From a42c3780d582a28ec4e6f10faf56639cef8020ac Mon Sep 17 00:00:00 2001 From: Jason Wang Date: Wed, 21 Oct 2026 09:32:32 +0800 Subject: [PATCH] Profile catalog data Local review preparation only. Reuse Draft PR #4118. --- .../core/model_service/profiles/LICENSE | 21 ++ .../core/model_service/profiles/README.md | 176 +++++++++ .../core/model_service/profiles/manifest.toml | 342 ++++++++++++++++++ .../models/alibaba/qwen3-coder-next.toml | 27 ++ .../profiles/models/alibaba/qwen3.5-27b.toml | 29 ++ .../models/alibaba/qwen3.5-35b-a3b.toml | 23 ++ .../models/alibaba/qwen3.5-397b-a17b.toml | 29 ++ .../models/alibaba/qwen3.5-flash.toml | 22 ++ .../profiles/models/alibaba/qwen3.5-plus.toml | 19 + .../profiles/models/alibaba/qwen3.7-max.toml | 82 +++++ .../profiles/models/alibaba/qwen3.7-plus.toml | 19 + .../models/alibaba/qwen3.8-2.4t-a95b.toml | 33 ++ .../profiles/models/alibaba/qwen3.8-27b.toml | 236 ++++++++++++ .../models/alibaba/qwen3.8-flash.toml | 19 + .../profiles/models/alibaba/qwen3.8-max.toml | 266 ++++++++++++++ .../models/bytedance-seed/seed-2.1-pro.toml | 22 ++ .../deepseek/deepseek-v4-flash-0731.toml | 105 ++++++ .../models/deepseek/deepseek-v4-flash.toml | 183 ++++++++++ .../models/deepseek/deepseek-v4-pro-0813.toml | 106 ++++++ .../models/deepseek/deepseek-v4-pro.toml | 217 +++++++++++ .../models/deepseek/deepseek-v4.1-flash.toml | 311 ++++++++++++++++ .../profiles/models/minimax/MiniMax-M2.7.toml | 46 +++ .../profiles/models/minimax/MiniMax-M3.toml | 67 ++++ .../profiles/models/moonshotai/kimi-k3.toml | 135 +++++++ .../profiles/models/zhipuai/glm-5.1.toml | 53 +++ .../profiles/models/zhipuai/glm-5.2.toml | 167 +++++++++ .../models/zhipuai/glm-5.3-flash.toml | 70 ++++ .../profiles/models/zhipuai/glm-5.3.toml | 158 ++++++++ .../profiles/models/zhipuai/glm-5.toml | 49 +++ .../models/MiniMax/MiniMax-M2.7.toml | 25 ++ .../alibaba-cn/models/deepseek-v4-flash.toml | 16 + .../alibaba-cn/models/deepseek-v4-pro.toml | 16 + .../models/deepseek-v4.1-flash.toml | 27 ++ .../providers/alibaba-cn/models/glm-5.1.toml | 45 +++ .../providers/alibaba-cn/models/glm-5.2.toml | 22 ++ .../providers/alibaba-cn/models/glm-5.3.toml | 25 ++ .../providers/alibaba-cn/models/glm-5.toml | 41 +++ .../providers/alibaba-cn/models/kimi-k3.toml | 27 ++ .../alibaba-cn/models/qwen3.5-397b-a17b.toml | 42 +++ .../alibaba-cn/models/qwen3.5-flash.toml | 53 +++ .../alibaba-cn/models/qwen3.5-plus.toml | 53 +++ .../alibaba-cn/models/qwen3.7-max.toml | 14 + .../alibaba-cn/models/qwen3.7-plus.toml | 21 ++ .../alibaba-cn/models/qwen3.8-flash.toml | 26 ++ .../alibaba-cn/models/qwen3.8-max.toml | 27 ++ .../providers/alibaba-cn/provider.toml | 20 + .../deepseek/models/deepseek-flash.toml | 32 ++ .../deepseek/models/deepseek-v4-flash.toml | 30 ++ .../deepseek/models/deepseek-v4-pro.toml | 35 ++ .../profiles/providers/deepseek/provider.toml | 12 + .../models/Pro/zai-org/GLM-5.1.toml | 29 ++ .../models/Pro/zai-org/GLM-5.toml | 27 ++ .../models/deepseek-ai/DeepSeek-V4-Flash.toml | 14 + .../models/deepseek-ai/DeepSeek-V4-Pro.toml | 25 ++ .../models/zai-org/GLM-5.2.toml | 15 + .../providers/siliconflow-cn/provider.toml | 16 + .../providers/tokenpony/provider.toml | 8 + .../models/deepseek-v4-flash.toml | 23 ++ .../models/deepseek-v4-pro.toml | 22 ++ .../models/glm-5.3-flash.toml | 17 + .../models/glm-5.3.toml | 30 ++ .../models/kimi-k3.toml | 28 ++ .../models/minimax-m3.toml | 21 ++ .../volcengine-coding-plan/provider.toml | 42 +++ .../models/deepseek-v4-flash-ga-260731.toml | 21 ++ .../models/deepseek-v4-pro-ga-260813.toml | 21 ++ .../models/doubao-seed-2-1-pro-260628.toml | 19 + .../volcengine/models/glm-5-2-260617.toml | 21 ++ .../models/glm-5-3-flash-260828.toml | 18 + .../providers/volcengine/provider.toml | 23 ++ .../profiles/rules/manifest.toml | 133 +++++++ .../rules/models/alibaba/qwen3.5-plus.toml | 12 + .../alibaba-cn/models/deepseek-v4-flash.toml | 32 ++ .../alibaba-cn/models/deepseek-v4-pro.toml | 32 ++ .../models/deepseek-v4.1-flash.toml | 32 ++ .../providers/alibaba-cn/models/glm-5.1.toml | 32 ++ .../providers/alibaba-cn/models/glm-5.3.toml | 19 + .../providers/alibaba-cn/models/glm-5.toml | 32 ++ .../providers/alibaba-cn/models/kimi-k3.toml | 19 + .../alibaba-cn/models/qwen3.5-397b-a17b.toml | 32 ++ .../alibaba-cn/models/qwen3.5-flash.toml | 32 ++ .../alibaba-cn/models/qwen3.7-max.toml | 32 ++ .../alibaba-cn/models/qwen3.7-plus.toml | 32 ++ .../alibaba-cn/models/qwen3.8-flash.toml | 45 +++ .../alibaba-cn/models/qwen3.8-max.toml | 45 +++ .../rules/providers/alibaba-cn/provider.toml | 8 + .../providers/alibaba-cn/qwen3.5-plus.toml | 117 ++++++ .../deepseek/models/deepseek-flash.toml | 32 ++ .../deepseek/models/deepseek-v4-flash.toml | 32 ++ .../deepseek/models/deepseek-v4-pro.toml | 32 ++ .../models/Pro/zai-org/GLM-5.toml | 19 + .../models/deepseek-ai/DeepSeek-V4-Flash.toml | 19 + .../models/zai-org/GLM-5.2.toml | 19 + .../models/deepseek-v4-flash.toml | 19 + .../models/deepseek-v4-pro.toml | 19 + .../models/glm-5.3-flash.toml | 19 + .../models/glm-5.3.toml | 19 + .../models/kimi-k3.toml | 32 ++ .../models/minimax-m3.toml | 19 + .../models/deepseek-v4-flash-ga-260731.toml | 32 ++ .../models/deepseek-v4-pro-ga-260813.toml | 32 ++ .../models/doubao-seed-2-1-pro-260628.toml | 19 + .../volcengine/models/glm-5-2-260617.toml | 32 ++ .../models/glm-5-3-flash-260828.toml | 19 + 104 files changed, 5181 insertions(+) create mode 100644 sdk/nexent/core/model_service/profiles/LICENSE create mode 100644 sdk/nexent/core/model_service/profiles/README.md create mode 100644 sdk/nexent/core/model_service/profiles/manifest.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3-coder-next.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-27b.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-35b-a3b.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-397b-a17b.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-plus.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.7-max.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.7-plus.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-2.4t-a95b.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-27b.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-max.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/bytedance-seed/seed-2.1-pro.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-flash-0731.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-pro-0813.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-pro.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4.1-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/minimax/MiniMax-M2.7.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/minimax/MiniMax-M3.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/moonshotai/kimi-k3.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.1.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.2.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.3-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.3.toml create mode 100644 sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/MiniMax/MiniMax-M2.7.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4-pro.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4.1-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.1.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.2.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.3.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/kimi-k3.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/qwen3.5-397b-a17b.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/qwen3.5-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/qwen3.5-plus.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/qwen3.7-max.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/qwen3.7-plus.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/qwen3.8-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/qwen3.8-max.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/alibaba-cn/provider.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/deepseek/models/deepseek-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/deepseek/models/deepseek-v4-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/deepseek/models/deepseek-v4-pro.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/deepseek/provider.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.1.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Pro.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/zai-org/GLM-5.2.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/provider.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/tokenpony/provider.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/deepseek-v4-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/deepseek-v4-pro.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/glm-5.3-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/glm-5.3.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/kimi-k3.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/minimax-m3.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/provider.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine/models/deepseek-v4-flash-ga-260731.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine/models/deepseek-v4-pro-ga-260813.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine/models/doubao-seed-2-1-pro-260628.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine/models/glm-5-2-260617.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine/models/glm-5-3-flash-260828.toml create mode 100644 sdk/nexent/core/model_service/profiles/providers/volcengine/provider.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/manifest.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/models/alibaba/qwen3.5-plus.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4-pro.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4.1-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.1.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.3.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/kimi-k3.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.5-397b-a17b.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.5-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.7-max.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.7-plus.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.8-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.8-max.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/provider.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/qwen3.5-plus.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-v4-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-v4-pro.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/zai-org/GLM-5.2.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/deepseek-v4-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/deepseek-v4-pro.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/glm-5.3-flash.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/glm-5.3.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/kimi-k3.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/minimax-m3.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/deepseek-v4-flash-ga-260731.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/deepseek-v4-pro-ga-260813.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/doubao-seed-2-1-pro-260628.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/glm-5-2-260617.toml create mode 100644 sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/glm-5-3-flash-260828.toml diff --git a/sdk/nexent/core/model_service/profiles/LICENSE b/sdk/nexent/core/model_service/profiles/LICENSE new file mode 100644 index 0000000000..9ef000844a --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/LICENSE @@ -0,0 +1,21 @@ +MIT License + +Copyright (c) 2025 models.dev + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. diff --git a/sdk/nexent/core/model_service/profiles/README.md b/sdk/nexent/core/model_service/profiles/README.md new file mode 100644 index 0000000000..a2ddae26dc --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/README.md @@ -0,0 +1,176 @@ +# Nexent model profile source catalog + +## Source and scope + +Selected upstream TOML files are copied byte-for-byte from +[models.dev](https://github.com/anomalyco/models.dev), commit +`e69e1844210e5cfe7c7e7f68d57328bc7451a51c`. +The original MIT copyright and permission notice is retained in [LICENSE](LICENSE). +All 66 selected upstream files were also checked against `dev` commit +`ab9af7ba4aa16ee65e9f42c01544393f4f49ec44` on 2026-10-16 and are unchanged. + +The inventory contains 26 model files, 35 provider-model files and six provider +files. Five provider files are copied from models.dev; the TokenPony provider +is authored by Nexent from its [official documentation](https://docs.tokenpony.cn/). +Its manifest entry explicitly records `origin = "nexent"`, a local revision and +source URL; it is not attributed to the upstream MIT data. +This is source material, not a verified support list. Pricing remains reference +data; this change does not implement cost estimation or change VLM consumers. + +## File ownership + +- Copied files under `models/` and `providers/` retain upstream paths, spelling, comments and bytes. +- Nexent-authored TOML uses the same schema; its manifest origin, documentation URL and local revision are mandatory. +- `manifest.toml` pins source kinds, paths, hashes and the upstream revision. +- Nexent normalization and API binding rules belong under `rules/` separately; + do not silently rewrite