diff = unified_diff_compacted(["a", "b"], ["a", "a", "b"], "a", "b", context_lines=3) + + assert diff == ["--- a", "+++ b", "@@ -1,2 +1,3 @@", " a", "+a", " b"] + + +def test_compacts_deletion_at_start(): + diff = unified_diff_compacted(["a", "a", "b"], ["a", "b"], "a", "b", context_lines=3) + + assert diff == ["--- a", "+++ b", "@@ -1,3 +1,2 @@", " a", "-a", " b"] diff --git a/tests/basic/test_main.py b/tests/basic/test_main.py index 392dfd5dd02..3248a37c54b 100644 --- a/tests/basic/test_main.py +++ b/tests/basic/test_main.py @@ -761,6 +761,9 @@ def test_boolean_flags(flag_arg, attr_name, expected, dummy_io, git_temp_dir): def test_no_verify_ssl_sets_model_info_manager(dummy_io, git_temp_dir, mocker): mock_set_verify_ssl = mocker.patch("cecli.models.ModelInfoManager.set_verify_ssl") mock_model = mocker.patch("cecli.models.Model") + # Startup now builds the default agent coder, so the fully mocked model + # must return a real int for map tokens (base_coder compares it to 0). + mock_model.return_value.get_repo_map_tokens.return_value = 0 mock_model.return_value.info = {} mock_model.return_value.name = "gpt-4" mock_model.return_value.validate_environment.return_value = { @@ -891,6 +894,31 @@ def test_invalid_edit_format(dummy_io, git_temp_dir, mocker, capsys): assert "not-a-real-format" in stderr_output +def test_default_edit_format_is_agent(dummy_io, git_temp_dir): + coder = main(["--exit", "--yes-always"], **dummy_io, return_coder=True) + assert coder.edit_format == "agent" + + +@pytest.mark.parametrize( + "flag_arg,expected_format", + [ + ("--edit-format", "whole"), + ("--chat-mode", "whole"), + (None, "agent"), + ], + ids=["edit_format_flag", "chat_mode_alias", "unset_defaults_to_agent"], +) +def test_edit_format_flag_overrides_agent_default( + flag_arg, expected_format, dummy_io, git_temp_dir, mock_coder +): + args = ["--exit", "--yes-always"] + if flag_arg: + args = [flag_arg, expected_format] + args + main(args, **dummy_io) + _, kwargs = mock_coder.call_args + assert kwargs["edit_format"] == expected_format + + @pytest.mark.parametrize( "api_key_env,expected_model_substr", [ diff --git a/tests/helpers/test_llms_reasoning_config.py b/tests/helpers/test_llms_reasoning_config.py index 4176ad88d76..023bd8ec46e 100644 --- a/tests/helpers/test_llms_reasoning_config.py +++ b/tests/helpers/test_llms_reasoning_config.py @@ -197,8 +197,13 @@ def test_responses_generic_reasoning_keys_stripped(): def test_responses_copilot_opt_in_to_summary(): """Copilot gpt-5 (responses mode) opts in to a reasoning summary even with no configured effort, so the model returns (and the capture logic exposes) the - readable ``summary`` block.""" + readable ``summary`` block. + + The bundled metadata currently resolves a default effort for this route, so + drop it to exercise the summary-only opt-in path. + """ resolved = resolve_model_config("github_copilot/gpt-5.1") + resolved["api_block"].pop("reasoning_effort", None) payload = _build("responses", resolved, {}) assert payload["reasoning"] == {"summary": "auto"} assert payload["include"] == ["reasoning.encrypted_content"] diff --git a/tests/helpers/workflows/test_config.py b/tests/helpers/workflows/test_config.py index 0d017999634..91a31b9b847 100644 --- a/tests/helpers/workflows/test_config.py +++ b/tests/helpers/workflows/test_config.py @@ -40,6 +40,14 @@ def _home_workflow(wdir: Path, filename: str, text: str) -> Path: return path +def _home_agents_workflow(home: Path, filename: str, text: str) -> Path: + wdir = home / ".agents" / "workflows" + wdir.mkdir(parents=True, exist_ok=True) + path = wdir / filename + path.write_text(text, encoding="utf-8") + return path + + def _mapping_text(name: str, prompt: str) -> str: return f"name: {name}\ntasks:\n - prompt: {prompt}\n" @@ -111,6 +119,28 @@ def test_glob_of_home_files_is_stably_sorted(isolated): assert [wf.name for wf in load_workflows()] == ["a", "b"] +def test_home_agents_dir_files_are_loaded(isolated): + _home_agents_workflow( + isolated["home"], "pipelines.yml", _mapping_text("home-agents-dir", "check") + ) + + configs = load_workflows() + + assert [wf.name for wf in configs] == ["home-agents-dir"] + assert configs[0].source_path == str( + isolated["home"] / ".agents" / "workflows" / "pipelines.yml" + ) + + +def test_home_cecli_dir_wins_over_home_agents_dir(isolated): + _home_agents_workflow(isolated["home"], "dup.yml", _mapping_text("dup", "agents")) + _home_workflow(isolated["wdir"], "dup.yml", _mapping_text("dup", "cecli")) + + configs = load_workflows() + + assert [wf.tasks[0].prompt for wf in configs] == ["cecli"] + + # ── Case 3: hyphen-case keys normalize ────────────────────────────────────── diff --git a/tests/subagents/test_service.py b/tests/subagents/test_service.py index f6ee8469a5b..3eede1493fc 100644 --- a/tests/subagents/test_service.py +++ b/tests/subagents/test_service.py @@ -1025,6 +1025,39 @@ async def test_wf_agent_inherits_and_merges(self, service): assert merged["skip_cli_confirmations"] is True assert merged["allow_nested_delegation"] is True # defaulted + @pytest.mark.asyncio + async def test_wf_servers_allowlist_forces_local(self, service): + """A wf: servers allowlist always keeps the Local server available.""" + service.coder.agent_config = {} + self._register( + "wf:demo", + {"servers_includelist": ["GitLab"], "skip_cli_confirmations": True}, + ) + + merged = await self._create(service, "wf:demo") + + assert merged["servers_includelist"] == ["gitlab", "local"] + + @pytest.mark.asyncio + async def test_wf_without_allowlist_does_not_force_local(self, service): + """Only an explicit allowlist gets the Local server added.""" + service.coder.agent_config = {} + self._register("wf:demo", {"skip_cli_confirmations": True}) + + merged = await self._create(service, "wf:demo") + + assert "servers_includelist" not in merged + + @pytest.mark.asyncio + async def test_ws_servers_allowlist_does_not_force_local(self, service): + """Local is forced for wf: agents only, not for other inherit-prefix agents.""" + service.coder.agent_config = {} + self._register("ws:app", {"servers_includelist": ["github"]}) + + merged = await self._create(service, "ws:app") + + assert merged["servers_includelist"] == ["github"] + @pytest.mark.asyncio async def test_wf_respects_explicit_optout(self, service): """wf: uses setdefault, so an explicit False in the config survives.""" @@ -1160,3 +1193,29 @@ async def contends(): assert "c1" in order assert "c2" in order assert "holder-released" in order + + +class TestStartGenerateTaskSwitch: + """A mid-turn command signal rebuilds the sub-agent coder in place.""" + + @pytest.mark.asyncio + async def test_switch_signal_switches_sub_agent_instead_of_escaping(self, service): + from cecli.signals import SwitchCoderSignal + + coder = MagicMock() + coder.uuid = "sub-uuid" + + async def raise_switch(**kwargs): + raise SwitchCoderSignal(edit_format="subagent") + + coder.generate = AsyncMock(side_effect=raise_switch) + info = SubAgentInfo(name="wf:demo", coder=coder, parent_uuid="parent-uuid") + info.detached = True + service.switch_sub_agent_coder = AsyncMock() + + task = service.start_generate_task(info, "/load-mcp github") + await task + + service.switch_sub_agent_coder.assert_awaited_once() + assert service.switch_sub_agent_coder.call_args.args[0] is info + assert info.status == SubAgentStatus.FINISHED