upstream files. +- Source references are provenance. Saved outlets retain their resolved + configuration and do not recompose it when the catalog changes. + +## Updating the selection + +1. Select one upstream commit and inspect its authored schema and generator. +2. Copy reviewed files and their complete `base_model` dependencies. +3. Preserve original bytes and paths; update manifest hashes and revision. +4. Review every field against the versioned Nexent mapping rules, including + fields retained as references or explicitly ignored. +5. Run integrity, parser, normalization, composition and installed-package + checks. Run the reusable real-model matrix before declaring support. +6. Keep deleted upstream entries locally unless deliberately removed in a Nexent + release; do not modify saved outlets automatically. + +No upstream TypeScript project, online synchronization or npm dependency is +required at runtime. Model weight licenses and service terms remain separate +from the license of these descriptive files. + + +## Nexent supplements and complete catalog loading + +`resolver.load_source_catalog()` reads source facts only. +`resolver.load_profile_catalog()` additionally validates `rules/manifest.toml`, +reads exact Nexent supplement declarations and publishes immutable atomic views. +The manifest pins each supplement file's digest separately from upstream data. + +Exact provider-model supplements bind the copied reasoning controls to registered +OpenAI Chat request fields. They preserve each source's effort enum and budget +bounds, keep unknown bounds unknown, and do not add numeric defaults. Alibaba +China and SiliconFlow toggles use `/enable_thinking`; Volcengine and DeepSeek +use `/thinking/type` through `thinking_toggle_v1`. Effort uses +`/reasoning_effort`, and the declared numeric budget uses `/thinking_budget`. +The separate `alibaba-cn/qwen3.5-plus` rule retains its earlier explicit defaults +and maximum-output declaration. Every known combination is tested through +composition and request rendering; that is not real-provider qualification. + +Keep upstream TOML unchanged. Put model rules under `rules/models/` and provider +or exact provider-model rules under `rules/providers/`; update their manifest +hashes when editing. Use the optional `json_fields` table only for JSON-only +values such as explicit null, never as a string alias or arbitrary path patch. + +## Complete selected inventory (2026-10-16) + +The tables enumerate available configuration references, not a provider/model +support certification. Provider selection does not imply every model is served +by that provider; the exact copied combinations are listed separately below. +Saved outlets are never recomposed by this source update. + +### Providers + +| Key | Name | Suggested API base URL | Origin | +| --- | --- | --- | --- | +| `alibaba-cn` | [Alibaba (China)](providers/alibaba-cn/provider.toml) | `https://dashscope.aliyuncs.com/compatible-mode/v1` | models_dev | +| `deepseek` | [DeepSeek](providers/deepseek/provider.toml) | `https://api.deepseek.com` | models_dev | +| `siliconflow-cn` | [SiliconFlow (China)](providers/siliconflow-cn/provider.toml) | `https://api.siliconflow.cn/v1` | models_dev | +| `tokenpony` | [TokenPony 小马算力](providers/tokenpony/provider.toml) | `https://api.tokenpony.cn/v1` | nexent | +| `volcengine-coding-plan` | [Volcengine Ark Coding Plan](providers/volcengine-coding-plan/provider.toml) | `https://ark.cn-beijing.volces.com/api/coding/v3` | models_dev | +| `volcengine` | [Volcengine Ark](providers/volcengine/provider.toml) | `https://ark.cn-beijing.volces.com/api/v3` | models_dev | + +### Models + +| Model key | Name | +| --- | --- | +| `alibaba/qwen3-coder-next` | [Qwen3 Coder Next](models/alibaba/qwen3-coder-next.toml) | +| `alibaba/qwen3.5-27b` | [Qwen3.5 27B](models/alibaba/qwen3.5-27b.toml) | +| `alibaba/qwen3.5-35b-a3b` | [Qwen3.5 35B-A3B](models/alibaba/qwen3.5-35b-a3b.toml) | +| `alibaba/qwen3.5-397b-a17b` | [Qwen3.5 397B-A17B](models/alibaba/qwen3.5-397b-a17b.toml) | +| `alibaba/qwen3.5-flash` | [Qwen3.5 Flash](models/alibaba/qwen3.5-flash.toml) | +| `alibaba/qwen3.5-plus` | [Qwen3.5 Plus](models/alibaba/qwen3.5-plus.toml) | +| `alibaba/qwen3.7-max` | [Qwen3.7 Max](models/alibaba/qwen3.7-max.toml) | +| `alibaba/qwen3.7-plus` | [Qwen3.7 Plus](models/alibaba/qwen3.7-plus.toml) | +| `alibaba/qwen3.8-2.4t-a95b` | [Qwen3.8 2.4T A95B](models/alibaba/qwen3.8-2.4t-a95b.toml) | +| `alibaba/qwen3.8-27b` | [Qwen3.8 27B](models/alibaba/qwen3.8-27b.toml) | +| `alibaba/qwen3.8-flash` | [Qwen3.8 Flash](models/alibaba/qwen3.8-flash.toml) | +| `alibaba/qwen3.8-max` | [Qwen3.8 Max](models/alibaba/qwen3.8-max.toml) | +| `bytedance-seed/seed-2.1-pro` | [Seed 2.1 Pro](models/bytedance-seed/seed-2.1-pro.toml) | +| `deepseek/deepseek-v4-flash-0731` | [DeepSeek V4 Flash 0731](models/deepseek/deepseek-v4-flash-0731.toml) | +| `deepseek/deepseek-v4-flash` | [DeepSeek V4 Flash](models/deepseek/deepseek-v4-flash.toml) | +| `deepseek/deepseek-v4-pro-0813` | [DeepSeek V4 Pro 0813](models/deepseek/deepseek-v4-pro-0813.toml) | +| `deepseek/deepseek-v4-pro` | [DeepSeek V4 Pro](models/deepseek/deepseek-v4-pro.toml) | +| `deepseek/deepseek-v4.1-flash` | [DeepSeek V4.1 Flash](models/deepseek/deepseek-v4.1-flash.toml) | +| `minimax/MiniMax-M2.7` | [MiniMax-M2.7](models/minimax/MiniMax-M2.7.toml) | +| `minimax/MiniMax-M3` | [MiniMax-M3](models/minimax/MiniMax-M3.toml) | +| `moonshotai/kimi-k3` | [Kimi K3](models/moonshotai/kimi-k3.toml) | +| `zhipuai/glm-5.1` | [GLM-5.1](models/zhipuai/glm-5.1.toml) | +| `zhipuai/glm-5.2` | [GLM-5.2](models/zhipuai/glm-5.2.toml) | +| `zhipuai/glm-5.3-flash` | [GLM-5.3-Flash](models/zhipuai/glm-5.3-flash.toml) | +| `zhipuai/glm-5.3` | [GLM-5.3](models/zhipuai/glm-5.3.toml) | +| `zhipuai/glm-5` | [GLM-5](models/zhipuai/glm-5.toml) | + +### Provider-model records + +| Provider/model request ID | Declared base model | +| --- | --- | +| [alibaba-cn/MiniMax/MiniMax-M2.7](providers/alibaba-cn/models/MiniMax/MiniMax-M2.7.toml) | Standalone upstream record; no base_model declared | +| [alibaba-cn/deepseek-v4-flash](providers/alibaba-cn/models/deepseek-v4-flash.toml) | `deepseek/deepseek-v4-flash` | +| [alibaba-cn/deepseek-v4-pro](providers/alibaba-cn/models/deepseek-v4-pro.toml) | `deepseek/deepseek-v4-pro` | +| [alibaba-cn/deepseek-v4.1-flash](providers/alibaba-cn/models/deepseek-v4.1-flash.toml) | `deepseek/deepseek-v4.1-flash` | +| [alibaba-cn/glm-5.1](providers/alibaba-cn/models/glm-5.1.toml) | Standalone upstream record; no base_model declared | +| [alibaba-cn/glm-5.2](providers/alibaba-cn/models/glm-5.2.toml) | `zhipuai/glm-5.2` | +| [alibaba-cn/glm-5.3](providers/alibaba-cn/models/glm-5.3.toml) | `zhipuai/glm-5.3` | +| [alibaba-cn/glm-5](providers/alibaba-cn/models/glm-5.toml) | Standalone upstream record; no base_model declared | +| [alibaba-cn/kimi-k3](providers/alibaba-cn/models/kimi-k3.toml) | `moonshotai/kimi-k3` | +| [alibaba-cn/qwen3.5-397b-a17b](providers/alibaba-cn/models/qwen3.5-397b-a17b.toml) | Standalone upstream record; no base_model declared | +| [alibaba-cn/qwen3.5-flash](providers/alibaba-cn/models/qwen3.5-flash.toml) | Standalone upstream record; no base_model declared | +| [alibaba-cn/qwen3.5-plus](providers/alibaba-cn/models/qwen3.5-plus.toml) | Standalone upstream record; no base_model declared | +| [alibaba-cn/qwen3.7-max](providers/alibaba-cn/models/qwen3.7-max.toml) | `alibaba/qwen3.7-max` | +| [alibaba-cn/qwen3.7-plus](providers/alibaba-cn/models/qwen3.7-plus.toml) | `alibaba/qwen3.7-plus` | +| [alibaba-cn/qwen3.8-flash](providers/alibaba-cn/models/qwen3.8-flash.toml) | `alibaba/qwen3.8-flash` | +| [alibaba-cn/qwen3.8-max](providers/alibaba-cn/models/qwen3.8-max.toml) | `alibaba/qwen3.8-max` | +| [deepseek/deepseek-flash](providers/deepseek/models/deepseek-flash.toml) | `deepseek/deepseek-v4.1-flash` | +| [deepseek/deepseek-v4-flash](providers/deepseek/models/deepseek-v4-flash.toml) | `deepseek/deepseek-v4.1-flash` | +| [deepseek/deepseek-v4-pro](providers/deepseek/models/deepseek-v4-pro.toml) | `deepseek/deepseek-v4-pro-0813` | +| [siliconflow-cn/Pro/zai-org/GLM-5.1](providers/siliconflow-cn/models/Pro/zai-org/GLM-5.1.toml) | Standalone upstream record; no base_model declared | +| [siliconflow-cn/Pro/zai-org/GLM-5](providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml) | Standalone upstream record; no base_model declared | +| [siliconflow-cn/deepseek-ai/DeepSeek-V4-Flash](providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml) | `deepseek/deepseek-v4-flash` | +| [siliconflow-cn/deepseek-ai/DeepSeek-V4-Pro](providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Pro.toml) | Standalone upstream record; no base_model declared | +| [siliconflow-cn/zai-org/GLM-5.2](providers/siliconflow-cn/models/zai-org/GLM-5.2.toml) | `zhipuai/glm-5.2` | +| [volcengine-coding-plan/deepseek-v4-flash](providers/volcengine-coding-plan/models/deepseek-v4-flash.toml) | `deepseek/deepseek-v4-flash` | +| [volcengine-coding-plan/deepseek-v4-pro](providers/volcengine-coding-plan/models/deepseek-v4-pro.toml) | `deepseek/deepseek-v4-pro` | +| [volcengine-coding-plan/glm-5.3-flash](providers/volcengine-coding-plan/models/glm-5.3-flash.toml) | `zhipuai/glm-5.3-flash` | +| [volcengine-coding-plan/glm-5.3](providers/volcengine-coding-plan/models/glm-5.3.toml) | `zhipuai/glm-5.3` | +| [volcengine-coding-plan/kimi-k3](providers/volcengine-coding-plan/models/kimi-k3.toml) | `moonshotai/kimi-k3` | +| [volcengine-coding-plan/minimax-m3](providers/volcengine-coding-plan/models/minimax-m3.toml) | `minimax/MiniMax-M3` | +| [volcengine/deepseek-v4-flash-ga-260731](providers/volcengine/models/deepseek-v4-flash-ga-260731.toml) | `deepseek/deepseek-v4-flash-0731` | +| [volcengine/deepseek-v4-pro-ga-260813](providers/volcengine/models/deepseek-v4-pro-ga-260813.toml) | `deepseek/deepseek-v4-pro-0813` | +| [volcengine/doubao-seed-2-1-pro-260628](providers/volcengine/models/doubao-seed-2-1-pro-260628.toml) | `bytedance-seed/seed-2.1-pro` | +| [volcengine/glm-5-2-260617](providers/volcengine/models/glm-5-2-260617.toml) | `zhipuai/glm-5.2` | +| [volcengine/glm-5-3-flash-260828](providers/volcengine/models/glm-5-3-flash-260828.toml) | `zhipuai/glm-5.3-flash` | + +DeepSeek official `deepseek-flash` declares `deepseek/deepseek-v4.1-flash` as +its base model. The retained `deepseek-v4-flash` request name is marked deprecated +upstream and now routes to V4.1 Flash. Preserve this distinction from the +provider-independent V4 Flash profile. TokenPony has no imported per-model +records; exact request model IDs remain operator-provided until documented. + +## Nexent experience precedence + +Source TOML remains unchanged. Outlet previews first combine authored model, +provider and exact hosted-model facts, then overlay Nexent supplement rules in +model, provider, exact-combination order. Higher layers replace a parameter's +complete declaration, a named implementation binding or an instruction strategy; +unmentioned keys inherit. Capability changes retain their evidence and previous +values. Permitted user values are applied last; fixed values, schema constraints, +conditions, registered references and protected request paths remain validated. + +Composer `profile_policy_composer_v6` participates in catalog revisions. Saving +revalidates the same draft before freezing the outlet; connection checks run after +saving. Existing outlets continue using their stored snapshot without recomposition. diff --git a/sdk/nexent/core/model_service/profiles/manifest.toml b/sdk/nexent/core/model_service/profiles/manifest.toml new file mode 100644 index 0000000000..cf61b9ae68 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/manifest.toml @@ -0,0 +1,342 @@ +schema_version = 1 +source_schema_ref = "models_dev_authored_v1" +upstream_url = "https://github.com/anomalyco/models.dev" +upstream_revision = "e69e1844210e5cfe7c7e7f68d57328bc7451a51c" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.5-27b.toml" +content_sha256 = "15e1c86123b937d3152f69e87650f3b3bbfd9600f8bbfbc5361a6f5e9f69fc16" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.5-35b-a3b.toml" +content_sha256 = "50a139d95791c9c5e145f4a0f115f18c3d38840202c7683e8861f05fe4ffc6be" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.5-397b-a17b.toml" +content_sha256 = "d064d5959f3dca64beb7108a38280877719e89709db7775f932a5ee791e3924e" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.5-flash.toml" +content_sha256 = "39a4329d9772432e58cf92044f143bd7e1445b434ee65830acad958b72322d5c" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.5-plus.toml" +content_sha256 = "15c6b3b7f334219ee6d43c76d1a004e57469e4a4e780e4b254c91474d5cb1bd4" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.8-27b.toml" +content_sha256 = "5a7c7ec1b513c3268756c02fa240f09ba438bce594cbb7d34ebb4f71c4a448e0" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.8-2.4t-a95b.toml" +content_sha256 = "067a17c3a34382a3e286414e8f6ef4ad31844716ed85af41fdab7061ba44add4" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.8-flash.toml" +content_sha256 = "7fc097dbfdcbcf827fc28c17e23bafda6d695ce5d16d96fb65612e4cae581dcc" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.8-max.toml" +content_sha256 = "ee87c9af3b4188b86dc8ee3d045e9a18a0b611b5db1e47cfc3f88b1672164634" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3-coder-next.toml" +content_sha256 = "1f4175a40524b382f1e9efac28e4454b477a2f4ccf5a460134b9ed12cb829229" + +[[sources]] +kind = "model" +path = "models/zhipuai/glm-5.1.toml" +content_sha256 = "8c7069a86e0ef764d2a1d0ebb30156abed0cf7b27e5b972c224a5cd9aee27b40" + +[[sources]] +kind = "model" +path = "models/zhipuai/glm-5.2.toml" +content_sha256 = "6ed303f1feeaed5496e919b6c7ae999e2743505a22ed97062b5c833adf37ce13" + +[[sources]] +kind = "model" +path = "models/zhipuai/glm-5.3.toml" +content_sha256 = "8022f2c087c0ad8bdf9f8ed4fe950fda5f7f37ef2a121c577e87fbe28d159c29" + +[[sources]] +kind = "model" +path = "models/zhipuai/glm-5.3-flash.toml" +content_sha256 = "7cdb7a1cfea9ab7e7c3e3392f0fdad0ac9c3f9c92e2f71a2d109788c5d02f843" + +[[sources]] +kind = "model" +path = "models/deepseek/deepseek-v4-flash.toml" +content_sha256 = "02ae90fe9be99047c373f3862afe72367c7a96a7aef7f43864c0bcd9e22da194" + +[[sources]] +kind = "model" +path = "models/deepseek/deepseek-v4-pro.toml" +content_sha256 = "bd08e3df5404ff5f795ac91b44ef34ac7c29a6911de0bd9fded3739501e47996" + +[[sources]] +kind = "model" +path = "models/deepseek/deepseek-v4-flash-0731.toml" +content_sha256 = "cb0133fb9f376da44fc6ee512c8a3ba0399626bb026805f608f265e504511c4e" + +[[sources]] +kind = "model" +path = "models/deepseek/deepseek-v4-pro-0813.toml" +content_sha256 = "e005b286813f8bcf39ecc4397b36a99895532c97bbc2c83f21a94dbaa85455ea" + +[[sources]] +kind = "model" +path = "models/deepseek/deepseek-v4.1-flash.toml" +content_sha256 = "ef594fbed1c96863ff728e31530dbc1987d8862e36c5f1b5f08e5376d631aee0" + +[[sources]] +kind = "model" +path = "models/minimax/MiniMax-M2.7.toml" +content_sha256 = "c63b15b0f281e217f466b973439cfd1876b23d636288cd4f7102ef28a57df761" + +[[sources]] +kind = "model" +path = "models/minimax/MiniMax-M3.toml" +content_sha256 = "a1d815dc34328bd0719fc557794d696205a7506a06515f8e2d84c24f3bce1b1e" + +[[sources]] +kind = "provider" +path = "providers/alibaba-cn/provider.toml" +content_sha256 = "0060172cf93a8db8b8427021030a3ad5c388c919f7fc47df9723ba8b08d5f54b" + +[[sources]] +kind = "provider" +path = "providers/volcengine/provider.toml" +content_sha256 = "b504a2ab81fccb036105dde4a626f486d82e37ceaf348249baaf0b5c8e234377" + +[[sources]] +kind = "provider" +path = "providers/volcengine-coding-plan/provider.toml" +content_sha256 = "15439498a42304ac424f73810fdf8593ee87d19c05bff271c64383afe6b06ee8" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/qwen3.5-397b-a17b.toml" +content_sha256 = "59342059889eac4ab4829384b954db58c72104f9ffc682441d7b39b319b80fb9" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/qwen3.5-flash.toml" +content_sha256 = "48d81041f3babeb002b73184e151df0bb4304b1679137c12e396951bbe9c16fe" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/qwen3.5-plus.toml" +content_sha256 = "5bd06ed609ab5ceeeecf8012a4ea587d5365fcd49c17f10bfb729f1a436fee0e" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/qwen3.8-flash.toml" +content_sha256 = "f496c5f293e2196f69bcd2f705cbb12548f14a1923d6b35d10400cbc7a1c817b" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/qwen3.8-max.toml" +content_sha256 = "1b1b47992952fb615bdce07fdeea93bdf445b36375acae6e5e812e6aecb9bd08" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/glm-5.1.toml" +content_sha256 = "f49b1f19c24db0d339b7bafb8c909c15bdaf5dc3765f9befef66b1166e10c520" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/glm-5.2.toml" +content_sha256 = "dbfbc406418c60aa33f33eed147f8d47a98f544b6f5cb50bf96e20fb5c38a2c2" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/glm-5.3.toml" +content_sha256 = "6ce1f84a6cd1af13f823e846b750e4f39d90aa4f008178091a118f88765dfe2f" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/deepseek-v4-flash.toml" +content_sha256 = "b7f6b5d17332d596883c8d5ac7de1c9e885dc4df8e9fc2ea996420e202a7febb" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/deepseek-v4-pro.toml" +content_sha256 = "db39defdd8ed98cf27ccb03f9884431c1e52f1b5d6a88591888d5801bb1e0ac3" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/deepseek-v4.1-flash.toml" +content_sha256 = "fe4c026f77d2a2b837d390dcef3da521b5dc3298eafb1e06199360cc0c6cd5d0" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/MiniMax/MiniMax-M2.7.toml" +content_sha256 = "d14f335b07a27d36e697d350d2aac28d6f9215e6ad4c2912ff9fe8bd5ac7c2f4" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine/models/deepseek-v4-flash-ga-260731.toml" +content_sha256 = "8012bc8a96ff1a96a56b64721e5e144d4ff997b70985c34483b95c7469de13c1" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine/models/deepseek-v4-pro-ga-260813.toml" +content_sha256 = "444e6d130d11e3132d34e3afe4d878f96120beef51db5946f206d96bc8dfdec8" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine/models/glm-5-2-260617.toml" +content_sha256 = "092fcf694c283b90a71cde982d4231d63e4a53544665d4a481e306b4cb1bed1a" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine/models/glm-5-3-flash-260828.toml" +content_sha256 = "34b35e1742e663e54d1231baa5282fc1497b255d60764df21e018c8a78da04a2" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine-coding-plan/models/deepseek-v4-flash.toml" +content_sha256 = "be24a803ea861218c4ad9fc60025a424cb275b95e7c02507699776ce37869a99" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine-coding-plan/models/deepseek-v4-pro.toml" +content_sha256 = "b359870df413e10509083d9b56a62453c530e0f600ca5628ae3279043df2ee57" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine-coding-plan/models/glm-5.3.toml" +content_sha256 = "6fb6ebfb16a106cfb22e5f887326d661b96c1b67df4f826fdf6d02cceded4782" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine-coding-plan/models/glm-5.3-flash.toml" +content_sha256 = "8ca9644166cb0cdc479049a0f7467a6cebded93cd0d3bef00a66e462b77111f3" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine-coding-plan/models/minimax-m3.toml" +content_sha256 = "c2998f64873231e71f62be2da7983afd4c1af76e7a129f232007e65a58c24e24" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.7-max.toml" +content_sha256 = "45e9afba10e6b6164914fff61758d748643d2b64a4be0bb54afc4fc33e91cb86" + +[[sources]] +kind = "model" +path = "models/alibaba/qwen3.7-plus.toml" +content_sha256 = "a6929b11d1652ee69c1c43dbe81f83beb959bcef9366deb4bc92118e8fca6fa2" + +[[sources]] +kind = "model" +path = "models/bytedance-seed/seed-2.1-pro.toml" +content_sha256 = "e617184014aea6221241eb84ed1cabda001564c3dd8e7a3878fded7915433915" + +[[sources]] +kind = "model" +path = "models/moonshotai/kimi-k3.toml" +content_sha256 = "6391cda163821bd0f6ae07b8ad26baa0d9c356da5e0584c37dd64b721a1a5513" + +[[sources]] +kind = "model" +path = "models/zhipuai/glm-5.toml" +content_sha256 = "efce81254124e5a529f9dc59fb8969fcb3b26189e0f292a17c03f8b60712f881" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/glm-5.toml" +content_sha256 = "f422ddc5053de0db7134e21b4dc1491004b787dc660b95c81ef84c83bff79d63" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/kimi-k3.toml" +content_sha256 = "d2622cbdc17360a200a292468ae8c4192a97b1e6191a4f811014511e23ba57a3" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/qwen3.7-max.toml" +content_sha256 = "b9cec088349a5b29fefb61d6cc471495f7f1a9a9c7b82747abfb94425c1f1610" + +[[sources]] +kind = "provider_model" +path = "providers/alibaba-cn/models/qwen3.7-plus.toml" +content_sha256 = "d0c2a5b15e0bd7ac8c140a142063558c9fa43ead040fe99513f3851f532e0203" + +[[sources]] +kind = "provider_model" +path = "providers/deepseek/models/deepseek-flash.toml" +content_sha256 = "d5e811b4af967dd239b008c62b55c7924de2bdb36cd53c75324c98a42796db8d" + +[[sources]] +kind = "provider_model" +path = "providers/deepseek/models/deepseek-v4-flash.toml" +content_sha256 = "b1910724c0240130f4676a90719ee321aa16d73ee354dbefb2ade27ce2e0c917" + +[[sources]] +kind = "provider_model" +path = "providers/deepseek/models/deepseek-v4-pro.toml" +content_sha256 = "0e4d752d27934327150916fbf5b62fa17b21a3fa6857a460988f89a16c7072a5" + +[[sources]] +kind = "provider" +path = "providers/deepseek/provider.toml" +content_sha256 = "3f2fbc840041ce2d9dc1a44cf9a7b82011e312bc8e71d973b3a3113f601817c2" + +[[sources]] +kind = "provider_model" +path = "providers/siliconflow-cn/models/Pro/zai-org/GLM-5.1.toml" +content_sha256 = "eca714ad1e4ca589ab5a065e10b1dbf0ded78241a8cf2d3614127f971731a963" + +[[sources]] +kind = "provider_model" +path = "providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml" +content_sha256 = "5947afc8c482ae9582875efb8bcbea2245eec898f92838e03ae36bfa192c206e" + +[[sources]] +kind = "provider_model" +path = "providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml" +content_sha256 = "925beb91bea20c55ecc4752e3e2e712785ec1e11468853587f1d5fe688b55b4c" + +[[sources]] +kind = "provider_model" +path = "providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Pro.toml" +content_sha256 = "59f2b4a5da97e4daf237ce295908258487ca592d1428158937e0b657981a43bf" + +[[sources]] +kind = "provider_model" +path = "providers/siliconflow-cn/models/zai-org/GLM-5.2.toml" +content_sha256 = "4967aa9c9948ce58b52d65599cb54ee7288ca80932f86b59b6bdae0842ab873d" + +[[sources]] +kind = "provider" +path = "providers/siliconflow-cn/provider.toml" +content_sha256 = "805d38042cf90ca3d94f944d3e218dde1b3f5babdf657d677d3ef4df566bae36" + +[[sources]] +kind = "provider" +path = "providers/tokenpony/provider.toml" +content_sha256 = "6eb8ed0f1b8d5af7887edaa983ade429d872e10ca495487d45cbda49ae4c8fd0" +origin = "nexent" +source_revision = "nexent_tokenpony_provider_v1" +source_url = "https://docs.tokenpony.cn/" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine-coding-plan/models/kimi-k3.toml" +content_sha256 = "dd4ea5240577cfe3a6dfc7860eb1e6cf558349ec562aa83bcd82519831368e0c" + +[[sources]] +kind = "provider_model" +path = "providers/volcengine/models/doubao-seed-2-1-pro-260628.toml" +content_sha256 = "4a9deef53960bd716d4b6078adfcc05719b9ef3491c1c41363ee31a3e4173bf4" diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3-coder-next.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3-coder-next.toml new file mode 100644 index 0000000000..cce2c7bb62 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3-coder-next.toml @@ -0,0 +1,27 @@ +# https://qwen.ai/blog?id=qwen3-coder-next +# https://huggingface.co/Qwen/Qwen3-Coder-Next +# https://www.qwencloud.com/models/qwen3-coder-next +name = "Qwen3 Coder Next" +description = "Open-weight Qwen coding model for agents, repository edits, and multi-turn tool use" +family = "qwen" +release_date = "2026-02-03" +last_updated = "2026-02-03" +attachment = false +reasoning = false +temperature = true +tool_call = true +structured_output = true +knowledge = "2025-09" +open_weights = true + +[limit] +context = 262_144 +output = 65_536 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/Qwen/Qwen3-Coder-Next" diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-27b.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-27b.toml new file mode 100644 index 0000000000..93ad247dd8 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-27b.toml @@ -0,0 +1,29 @@ +name = "Qwen3.5 27B" +description = "Qwen vision-language model for visual reasoning, documents, and agent tasks" +family = "qwen" +release_date = "2026-02-23" +last_updated = "2026-02-23" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[limit] +context = 262_144 +output = 65_536 + +[modalities] +input = ["text", "image", "video", "audio"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/Qwen/Qwen3.5-27B" + +[[benchmarks]] +name = "SWE-Bench Verified" +score = 72.4 +metric = "resolved" +source = "https://huggingface.co/Qwen/Qwen3.5-27B" diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-35b-a3b.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-35b-a3b.toml new file mode 100644 index 0000000000..36700f959e --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-35b-a3b.toml @@ -0,0 +1,23 @@ +name = "Qwen3.5 35B-A3B" +description = "Qwen vision-language model for visual reasoning, documents, and agent tasks" +family = "qwen" +release_date = "2026-02-23" +last_updated = "2026-02-23" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[limit] +context = 262_144 +output = 65_536 + +[modalities] +input = ["text", "image", "video", "audio"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/Qwen/Qwen3.5-35B-A3B" diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-397b-a17b.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-397b-a17b.toml new file mode 100644 index 0000000000..103227c16b --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-397b-a17b.toml @@ -0,0 +1,29 @@ +name = "Qwen3.5 397B-A17B" +description = "Large open Qwen multimodal MoE for visual agents and long technical tasks" +family = "qwen" +release_date = "2026-02-15" +last_updated = "2026-02-15" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[limit] +context = 262_144 +output = 65_536 + +[modalities] +input = ["text", "image", "video", "audio"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/Qwen/Qwen3.5-397B-A17B" + +[[benchmarks]] +name = "SWE-Bench Verified" +score = 76.4 +metric = "resolved" +source = "https://huggingface.co/Qwen/Qwen3.5-397B-A17B" diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-flash.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-flash.toml new file mode 100644 index 0000000000..800ff85f43 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-flash.toml @@ -0,0 +1,22 @@ +# https://help.aliyun.com/en/model-studio/qwen3-5-flash +# https://www.alibabacloud.com/help/en/model-studio/deep-thinking +name = "Qwen3.5 Flash" +description = "Qwen vision-language model for visual reasoning, documents, and agent tasks" +family = "qwen" +release_date = "2026-02-23" +last_updated = "2026-02-23" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = false + + +[limit] +context = 1_000_000 +output = 65_536 + +[modalities] +input = ["text", "image", "video"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-plus.