diff --git a/tests/tui/test_floating_panels.py b/tests/tui/test_floating_panels.py new file mode 100644 index 00000000000..860420f867a --- /dev/null +++ b/tests/tui/test_floating_panels.py @@ -0,0 +1,222 @@ +import asyncio +from types import SimpleNamespace +from unittest.mock import patch + +from textual.app import App, ComposeResult +from textual.binding import Binding +from textual.containers import Horizontal, Vertical, VerticalScroll +from textual.widgets import Input, Static + +from cecli.tui.widgets.floating_panels import ( + FloatingPanels, + StatusColumn, + StatusItem, + StatusItemClicked, +) + + +class FakeSkill: + def __init__(self, name): + self.name = name + + +class FakeSkills: + def __init__(self): + self._loaded_skills = {"pdf"} + self.names = ["pdf", "docx", "spark"] + + def find_skills(self): + return [FakeSkill(name) for name in self.names] + + +class FakeServer: + def __init__(self, name, connected): + self.name = name + self.is_connected = connected + + +class FakeMcp: + def __init__(self): + self.servers = [FakeServer("local", True), FakeServer("github", False)] + + +class FakeCoder: + registered_servers = {"included": set(), "excluded": set()} + + def __init__(self): + self.skills_manager = FakeSkills() + self.mcp_manager = FakeMcp() + + +class _PanelApp(App): + CSS = """ + #input-row { height: auto; width: 100%; align-vertical: top; } + #input-container { height: 5; width: 1fr; } + #key-hints { height: 1; margin: 0 0 1 0; } + """ + + BINDINGS = [Binding("alt+p", "toggle_floating_panels", "Toggle panels", show=False)] + + def __init__(self): + super().__init__() + self.worker = SimpleNamespace(coder=FakeCoder()) + + def compose(self) -> ComposeResult: + with Horizontal(id="input-row"): + with Vertical(id="input-column"): + with Vertical(id="input-container"): + yield Input(id="input") + yield Static("hints", id="key-hints") + with Vertical(id="extensions-column"): + yield FloatingPanels(id="floating-panels") + yield Static("alt+p to toggle", id="extensions-hint") + + def action_toggle_floating_panels(self): + self.query_one(FloatingPanels).toggle_visible() + + +def _column_text(panels: FloatingPanels, column_id: str) -> str: + column = panels.query_one(f"#{column_id}", StatusColumn) + return " ".join(str(item.render()) for item in column.query(StatusItem)) + + +async def _run_panel_checks(): + app = _PanelApp() + async with app.run_test(size=(100, 40)) as pilot: + panels = app.query_one("#floating-panels", FloatingPanels) + input_container = app.query_one("#input-container") + extensions_column = app.query_one("#extensions-column") + input_column = app.query_one("#input-column") + input_area = app.query_one("#input", Input) + panels.refresh_data() + await pilot.pause() + await asyncio.sleep(0.1) + await pilot.pause() + + visible_input_width = input_container.region.width + + assert panels.border_title == "agents ───── skills ───── servers" + assert panels.region.x >= input_container.region.right + assert panels.region.bottom == input_container.region.bottom + assert panels.region.height == input_container.region.height + + columns = list(panels.query(".status-column")) + assert len(columns) == 3 + column_widths = [column.region.width for column in columns] + assert len(set(column_widths)) == 1 + assert columns[0].region.x < columns[1].region.x < columns[2].region.x + assert columns[1].region.x - columns[0].region.right == 1 + assert columns[2].region.x - columns[1].region.right == 1 + assert len({column.region.y for column in columns}) == 1 + + assert _column_text(panels, "subagents-column") == "● worker ○ tester" + assert _column_text(panels, "skills-column") == "● pdf ○ docx ○ spark" + assert _column_text(panels, "servers-column") == "● local ○ github" + + long_name = "a-very-long-skill-name-that-keeps-going" + formatted = FloatingPanels._format_name(long_name) + assert len(formatted) == FloatingPanels.MAX_NAME_WIDTH + assert formatted.endswith("…") + + captured = [] + panels._run_coder_command = lambda category, name, active: captured.append( + (category, name, active) + ) + skill_items = list(panels.query_one("#skills-column", StatusColumn).query(StatusItem)) + input_area.blur() + await pilot.pause() + await pilot.click(skill_items[0]) + await pilot.pause() + await pilot.click(skill_items[1]) + await pilot.pause() + assert captured == [("skills", "pdf", True), ("skills", "docx", False)] + assert app.focused is input_area + + panels.on_status_item_clicked(StatusItemClicked("subagents", "worker", False)) + panels.on_status_item_clicked(StatusItemClicked("servers", "github", False)) + assert captured[-2:] == [ + ("subagents", "worker", False), + ("servers", "github", False), + ] + + assert panels._command_class("subagents", False).__name__ == "SpawnAgentCommand" + assert panels._command_class("skills", True).__name__ == "RemoveSkillCommand" + assert panels._command_class("skills", False).__name__ == "LoadSkillCommand" + assert panels._command_class("servers", True).__name__ == "RemoveMcpCommand" + assert panels._command_class("servers", False).__name__ == "LoadMcpCommand" + + panels._collect_skills = lambda coder: [("z", True), ("a", True), ("b", False)] + panels.refresh_data() + await pilot.pause() + assert _column_text(panels, "skills-column") == "● a ● z ○ b" + + panels._collect_skills = lambda coder: [(f"skill-{i}", True) for i in range(20)] + panels.refresh_data() + await pilot.pause() + await asyncio.sleep(0.1) + await pilot.pause() + + scroll = panels.query_one("#status-scroll", VerticalScroll) + assert scroll.max_scroll_y > 0 + assert scroll.styles.scrollbar_visibility == "hidden" + assert "skill-19" in _column_text(panels, "skills-column") + assert panels.region.height == input_container.region.height + + scroll.scroll_to(y=5, animate=False) + await pilot.pause() + assert scroll.scroll_y > 0 + + panels.on_status_item_clicked(StatusItemClicked("skills", "skill-19", True)) + await pilot.pause() + assert scroll.scroll_y == 0 + + input_container.styles.height = 8 + await pilot.pause() + await asyncio.sleep(0.35) + await pilot.pause() + assert panels.region.height == 8 + assert input_column.region.height == 10 + + await pilot.press("alt+p") + await pilot.pause() + await asyncio.sleep(0.1) + await pilot.pause() + assert extensions_column.display is False + assert input_container.region.width > visible_input_width + + await pilot.press("alt+p") + await pilot.pause() + await asyncio.sleep(0.1) + await pilot.pause() + assert extensions_column.display is True + assert input_container.region.width == visible_input_width + + +def test_floating_panels_layout_scroll_and_toggle(): + agent_service = SimpleNamespace( + sub_agents={"u1": SimpleNamespace(name="worker")}, + foreground_uuid="u1", + foreground_coder=None, + ) + with ( + patch( + "cecli.helpers.agents.service.AgentService.get_instance", + return_value=agent_service, + ), + patch( + "cecli.helpers.agents.service.AgentService.get_registry", + return_value={"tester": None, "worker": None}, + ), + ): + asyncio.run(_run_panel_checks()) + + +def test_floating_panels_key_registration(): + from cecli.tui.app import TUI + + tui = object.__new__(TUI) + tui.args = SimpleNamespace(tui_config=None) + config = tui._get_config() + + assert config["key_bindings"]["panels"] == "alt+p" + assert hasattr(TUI, "action_toggle_floating_panels")