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-plus.toml new file mode 100644 index 0000000000..7f04f4c2fe --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.5-plus.toml @@ -0,0 +1,19 @@ +name = "Qwen3.5 Plus" +description = "Qwen vision-language model for visual reasoning, documents, and agent tasks" +family = "qwen" +release_date = "2026-02-16" +last_updated = "2026-02-16" +attachment = false +reasoning = true +temperature = true +tool_call = true +knowledge = "2025-04" +open_weights = false + +[limit] +context = 1_000_000 +output = 65_536 + +[modalities] +input = ["text", "image", "video"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.7-max.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.7-max.toml new file mode 100644 index 0000000000..65b09455d6 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.7-max.toml @@ -0,0 +1,82 @@ +name = "Qwen3.7 Max" +description = "Qwen frontier model tuned for agent frameworks, coding assistants, and long tasks" +family = "qwen" +release_date = "2026-05-21" +last_updated = "2026-05-21" +attachment = false +reasoning = true +temperature = true +tool_call = true +open_weights = false + +[limit] +context = 1_000_000 +output = 65_536 + +[modalities] +input = ["text"] +output = ["text"] + +[[benchmarks]] +name = "SWE-Bench Verified" +score = 80.4 +metric = "resolved" +source = "https://qwen.ai/blog?id=qwen3.7" +date = "2026-05-19" + +[[benchmarks]] +name = "SWE-Bench Pro" +score = 60.6 +metric = "resolve rate" +source = "https://qwen.ai/blog?id=qwen3.7" +date = "2026-05-19" + +[[benchmarks]] +name = "SWE-Bench Multilingual" +score = 78.3 +metric = "resolve rate" +source = "https://qwen.ai/blog?id=qwen3.7" +date = "2026-05-19" + +[[benchmarks]] +name = "Terminal-Bench" +score = 69.7 +metric = "success rate" +harness = "Terminus-2" +version = "2.0" +source = "https://qwen.ai/blog?id=qwen3.7" +date = "2026-05-19" + +[[benchmarks]] +name = "GPQA Diamond" +score = 92.4 +metric = "accuracy" +source = "https://qwen.ai/blog?id=qwen3.7" +date = "2026-05-19" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 41.4 +metric = "accuracy" +source = "https://qwen.ai/blog?id=qwen3.7" +date = "2026-05-19" + +[[benchmarks]] +name = "SciCode" +score = 53.5 +source = "https://qwen.ai/blog?id=qwen3.7" +date = "2026-05-19" + +[[benchmarks]] +name = "MCP Atlas" +score = 76.4 +metric = "success rate" +source = "https://qwen.ai/blog?id=qwen3.7" +date = "2026-05-19" + +[[benchmarks]] +name = "NL2Repo" +score = 47.2 +harness = "Claude Code" +source = "https://qwen.ai/blog?id=qwen3.7" +date = "2026-05-19" diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.7-plus.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.7-plus.toml new file mode 100644 index 0000000000..42ce95a219 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.7-plus.toml @@ -0,0 +1,19 @@ +name = "Qwen3.7 Plus" +description = "Multimodal Qwen workhorse for long-context agents, visual inputs, and coding" +family = "qwen" +release_date = "2026-06-02" +last_updated = "2026-06-02" +attachment = true +reasoning = true +temperature = true +tool_call = true +knowledge = "2025-04" +open_weights = false + +[limit] +context = 1_000_000 +output = 64_000 + +[modalities] +input = ["text", "image", "video"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-2.4t-a95b.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-2.4t-a95b.toml new file mode 100644 index 0000000000..b30a03ea6b --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-2.4t-a95b.toml @@ -0,0 +1,33 @@ +# Sources (accessed 2026-08-16): +# https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B +# https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/raw/main/README.md +# https://qwen.ai/blog?id=qwen3.8 +# https://openrouter.ai/qwen/qwen3.8-2.4t-a95b +# Open-weight twin of Qwen3.8 Max: text-only, thinking always on, +# reasoning_effort low|medium|xhigh (default xhigh). Native context 262K, +# extensible to ~1.01M. Distinct from closed multimodal qwen3.8-max. + +name = "Qwen3.8 2.4T A95B" +description = "Open-weight sparse MoE (2.4T total, 95B active), the open-weight twin of Qwen3.8 Max for coding, research, complex reasoning, and agentic workflows" +family = "qwen" +release_date = "2026-08-12" +last_updated = "2026-08-12" +attachment = false +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true +license = "qwen3.8-max" + +[limit] +context = 262_144 +output = 131_072 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-27b.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-27b.toml new file mode 100644 index 0000000000..4517f55223 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-27b.toml @@ -0,0 +1,236 @@ +# Benchmark caveats and evaluation harnesses: https://huggingface.co/Qwen/Qwen3.8-27B#evaluation +# Sources (accessed 2026-08-15): +# https://huggingface.co/Qwen/Qwen3.8-27B +# https://huggingface.co/api/models/Qwen/Qwen3.8-27B +# https://qwen.ai/blog?id=qwen3.8 +# Hub lastModified 2026-08-14T15:00:01Z is the open-weight drop. +# Do not use Hub createdAt 2026-08-05 (staged countdown page). + +name = "Qwen3.8 27B" +description = "Dense 27B vision-language model for coding, agent tasks, and image and video understanding" +family = "qwen" +release_date = "2026-08-14" +last_updated = "2026-08-14" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[limit] +context = 262_144 +output = 32_768 + +[modalities] +input = ["text", "image", "video"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "SWE-bench Pro" +score = 61.7 +metric = "resolved" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "Terminal-Bench" +score = 73 +metric = "score" +version = "2.1" +harness = "Terminus" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "NL2Repo" +score = 42.3 +metric = "score" +harness = "Claude Code" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "DeepSWE" +score = 42.2 +metric = "score" +version = "1.1" +harness = "Claude Code" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "QwenSWEBench" +score = 79 +metric = "avg@3" +harness = "Claude Code" +dataset = "internal" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "CoWorkBench" +score = 70.7 +metric = "score" +dataset = "internal" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "JobBench" +score = 33.4 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "Agents' Last Exam" +score = 20.4 +metric = "pass@1" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "Agents' Last Exam" +score = 42.9 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "IFBench" +score = 79.5 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "GPQA Diamond" +score = 89.2 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 30.8 +metric = "score" +variant = "without tools; GPT-4o judge" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "LiveCodeBench" +score = 90.3 +metric = "score" +version = "6" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "OSWorld-Verified" +score = 84.3 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "WebArena-Verified" +score = 64.8 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" +harness = "OSWorld scaffold; official WebArena-Verified grader" + +[[benchmarks]] +name = "AndroidWorld" +score = 81.9 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "SWE-MM" +score = 38.6 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "RealWorldQA" +score = 85.9 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "ERQA" +score = 65.5 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "RecreationBench" +score = 47.1 +metric = "score" +dataset = "internal" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "ClawEval-MM" +score = 57.4 +metric = "pass@3" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "ClawEval-MM" +score = 56.9 +metric = "average score" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "Vision2Web" +score = 62.9 +metric = "score" +harness = "Claude Code" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" +variant = "gpt-5.4-2026-03-05 judge" + +[[benchmarks]] +name = "MathVision" +score = 90 +metric = "score" +variant = "without code interpreter" +dataset = "corrected annotations" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "MathVision" +score = 94.6 +metric = "score" +variant = "with code interpreter" +dataset = "corrected annotations" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "BabyVision" +score = 65.7 +metric = "score" +variant = "without code interpreter" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "BabyVision" +score = 85.6 +metric = "score" +variant = "with code interpreter" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" + +[[benchmarks]] +name = "CharXiv Reasoning" +score = 83.7 +metric = "score" +variant = "without code interpreter" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" +dataset = "corrected annotations" + +[[benchmarks]] +name = "CharXiv Reasoning" +score = 90.2 +metric = "score" +variant = "with code interpreter" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" +dataset = "corrected annotations" + +[[benchmarks]] +name = "OmniDocBench" +score = 91.1 +metric = "score" +version = "1.5" +source = "https://huggingface.co/Qwen/Qwen3.8-27B" diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-flash.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-flash.toml new file mode 100644 index 0000000000..2ee8a5ff53 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-flash.toml @@ -0,0 +1,19 @@ +# Source: https://www.qwencloud.com/models/qwen3.8-flash +name = "Qwen3.8 Flash" +description = "Qwen vision-language model for visual reasoning, documents, and agent tasks" +family = "qwen" +release_date = "2026-08-26" +last_updated = "2026-08-26" +attachment = true +reasoning = true +tool_call = true +structured_output = true +open_weights = false + +[limit] +context = 1_000_000 +output = 131_072 + +[modalities] +input = ["text", "image", "video"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-max.toml b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-max.toml new file mode 100644 index 0000000000..76a956fd3f --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/alibaba/qwen3.8-max.toml @@ -0,0 +1,266 @@ +# These are the Qwen3.8-Max column of the official model card, not a claim about a separately evaluated local checkpoint. +# Sources (accessed 2026-08-06): +# https://www.qwencloud.com/models/qwen3.8-max +# https://www.qianwenai.com/models/qwen3.8-max +# https://help.aliyun.com/zh/model-studio/qwen3-8-max +# https://www.alibabacloud.com/help/en/model-studio/qwen3-8-max +# https://help.aliyun.com/zh/model-studio/pdf-understanding +# https://platform.qianwenai.com/docs/developer-guides/tool-calling/pdf-understanding +# https://docs.qwencloud.com/token-plan/personal/token-plan-personal-overview +# https://help.aliyun.com/zh/model-studio/token-plan-personal-overview +# https://help.aliyun.com/en/model-studio/token-plan-personal-overview +# https://docs.qwencloud.com/developer-guides/getting-started/text-generation-models +# https://docs.qwencloud.com/developer-guides/text-generation/thinking +# https://docs.qwencloud.com/developer-guides/clients-and-developer-tools/opencode +# https://platform.qianwenai.com/docs/developer-guides/clients-and-developer-tools/opencode +# https://qwen.ai/blog?id=qwen3.8 +# PDF input: Model Studio / 千问AI docs list only qwen3.8-max under PDF理解 +# (type:file / file_url|file_data). Model pages list Image/Text/Video badges +# and separately list PDF理解 as a Completions built-in tool. Beijing-region +# availability note on help.aliyun.com; lab capability still includes pdf. + +name = "Qwen3.8 Max" +description = "2.4-trillion-parameter MoE flagship for coding, professional work, multimodal understanding, and long-horizon agentic workflows" +family = "qwen" +release_date = "2026-08-03" +last_updated = "2026-08-03" +attachment = true +reasoning = true +temperature = true +tool_call = true +open_weights = false + +[limit] +context = 1_000_000 +output = 131_072 + +[modalities] +input = ["text", "image", "video", "pdf"] +output = ["text"] + +[[benchmarks]] +name = "Terminal-Bench" +score = 86.6 +metric = "avg@10" +version = "2.1" +harness = "Claude Code" +variant = "5h timeout" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "SWE-Bench Pro" +score = 67.7 +metric = "resolved" +harness = "Claude Code" +dataset = "Qwen refined and corrected task set" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "DeepSWE" +score = 56.6 +metric = "score" +version = "1.1" +harness = "Claude Code" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "NL2Repo" +score = 55.9 +metric = "score" +harness = "Claude Code" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "FrontierSWE" +score = 73.5 +metric = "dominance score" +date = "2026-08-03" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "MLS-Bench-Lite" +score = 41 +metric = "score" +harness = "Claude Code" +variant = "5h timeout" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "PaperBench" +score = 93 +metric = "score" +harness = "BasicAgent" +variant = "Code-Dev; 3 runs; 12h timeout; Opus 4.6 judge" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "AndroidBench" +score = 75.1 +metric = "avg@3" +dataset = "95 public tasks" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "QwenSWEBench" +score = 80.7 +metric = "avg@3" +harness = "Claude Code" +dataset = "internal" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "QwenQoderBench" +score = 58.4 +metric = "avg@5" +harness = "Claude Code" +dataset = "internal" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "QwenReactBench" +score = 1724 +metric = "Elo" +dataset = "internal" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "QwenSVGBench" +score = 1713 +metric = "Elo" +dataset = "internal" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "CoWorkBench" +score = 74.8 +metric = "score" +dataset = "internal" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "WorkSpaceBench" +score = 67.7 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "JobBench" +score = 53.4 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "SkillsBench" +score = 70.2 +metric = "avg@3" +version = "1.1" +dataset = "87 tasks" +harness = "OpenCode" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "Agents' Last Exam" +score = 27 +metric = "pass@1" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "Agents' Last Exam" +score = 52.4 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "AutomationBench" +score = 27.3 +metric = "pass@1" +dataset = "600 public tasks" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "Toolathlon-Verified" +score = 72.5 +metric = "pass@1" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "WideSearch" +score = 81.9 +metric = "average item F1" +harness = "Qwen-Agent" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 56.2 +metric = "score" +variant = "with tools" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 43.6 +metric = "score" +variant = "without tools" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "GPQA Diamond" +score = 92.6 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "IFBench" +score = 82.8 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "$OneMillion-Bench" +score = 52.5 +metric = "expert score" +variant = "gemini-3.1-pro-preview judge" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "HealthBench" +score = 60.2 +metric = "score" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "PLawBench" +score = 73.2 +metric = "score" +variant = "gemini-3.1-pro-preview judge" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "PRBench" +score = 57.6 +metric = "score" +variant = "legal" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "PRBench" +score = 58.3 +metric = "score" +variant = "finance" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "MRCR" +score = 92.9 +metric = "score" +version = "2" +dataset = "256K; 8-needle" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" + +[[benchmarks]] +name = "LongBench" +score = 66.3 +metric = "score" +version = "2" +source = "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" diff --git a/sdk/nexent/core/model_service/profiles/models/bytedance-seed/seed-2.1-pro.toml b/sdk/nexent/core/model_service/profiles/models/bytedance-seed/seed-2.1-pro.toml new file mode 100644 index 0000000000..0f7dea4d8b --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/bytedance-seed/seed-2.1-pro.toml @@ -0,0 +1,22 @@ +# Sources (accessed 2026-08-14): +# - https://seed.bytedance.com/en/seed2 +# - https://www.volcengine.com/docs/82379/1330310 +name = "Seed 2.1 Pro" +description = "Flagship ByteDance Seed 2.1 model for complex multimodal reasoning, coding, and agents" +family = "seed" +release_date = "2026-06-23" +last_updated = "2026-06-23" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = false + +[limit] +context = 256_000 +output = 256_000 + +[modalities] +input = ["text", "image", "video"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-flash-0731.toml b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-flash-0731.toml new file mode 100644 index 0000000000..61479ce86a --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-flash-0731.toml @@ -0,0 +1,105 @@ +name = "DeepSeek V4 Flash 0731" +description = "Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding" +family = "deepseek-flash" +release_date = "2026-07-31" +last_updated = "2026-07-31" +attachment = false +reasoning = true +temperature = true +tool_call = true +structured_output = true +knowledge = "2025-05" +open_weights = true +license = "MIT" + +[limit] +context = 1_000_000 +output = 384_000 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731" + +[[benchmarks]] +name = "Terminal-Bench" +score = 82.7 +metric = "pass@1" +variant = "max" +version = "2.1" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731" + +[[benchmarks]] +name = "NL2Repo" +score = 54.2 +metric = "resolve rate" +variant = "max effort" +harness = "DeepSeek Harness minimal mode" +source = "https://api-docs.deepseek.com/updates/" +date = "2026-07-31" + +[[benchmarks]] +name = "CyberGym" +score = 76.7 +metric = "score" +variant = "max effort" +harness = "DeepSeek Harness minimal mode" +source = "https://api-docs.deepseek.com/updates/" +date = "2026-07-31" + +[[benchmarks]] +name = "DeepSWE" +score = 54.4 +metric = "resolve rate" +variant = "max effort" +harness = "DeepSeek Harness minimal mode" +source = "https://api-docs.deepseek.com/updates/" +date = "2026-07-31" + +[[benchmarks]] +name = "Toolathlon-Verified" +score = 70.3 +metric = "score" +variant = "max effort" +harness = "DeepSeek Harness minimal mode" +source = "https://api-docs.deepseek.com/updates/" +date = "2026-07-31" + +[[benchmarks]] +name = "Agents' Last Exam" +score = 25.2 +metric = "score" +variant = "max effort" +harness = "DeepSeek Harness minimal mode" +source = "https://api-docs.deepseek.com/updates/" +date = "2026-07-31" + +[[benchmarks]] +name = "AutomationBench" +score = 25.1 +metric = "success rate" +variant = "max effort" +dataset = "public" +source = "https://api-docs.deepseek.com/updates/" +date = "2026-07-31" + +[[benchmarks]] +name = "DSBench-FullStack" +score = 68.7 +metric = "score" +variant = "max effort" +dataset = "internal" +source = "https://api-docs.deepseek.com/updates/" +date = "2026-07-31" + +[[benchmarks]] +name = "DSBench-Hard" +score = 59.6 +metric = "score" +variant = "max effort" +dataset = "internal" +source = "https://api-docs.deepseek.com/updates/" +date = "2026-07-31" diff --git a/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-flash.toml b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-flash.toml new file mode 100644 index 0000000000..5f521d77d2 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-flash.toml @@ -0,0 +1,183 @@ +# Added benchmarks are from the original V4 preview instruct-model comparison, not the later 0731/0813 releases. +name = "DeepSeek V4 Flash" +description = "Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work" +family = "deepseek-flash" +release_date = "2026-04-24" +last_updated = "2026-04-24" +attachment = false +reasoning = true +temperature = true +tool_call = true +structured_output = true +knowledge = "2025-05" +open_weights = true + +[limit] +context = 1_000_000 +output = 384_000 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "SWE-Bench Verified" +score = 79 +metric = "resolved" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "MMLU-Pro" +score = 86.2 +metric = "EM" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "SimpleQA-Verified" +score = 34.1 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "Chinese SimpleQA" +score = 78.9 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "GPQA Diamond" +score = 88.1 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 34.8 +metric = "pass@1" +variant = "preview checkpoint; max effort; without tools" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "LiveCodeBench" +score = 91.6 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "Codeforces" +score = 3052 +metric = "rating" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "HMMT" +score = 94.8 +metric = "pass@1" +variant = "preview checkpoint; max effort" +dataset = "February 2026" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "IMOAnswerBench" +score = 88.4 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "MathArena Apex" +score = 33 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "MathArena Apex Shortlist" +score = 85.7 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "MRCR" +score = 78.7 +metric = "MMR" +variant = "preview checkpoint; max effort" +dataset = "1M context" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "CorpusQA" +score = 60.5 +metric = "accuracy" +variant = "preview checkpoint; max effort" +dataset = "1M context" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 56.9 +metric = "accuracy" +variant = "preview checkpoint; max effort" +version = "2.0" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "SWE-Bench Pro" +score = 52.6 +metric = "resolved" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "SWE-Bench Multilingual" +score = 73.3 +metric = "resolved" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "BrowseComp" +score = 73.2 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 45.1 +metric = "pass@1" +variant = "preview checkpoint; max effort; with tools" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "MCP Atlas" +score = 69 +metric = "pass@1" +variant = "preview checkpoint; max effort" +dataset = "public" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "GDPval-AA" +score = 1395 +metric = "Elo" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" + +[[benchmarks]] +name = "Toolathlon" +score = 47.8 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" diff --git a/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-pro-0813.toml b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-pro-0813.toml new file mode 100644 index 0000000000..dade234dbb --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-pro-0813.toml @@ -0,0 +1,106 @@ +# Public code-agent benchmarks use DeepSeek Harness minimal mode, max effort, temperature=1.0, top_p=0.95. +# https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813 +name = "DeepSeek V4 Pro 0813" +description = "DeepSeek V4 Pro snapshot with million-token context and support for thinking and non-thinking modes" +family = "deepseek-thinking" +release_date = "2026-08-12" +last_updated = "2026-08-22" +attachment = false +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true +license = "MIT" + +[limit] +context = 1_000_000 +output = 384_000 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 42.7 +metric = "score" +variant = "without tools" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 60 +metric = "score" +variant = "with tools" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "Terminal-Bench" +score = 87.9 +metric = "score" +harness = "DeepSeek Harness minimal mode" +variant = "max effort" +version = "2.1" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "NL2Repo" +score = 61.5 +metric = "score" +harness = "DeepSeek Harness minimal mode" +variant = "max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "CyberGym" +score = 83.3 +metric = "score" +harness = "DeepSeek Harness minimal mode" +variant = "max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "DeepSWE" +score = 62.7 +metric = "score" +harness = "DeepSeek Harness minimal mode" +variant = "max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "Toolathlon-Verified" +score = 74.1 +metric = "score" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "Agents' Last Exam" +score = 25.7 +metric = "score" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "AutomationBench" +score = 31.8 +metric = "score" +dataset = "public" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "DSBench-FullStack" +score = 71.1 +metric = "score" +dataset = "internal" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" + +[[benchmarks]] +name = "DSBench-Hard" +score = 67.2 +metric = "score" +dataset = "internal" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" diff --git a/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-pro.toml b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-pro.toml new file mode 100644 index 0000000000..c0bb4099ca --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4-pro.toml @@ -0,0 +1,217 @@ +# Added benchmarks are from the original V4 preview instruct-model comparison, not the later 0731/0813 releases. +name = "DeepSeek V4 Pro" +description = "Open MoE flagship with million-token context for coding and long agent runs" +family = "deepseek-thinking" +release_date = "2026-04-24" +last_updated = "2026-04-24" +attachment = false +reasoning = true +temperature = true +tool_call = true +structured_output = true +knowledge = "2025-05" +open_weights = true + +[limit] +context = 1_000_000 +output = 384_000 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "SWE-Bench Verified" +score = 80.6 +metric = "resolved" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "Artificial Analysis Coding Agent Index" +score = 50.1 +metric = "average pass@1" +harness = "Claude Code" +variant = "high" +source = "https://artificialanalysis.ai/agents/coding-agents" + +[[benchmarks]] +name = "SWE-Atlas Codebase QnA" +score = 67.8 +metric = "pass@1" +harness = "Claude Code" +variant = "high" +source = "https://artificialanalysis.ai/agents/coding-agents" + +[[benchmarks]] +name = "SWE-Bench Pro" +score = 18 +metric = "pass@1" +harness = "Claude Code" +variant = "high" +dataset = "hard-aa" +source = "https://artificialanalysis.ai/agents/coding-agents" + +[[benchmarks]] +name = "Terminal-Bench" +score = 64.7 +metric = "pass@1" +harness = "Claude Code" +variant = "high" +version = "2.1" +source = "https://artificialanalysis.ai/agents/coding-agents" + +[[benchmarks]] +name = "MMLU-Pro" +score = 87.5 +metric = "EM" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "SimpleQA-Verified" +score = 57.9 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "Chinese SimpleQA" +score = 84.4 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "GPQA Diamond" +score = 90.1 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 37.7 +metric = "pass@1" +variant = "preview checkpoint; max effort; without tools" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "LiveCodeBench" +score = 93.5 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "Codeforces" +score = 3206 +metric = "rating" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "HMMT" +score = 95.2 +metric = "pass@1" +variant = "preview checkpoint; max effort" +dataset = "February 2026" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "IMOAnswerBench" +score = 89.8 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "MathArena Apex" +score = 38.3 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "MathArena Apex Shortlist" +score = 90.2 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "MRCR" +score = 83.5 +metric = "MMR" +variant = "preview checkpoint; max effort" +dataset = "1M context" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "CorpusQA" +score = 62 +metric = "accuracy" +variant = "preview checkpoint; max effort" +dataset = "1M context" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "Terminal-Bench" +score = 67.9 +metric = "accuracy" +variant = "preview checkpoint; max effort" +version = "2.0" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "SWE-Bench Pro" +score = 55.4 +metric = "resolved" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "SWE-Bench Multilingual" +score = 76.2 +metric = "resolved" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "BrowseComp" +score = 83.4 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 48.2 +metric = "pass@1" +variant = "preview checkpoint; max effort; with tools" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "MCP Atlas" +score = 73.6 +metric = "pass@1" +variant = "preview checkpoint; max effort" +dataset = "public" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "GDPval-AA" +score = 1554 +metric = "Elo" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" + +[[benchmarks]] +name = "Toolathlon" +score = 51.8 +metric = "pass@1" +variant = "preview checkpoint; max effort" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" diff --git a/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4.1-flash.toml b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4.1-flash.toml new file mode 100644 index 0000000000..16a1174cf7 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/deepseek/deepseek-v4.1-flash.toml @@ -0,0 +1,311 @@ +# Benchmarks use instruct weights, temperature=1.0, top_p=0.95; not the base-model table. +# Source: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash#instruct-model +# Open weights (MIT): https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash +name = "DeepSeek V4.1 Flash" +description = "DeepSeek V4.1 Flash model for reasoning and agentic coding" +family = "deepseek-flash" +release_date = "2026-09-10" +last_updated = "2026-09-10" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +knowledge = "2025-05" +open_weights = true +license = "MIT" + +[limit] +context = 1_000_000 +output = 384_000 + +[modalities] +input = ["text", "image"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "GPQA Diamond" +score = 90.9 +metric = "pass@1" +variant = "reasoning_effort=100" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "MathArena Apex" +score = 65.6 +metric = "pass@1" +variant = "reasoning_effort=100" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 36.8 +metric = "pass@1" +variant = "reasoning_effort=100" +dataset = "full set" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 39.1 +metric = "pass@1" +variant = "reasoning_effort=100" +dataset = "text-only subset" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Codeforces" +score = 3471 +metric = "rating" +variant = "reasoning_effort=100" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 90.6 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "2.1" +harness = "DeepSeek Harness minimal mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 30.0 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "3.0" +harness = "DeepSeek Harness minimal mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 31.2 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "4.0" +harness = "DeepSeek Harness minimal mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "DeepSWE" +score = 74.2 +metric = "resolved" +variant = "reasoning_effort=100" +version = "1.1" +harness = "mini-swe-agent" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "ProgramBench" +score = 20.3 +metric = "almost@1" +variant = "reasoning_effort=100" +harness = "DeepSeek Harness minimal mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "NL2Repo" +score = 64.0 +metric = "score" +variant = "reasoning_effort=100" +harness = "DeepSeek Harness minimal mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "CyberGym" +score = 88.1 +metric = "pass@1" +variant = "reasoning_effort=100" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "ExploitGym" +score = 15.3 +metric = "pass@1" +variant = "reasoning_effort=100" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "AutomationBench" +score = 54.8 +metric = "pass@1" +variant = "reasoning_effort=100" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Agents' Last Exam" +score = 31.8 +metric = "pass@1" +variant = "reasoning_effort=100" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "SEC-Bench Pro" +score = 62.8 +metric = "pass@1" +variant = "reasoning_effort=100" +harness = "Claude Code" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 63.9 +metric = "pass@1" +variant = "reasoning_effort=100; with tools" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Chartography" +score = 78.9 +metric = "pass@1" +variant = "reasoning_effort=100; with tools" +harness = "Claude Code" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "BabyVision" +score = 89.6 +metric = "pass@1" +variant = "reasoning_effort=100; with tools" +harness = "Claude Code" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "ZeroBench" +score = 49.0 +metric = "pass@5" +variant = "reasoning_effort=100; with tools" +harness = "Claude Code" +dataset = "main" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "DeepSWE" +score = 69.8 +metric = "resolved" +variant = "reasoning_effort=100" +version = "1.1" +harness = "Claude Code" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 88.0 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "2.1" +harness = "Claude Code" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "DeepSWE" +score = 65.6 +metric = "resolved" +variant = "reasoning_effort=100" +version = "1.1" +harness = "Codex" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 84.1 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "2.1" +harness = "Codex" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "DeepSWE" +score = 65.5 +metric = "resolved" +variant = "reasoning_effort=100" +version = "1.1" +harness = "OpenCode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 85.0 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "2.1" +harness = "OpenCode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "DeepSWE" +score = 66.2 +metric = "resolved" +variant = "reasoning_effort=100" +version = "1.1" +harness = "Pi" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 86.1 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "2.1" +harness = "Pi" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 90.3 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "2.1" +harness = "mini-swe-agent" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "DeepSWE" +score = 72.6 +metric = "resolved" +variant = "reasoning_effort=100" +version = "1.1" +harness = "DeepSeek Harness minimal mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "DeepSWE" +score = 70.5 +metric = "resolved" +variant = "reasoning_effort=100" +version = "1.1" +harness = "DeepSeek Harness standard mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 85.8 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "2.1" +harness = "DeepSeek Harness standard mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "DeepSWE" +score = 67.6 +metric = "resolved" +variant = "reasoning_effort=100" +version = "1.1" +harness = "DeepSeek Harness PTC mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" + +[[benchmarks]] +name = "Terminal-Bench" +score = 85.8 +metric = "pass@1" +variant = "reasoning_effort=100" +version = "2.1" +harness = "DeepSeek Harness PTC mode" +source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash" diff --git a/sdk/nexent/core/model_service/profiles/models/minimax/MiniMax-M2.7.toml b/sdk/nexent/core/model_service/profiles/models/minimax/MiniMax-M2.7.toml new file mode 100644 index 0000000000..e72eaabdce --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/minimax/MiniMax-M2.7.toml @@ -0,0 +1,46 @@ +name = "MiniMax-M2.7" +description = "Open MiniMax flagship for coding agents, office automation, and complex environments" +family = "minimax" +release_date = "2026-03-18" +last_updated = "2026-03-18" +attachment = false +reasoning = true +temperature = true +tool_call = true +open_weights = true + +[limit] +context = 204_800 +output = 131_072 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/MiniMaxAI/MiniMax-M2.7" + +[[benchmarks]] +name = "SWE-Bench Verified" +score = 79.9 +metric = "resolved" +harness = "Claude Code" +source = "https://www.minimax.io/blog/minimax-m3" +date = "2026-06-01" + +[[benchmarks]] +name = "SWE-Bench Pro" +score = 56.2 +metric = "resolve rate" +harness = "Claude Code" +source = "https://www.minimax.io/blog/minimax-m3" +date = "2026-06-01" + +[[benchmarks]] +name = "Terminal-Bench" +score = 51.1 +metric = "success rate" +version = "2.1" +source = "https://www.minimax.io/blog/minimax-m3" +date = "2026-06-01" diff --git a/sdk/nexent/core/model_service/profiles/models/minimax/MiniMax-M3.toml b/sdk/nexent/core/model_service/profiles/models/minimax/MiniMax-M3.toml new file mode 100644 index 0000000000..31ad8eeb00 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/minimax/MiniMax-M3.toml @@ -0,0 +1,67 @@ +name = "MiniMax-M3" +description = "MiniMax multimodal model for long-context coding, perception, and agent planning" +family = "minimax" +release_date = "2026-06-01" +last_updated = "2026-06-01" +attachment = true +reasoning = true +temperature = true +tool_call = true +open_weights = true + +[limit] +context = 1_048_576 +output = 512_000 + +[modalities] +input = ["text", "image", "video"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/MiniMaxAI/MiniMax-M3" + +[[benchmarks]] +name = "SWE-Bench Verified" +score = 80.5 +metric = "resolved" +harness = "Claude Code" +source = "https://www.minimax.io/blog/minimax-m3" +date = "2026-06-01" + +[[benchmarks]] +name = "SWE-Bench Pro" +score = 59.0 +metric = "resolve rate" +harness = "Claude Code" +source = "https://www.minimax.io/blog/minimax-m3" +date = "2026-06-01" + +[[benchmarks]] +name = "Terminal-Bench" +score = 66.0 +metric = "success rate" +version = "2.1" +source = "https://www.minimax.io/blog/minimax-m3" +date = "2026-06-01" + +[[benchmarks]] +name = "BrowseComp" +score = 83.52 +metric = "accuracy" +source = "https://www.minimax.io/blog/minimax-m3" +date = "2026-06-01" + +[[benchmarks]] +name = "MCP Atlas" +score = 74.2 +metric = "success rate" +source = "https://www.minimax.io/blog/minimax-m3" +date = "2026-06-01" + +[[benchmarks]] +name = "OSWorld-Verified" +score = 70.06 +metric = "success rate" +source = "https://www.minimax.io/blog/minimax-m3" +date = "2026-06-01" diff --git a/sdk/nexent/core/model_service/profiles/models/moonshotai/kimi-k3.toml b/sdk/nexent/core/model_service/profiles/models/moonshotai/kimi-k3.toml new file mode 100644 index 0000000000..52ace80d25 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/moonshotai/kimi-k3.toml @@ -0,0 +1,135 @@ +name = "Kimi K3" +description = "Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work" +family = "kimi-k3" +release_date = "2026-07-16" +last_updated = "2026-07-16" +attachment = true +reasoning = true +temperature = false +tool_call = true +structured_output = true +open_weights = true + +[limit] +context = 1_048_576 +output = 131_072 + +[modalities] +input = ["text", "image", "video"] +output = ["text"] + +[[benchmarks]] +name = "DeepSWE" +score = 67.5 +metric = "resolve rate" +variant = "max effort" +harness = "Kimi Code" +version = "1.1" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "Terminal-Bench" +score = 88.3 +metric = "accuracy" +variant = "max effort" +harness = "Kimi Code" +version = "2.1" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "FrontierSWE" +score = 81.2 +metric = "dominance score" +variant = "max effort" +harness = "Kimi Code" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "Program Bench" +score = 77.8 +metric = "score" +variant = "max effort" +harness = "Kimi Code" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "SWE Marathon" +score = 42.0 +metric = "resolve rate" +variant = "max effort" +harness = "Claude Code" +version = "1.1" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "GDPval-AA" +score = 1668 +metric = "Elo" +variant = "max effort" +version = "v2" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "AA-Briefcase" +score = 1548 +metric = "Elo" +variant = "max effort" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "AutomationBench" +score = 30.8 +metric = "success rate" +variant = "max effort" +dataset = "600-task public subset" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "JobBench" +score = 52.9 +metric = "score" +variant = "max effort" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "SpreadsheetBench" +score = 34.8 +metric = "score" +variant = "max effort" +harness = "Claude Code" +version = "2" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "BrowseComp" +score = 91.2 +metric = "accuracy" +variant = "max effort, context compaction" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "CharXiv Reasoning" +score = 91.3 +metric = "accuracy" +variant = "max effort, with tools" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" + +[[benchmarks]] +name = "ZeroBench" +score = 41.0 +metric = "pass@5" +variant = "max effort, with tools" +source = "https://www.kimi.com/blog/kimi-k3" +date = "2026-07-16" diff --git a/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.1.toml b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.1.toml new file mode 100644 index 0000000000..61860d5017 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.1.toml @@ -0,0 +1,53 @@ +name = "GLM-5.1" +description = "Strong GLM coding model for agentic engineering, terminals, and repository generation" +family = "glm" +release_date = "2026-04-07" +last_updated = "2026-04-07" +attachment = false +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[limit] +context = 200_000 +output = 131_072 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/zai-org/GLM-5.1" + +[[benchmarks]] +name = "Artificial Analysis Coding Agent Index" +score = 52.7 +metric = "average pass@1" +harness = "Claude Code" +source = "https://artificialanalysis.ai/agents/coding-agents" + +[[benchmarks]] +name = "SWE-Atlas Codebase QnA" +score = 73.2 +metric = "pass@1" +harness = "Claude Code" +source = "https://artificialanalysis.ai/agents/coding-agents" + +[[benchmarks]] +name = "SWE-Bench Pro" +score = 19.8 +metric = "pass@1" +harness = "Claude Code" +dataset = "hard-aa" +source = "https://artificialanalysis.ai/agents/coding-agents" + +[[benchmarks]] +name = "Terminal-Bench" +score = 65.1 +metric = "pass@1" +harness = "Claude Code" +version = "2.1" +source = "https://artificialanalysis.ai/agents/coding-agents" diff --git a/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.2.toml b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.2.toml new file mode 100644 index 0000000000..292a8dff6d --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.2.toml @@ -0,0 +1,167 @@ +name = "GLM-5.2" +description = "Open flagship GLM for long-horizon coding agents and million-token context work" +family = "glm" +release_date = "2026-06-13" +last_updated = "2026-06-13" +attachment = false +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[limit] +context = 1_000_000 +output = 131_072 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/zai-org/GLM-5.2" + +[[benchmarks]] +name = "SWE-Bench Pro" +score = 62.1 +metric = "resolve rate" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "Terminal-Bench" +score = 82.7 +metric = "success rate" +harness = "Claude Code" +version = "2.1" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "FrontierSWE" +score = 74.4 +metric = "dominance" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 40.5 +metric = "accuracy" +dataset = "text-only subset" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 54.7 +metric = "accuracy" +variant = "with tools" +dataset = "text-only subset" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "CritPt" +score = 20.9 +metric = "accuracy" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "AIME" +score = 99.2 +metric = "accuracy" +version = "2026" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "HMMT" +score = 94.4 +metric = "accuracy" +version = "November 2025" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "HMMT" +score = 92.5 +metric = "accuracy" +version = "February 2026" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "IMOAnswerBench" +score = 91.0 +metric = "accuracy" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "GPQA Diamond" +score = 91.2 +metric = "accuracy" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "NL2Repo" +score = 48.9 +metric = "resolve rate" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "DeepSWE" +score = 46.2 +metric = "resolve rate" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "Program Bench" +score = 63.7 +metric = "score" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "Terminal-Bench" +score = 81.0 +metric = "success rate" +harness = "Terminus 2" +version = "2.1" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "PostTrainBench" +score = 34.3 +metric = "score" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "SWE Marathon" +score = 13.0 +metric = "resolve rate" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "MCP Atlas" +score = 76.8 +metric = "score" +dataset = "public subset" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" + +[[benchmarks]] +name = "Tool-Decathlon" +score = 48.2 +metric = "score" +source = "https://z.ai/blog/glm-5.2" +date = "2026-06-16" diff --git a/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.3-flash.toml b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.3-flash.toml new file mode 100644 index 0000000000..6f713eea56 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.3-flash.toml @@ -0,0 +1,70 @@ +# Scores: https://raw.githubusercontent.com/zai-org/GLM-5/refs/heads/main/resources/bench_53.png +# Open weights: https://huggingface.co/zai-org/GLM-5.3-Flash (MIT, safetensors) +name = "GLM-5.3-Flash" +description = "Native multimodal GLM model for efficient coding and long-horizon agent tasks" +family = "glm-flash" +release_date = "2026-08-26" +last_updated = "2026-08-26" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[limit] +context = 1_000_000 +output = 131_072 + +[modalities] +input = ["text", "image", "video", "pdf"] +output = ["text"] + +[[benchmarks]] +name = "Terminal-Bench" +score = 84.3 +metric = "pass@1" +variant = "max effort" +version = "2.1" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3-Flash" + +[[benchmarks]] +name = "DeepSWE" +score = 63.4 +metric = "resolved" +variant = "max effort" +version = "1.1" +harness = "mini-swe-agent" +source = "https://huggingface.co/zai-org/GLM-5.3-Flash" + +[[benchmarks]] +name = "Agents' Last Exam" +score = 26.3 +metric = "score" +variant = "max effort" +source = "https://huggingface.co/zai-org/GLM-5.3-Flash" + +[[benchmarks]] +name = "AutomationBench" +score = 48.8 +metric = "pass@1" +variant = "max effort" +version = "1.0.6" +source = "https://huggingface.co/zai-org/GLM-5.3-Flash" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 55.3 +metric = "accuracy" +variant = "max effort; with tools" +dataset = "full set" +source = "https://huggingface.co/zai-org/GLM-5.3-Flash" + +[[benchmarks]] +name = "GDPval-AA" +score = 1773 +metric = "Elo" +variant = "max effort" +version = "2" +source = "https://huggingface.co/zai-org/GLM-5.3-Flash" diff --git a/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.3.toml b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.3.toml new file mode 100644 index 0000000000..e8c11c17ae --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.3.toml @@ -0,0 +1,158 @@ +# Benchmark settings and caveats: https://huggingface.co/zai-org/GLM-5.3#footnotes +name = "GLM-5.3" +description = "Flagship GLM model for long-horizon coding, agents, and complex project delivery" +family = "glm" +release_date = "2026-08-14" +last_updated = "2026-08-14" +attachment = false +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[limit] +context = 1_000_000 +output = 131_072 + +[modalities] +input = ["text"] +output = ["text"] + +[[benchmarks]] +name = "Terminal-Bench" +score = 88.2 +metric = "pass@1" +variant = "max effort" +version = "2.1" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "Terminal-Bench" +score = 28.3 +metric = "avg@3" +variant = "max effort" +version = "3.0" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "DeepSWE" +score = 66.9 +metric = "resolved" +variant = "max effort; 6h timeout; 400K context" +version = "1.1" +source = "https://huggingface.co/zai-org/GLM-5.3" +harness = "mini-swe-agent" + +[[benchmarks]] +name = "NL2Repo" +score = 58.0 +metric = "score" +variant = "max effort" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "ProgramBench" +score = 19.0 +metric = "almost solved" +variant = "max effort" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "FrontierSWE" +score = 78.1 +metric = "dominance score" +variant = "max effort" +date = "2026-08-14" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "SWE-Marathon" +score = 42.5 +metric = "score" +variant = "max effort; modified anti-cheat checks" +version = "1.1" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "PostTrainBench" +score = 39.8 +metric = "weighted average" +variant = "max effort; 3 runs; modified external-API checks" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "CyberGym" +score = 84.5 +metric = "pass@1" +variant = "max effort" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "ExploitGym" +score = 105 +metric = "tasks solved" +variant = "max effort; 2h rescaled timeout" +dataset = "869 tasks" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "ExploitGym" +score = 130 +metric = "tasks solved" +variant = "max effort; 6h rescaled timeout" +dataset = "869 tasks" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "ExploitBench" +score = 54.4 +metric = "average coverage" +variant = "max effort; union over 3 revisions" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "Toolathlon-Verified" +score = 73 +metric = "pass@1" +variant = "max effort" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "AutomationBench" +score = 48.2 +metric = "pass@1" +variant = "max effort" +version = "1.0.6" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "Agents' Last Exam" +score = 28.5 +metric = "score" +variant = "max effort; CLI" +harness = "Claude Code 2.1.207" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "Humanity's Last Exam" +score = 62.5 +metric = "accuracy" +variant = "max effort; with tools" +source = "https://huggingface.co/zai-org/GLM-5.3" + +[[benchmarks]] +name = "GDPval-AA" +score = 1769 +metric = "Elo" +variant = "max effort" +version = "2" +source = "https://huggingface.co/zai-org/GLM-5.3" diff --git a/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.toml b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.toml new file mode 100644 index 0000000000..111cc20f89 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/models/zhipuai/glm-5.toml @@ -0,0 +1,49 @@ +name = "GLM-5" +description = "General GLM flagship for coding, analysis, and tool-heavy engineering workflows" +family = "glm" +release_date = "2026-02-12" +last_updated = "2026-02-12" +attachment = false +reasoning = true +temperature = true +tool_call = true +open_weights = true + +[limit] +context = 204_800 +output = 131_072 + +[modalities] +input = ["text"] +output = ["text"] + +[[weights]] +label = "Hugging Face" +url = "https://huggingface.co/zai-org/GLM-5" + +[[benchmarks]] +name = "SWE-Bench Verified" +score = 72.8 +metric = "resolved" +source = "https://www.swebench.com/" + +[[benchmarks]] +name = "SWE-Atlas Codebase QnA" +score = 20.5 +metric = "score" +harness = "Mini-SWE-Agent" +source = "https://labs.scale.com/leaderboard/sweatlas-qna" + +[[benchmarks]] +name = "SWE-Atlas Refactoring" +score = 24.24 +metric = "score" +harness = "Mini-SWE-Agent" +source = "https://labs.scale.com/leaderboard/sweatlas-refactoring" + +[[benchmarks]] +name = "SWE-Atlas Test Writing" +score = 28.74 +metric = "score" +harness = "Mini-SWE-Agent" +source = "https://labs.scale.com/leaderboard/sweatlas-tw" diff --git a/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/MiniMax/MiniMax-M2.7.toml b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/MiniMax/MiniMax-M2.7.toml new file mode 100644 index 0000000000..80d471831f --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/MiniMax/MiniMax-M2.7.toml @@ -0,0 +1,25 @@ +name = "MiniMax-M2.7" +description = "MiniMax model for chat, coding, office work, and agentic tasks" +family = "minimax" +release_date = "2026-03-18" +last_updated = "2026-03-18" +attachment = false +reasoning = true +temperature = true +tool_call = true +open_weights = true +reasoning_options = [] + +[cost] +input = 0.30 +output = 1.20 +cache_read = 0.06 +cache_write = 0.375 + +[limit] +context = 204_800 +output = 131_072 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4-flash.toml b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4-flash.toml new file mode 100644 index 0000000000..f37c5ab26f --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4-flash.toml @@ -0,0 +1,16 @@ +base_model = "deepseek/deepseek-v4-flash" + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.14 +output = 0.28 +cache_read = 0.0028 diff --git a/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4-pro.toml b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4-pro.toml new file mode 100644 index 0000000000..722857a529 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4-pro.toml @@ -0,0 +1,16 @@ +base_model = "deepseek/deepseek-v4-pro" + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.435 +output = 0.87 +cache_read = 0.003625 diff --git a/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4.1-flash.toml b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4.1-flash.toml new file mode 100644 index 0000000000..6fe0ab8159 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/deepseek-v4.1-flash.toml @@ -0,0 +1,27 @@ +# CNY→USD rate: 6.721845, 2026-09-18, source: https://open.er-api.com/v6/latest/USD +# Sources (accessed 2026-09-18): +# https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/deepseek-v4.1-flash?serviceSite=asia-pacific-china +# https://help.aliyun.com/zh/model-studio/deepseek-api — capability table, +# reasoning_effort ladder, max_tokens default (393_216, shared with thinking) +# https://help.aliyun.com/zh/model-studio/billing-for-model-studio — Beijing +# peak/off-peak pricing: CNY 2/1 input, 8/4 output per 1M tokens with context +# cache discount; peak price converted to USD using the rate above (off-peak +# is half). +# Hybrid reasoning: enable_thinking = true|false toggles thinking; +# reasoning_effort = low|high|max (default high) — low is only supported on +# v4.1-flash / v4-flash-0731 / v4-pro-0813, and v4.1-flash is the current +# low-tier bearer. reasoning_content streams in deltas. Responses API +# supported (Beijing and Singapore only). Limits and modalities are identical +# to the deepseek lab entry (1M context, 384_000 output, text+image input), +# so they are inherited and not restated. +base_model = "deepseek/deepseek-v4.1-flash" + +reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "high", "max"] }] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.29754 +output = 1.19015 +cache_read = 0.01488 diff --git a/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.1.toml b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.1.toml new file mode 100644 index 0000000000..e19b35f1e8 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.1.toml @@ -0,0 +1,45 @@ +# Chinese mainland deployment is tiered on input tokens per request; every +# token in a request bills at its tier's unit price. The 0.87/3.48 previously +# published here matches neither band; both bands below are taken from the +# vendor page. No cache-read price is published for either band, so the +# existing curated cache_read is retained on the base and not restated. +# https://www.alibabacloud.com/help/en/model-studio/model-pricing (accessed 2026-08-26) +name = "GLM-5.1" +description = "Flagship GLM model for hybrid reasoning, coding, and agentic engineering" +family = "glm" +release_date = "2026-04-14" +last_updated = "2026-04-14" +attachment = false +reasoning = true +structured_output = true +temperature = true +tool_call = true +open_weights = true + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "budget_tokens" +max = 131_072 + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.825 +output = 3.301 +cache_read = 0.17 + +[[cost.tiers]] +tier = { type = "context", size = 32_000 } +input = 1.1 +output = 3.851 + +[limit] +context = 202_752 +output = 128_000 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.2.toml b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.2.toml new file mode 100644 index 0000000000..b847c801b2 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.2.toml @@ -0,0 +1,22 @@ +base_model = "zhipuai/glm-5.2" + +reasoning_options = [] + +[interleaved] +field = "reasoning_content" + +# First party price citation here +# https://bailian.console.alibabacloud.com/cn-beijing?tab=model#/model-market/detail/glm-5.2?serviceSite=asia-pacific-china +[cost] +input = 1.1 +output = 3.851 +cache_read = 0.275 +cache_write = 0 + +[limit] +context = 1000000 +output = 128_000 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.3.toml b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.3.toml new file mode 100644 index 0000000000..f6d0f6e67e --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.3.toml @@ -0,0 +1,25 @@ +# GLM-5.3 on Alibaba Bailian (Model Studio, Beijing): always-on thinking with +# reasoning_effort low|high|max (default max). Thinking cannot be turned off +# (enable_thinking only accepts true; passing false errors) and there is no +# thinking_budget — so a single effort option, no toggle, no budget. +# reasoning_content is streamed in deltas. +# Pricing: Beijing CNY 8 / 28 / 2 (cache hit) per 1M tokens, same list price +# as glm-5.2 on the same host (see providers/alibaba-cn/models/glm-5.2.toml). +# Limits and modalities are identical to the zhipuai lab entry +# (1M context, 131_072 output, text-only), so they are inherited and not restated. +# https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/glm-5.3?serviceSite=asia-pacific-china +# https://docs.bigmodel.cn/cn/guide/models/text/glm-5.3 +base_model = "zhipuai/glm-5.3" + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1.1 +output = 3.851 +cache_read = 0.275 +cache_write = 0 diff --git a/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.toml b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.toml new file mode 100644 index 0000000000..5d48f4a4ce --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/alibaba-cn/models/glm-5.toml @@ -0,0 +1,41 @@ +# Chinese mainland deployment is tiered on input tokens per request; every +# token in a request bills at its tier's unit price. The 0.86/3.15 previously +# published here is the 32K low (unlike Pro). xhigh -> high. +# https://api-docs.deepseek.com/quick_start/pricing (accessed 2026-09-10) +# https://api-docs.deepseek.com/guides/thinking_mode/ (accessed 2026-09-10) +# Chat API max_tokens: 1..393216 (384K): https://api-docs.deepseek.com/api/create-chat-completion/ +base_model = "deepseek/deepseek-v4.1-flash" + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.15 +output = 0.6 +reasoning = 0.6 +cache_read = 0.003 + +[limit] +output = 393_216 diff --git a/sdk/nexent/core/model_service/profiles/providers/deepseek/models/deepseek-v4-flash.toml b/sdk/nexent/core/model_service/profiles/providers/deepseek/models/deepseek-v4-flash.toml new file mode 100644 index 0000000000..e8b1624445 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/deepseek/models/deepseek-v4-flash.toml @@ -0,0 +1,30 @@ +# Legacy alias. The DeepSeek-V4-Flash model is retired; requests to +# `deepseek-v4-flash` are still accepted but served by DeepSeek-V4.1-Flash and +# billed at the Flash price. +# Toggle: thinking.type = enabled|disabled +# Effort: reasoning_effort = low|high|max +# https://api-docs.deepseek.com/quick_start/pricing (accessed 2026-09-10) +# Served by deepseek-flash; Chat API max_tokens <= 393216: +# https://api-docs.deepseek.com/api/create-chat-completion/ +base_model = "deepseek/deepseek-v4.1-flash" +name = "DeepSeek V4 Flash" +status = "deprecated" + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.15 +output = 0.6 +reasoning = 0.6 +cache_read = 0.003 + +[limit] +output = 393_216 diff --git a/sdk/nexent/core/model_service/profiles/providers/deepseek/models/deepseek-v4-pro.toml b/sdk/nexent/core/model_service/profiles/providers/deepseek/models/deepseek-v4-pro.toml new file mode 100644 index 0000000000..9fadc4b121 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/deepseek/models/deepseek-v4-pro.toml @@ -0,0 +1,35 @@ +# Toggle: thinking.type = enabled|disabled on /chat/completions. +# Effort: reasoning_effort = low|high|max; default high. +# Anthropic: output_config.effort = low|high|max; budget ignored. +# The 2026-08-13 V4-Pro GA announcement explicitly introduces all three levels; +# the pricing page maps API model deepseek-v4-pro to DeepSeek-V4-Pro-0813. +# Pricing (off-peak, USD per 1M tokens): cache miss $0.66, cache hit $0.022, output $1.98; peak is 2x. +# Peak 01:00-04:00 and 06:00-10:00 UTC Mon-Fri excl. Chinese holidays; models.dev records off-peak like deepseek-flash. +# https://api-docs.deepseek.com/news/news260813/ (accessed 2026-09-20) +# https://api-docs.deepseek.com/guides/thinking_mode/ (accessed 2026-09-20) +# https://api-docs.deepseek.com/quick_start/pricing (accessed 2026-10-02) +# Chat API max_tokens: 1..393216 (384K): https://api-docs.deepseek.com/api/create-chat-completion/ +# Reasoning tokens are billed at the output rate (no separate CoT price). +# `completion_tokens_details.reasoning_tokens` is a subset of completion_tokens. +# https://api-docs.deepseek.com/quick_start/pricing/ (accessed 2026-08-12) +base_model = "deepseek/deepseek-v4-pro-0813" +name = "DeepSeek V4 Pro" + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.66 +output = 1.98 +reasoning = 1.98 +cache_read = 0.022 + +[limit] +output = 393_216 diff --git a/sdk/nexent/core/model_service/profiles/providers/deepseek/provider.toml b/sdk/nexent/core/model_service/profiles/providers/deepseek/provider.toml new file mode 100644 index 0000000000..76274ae9de --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/deepseek/provider.toml @@ -0,0 +1,12 @@ +name = "DeepSeek" +env = ["DEEPSEEK_API_KEY"] +npm = "@ai-sdk/openai-compatible" +# OpenAI Chat is POST `/chat/completions`: `thinking.type = enabled|disabled` +# and `reasoning_effort = low|high|max`. Both current models map minimal→low, +# medium/xhigh→high, and ultra→max; low/high/max remain distinct native levels. +# Anthropic Messages is POST `/anthropic/v1/messages`: `thinking.type` and +# `output_config.effort = low|high|max`; `thinking.budget_tokens` is ignored. +# https://api-docs.deepseek.com/guides/thinking_mode/ (accessed 2026-09-20) +# https://api-docs.deepseek.com/guides/anthropic_api (accessed 2026-06-25) +doc = "https://api-docs.deepseek.com/quick_start/pricing" +api = "https://api.deepseek.com" diff --git a/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.1.toml b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.1.toml new file mode 100644 index 0000000000..634f29d893 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.1.toml @@ -0,0 +1,29 @@ +name = "Pro/zai-org/GLM-5.1" +description = "Flagship GLM model for hybrid reasoning, coding, and agentic engineering" +family = "glm" +release_date = "2026-04-08" +last_updated = "2026-04-08" +attachment = false +reasoning = true +reasoning_options = [] +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1.4 +output = 4.4 +cache_read = 0.26 +cache_write = 0 + +[limit] +context = 205_000 +output = 205_000 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml new file mode 100644 index 0000000000..2bd7f6ee36 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml @@ -0,0 +1,27 @@ +name = "Pro/zai-org/GLM-5" +description = "Flagship GLM model for hybrid reasoning, coding, and agentic engineering" +family = "glm" +release_date = "2026-02-12" +last_updated = "2026-02-12" +attachment = false +reasoning = true +reasoning_options = [{ type = "toggle" }] +temperature = true +tool_call = true +structured_output = true +open_weights = true + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1 +output = 3.20 + +[limit] +context = 205_000 +output = 205_000 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml new file mode 100644 index 0000000000..cafd58c6df --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml @@ -0,0 +1,14 @@ +base_model = "deepseek/deepseek-v4-flash" +# thinking_budget = 128..32768. This model alone also accepts +# reasoning_effort = "high"|"max"; low|medium map to high and xhigh maps to +# max (accessed 2026-06-25). +# https://docs.siliconflow.cn/cn/api-reference/chat-completions/chat-completions +reasoning_options = [{ type = "budget_tokens", min = 128, max = 32_768 }] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.14 +output = 0.28 +cache_read = 0.003 diff --git a/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Pro.toml b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Pro.toml new file mode 100644 index 0000000000..0e72e331f3 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Pro.toml @@ -0,0 +1,25 @@ +name = "deepseek-ai/DeepSeek-V4-Pro" +description = "Flagship DeepSeek model for coding, reasoning, and agentic work" +family = "deepseek-thinking" +release_date = "2026-04-24" +last_updated = "2026-04-24" +attachment = false +reasoning = true +reasoning_options = [] +temperature = true +tool_call = true +structured_output = false +open_weights = true + +[cost] +input = 1.74 +output = 3.48 +cache_read = 0.145 + +[limit] +context = 1_049_000 +output = 393_000 + +[modalities] +input = ["text"] +output = ["text"] diff --git a/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/zai-org/GLM-5.2.toml b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/zai-org/GLM-5.2.toml new file mode 100644 index 0000000000..1a6b9e2674 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/models/zai-org/GLM-5.2.toml @@ -0,0 +1,15 @@ +base_model = "zhipuai/glm-5.2" +reasoning_options = [{ type = "effort", values = ["high", "max"] }] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1.4 +output = 4.4 +cache_read = 0.26 +cache_write = 0 + +[limit] +context = 1_049_000 +output = 262_000 diff --git a/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/provider.toml b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/provider.toml new file mode 100644 index 0000000000..9f571e375e --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/siliconflow-cn/provider.toml @@ -0,0 +1,16 @@ +name = "SiliconFlow (China)" +env = ["SILICONFLOW_CN_API_KEY"] +npm = "@ai-sdk/openai-compatible" +api = "https://api.siliconflow.cn/v1" +# Reasoning HTTP format (accessed 2026-06-25): +# POST /v1/chat/completions uses top-level enable_thinking = true|false only +# for the schema's enumerated models, and thinking_budget = 128..32768 for +# most reasoning models. No zero/negative disable sentinel is documented. +# reasoning_effort is only for deepseek-ai/DeepSeek-V4-Flash: "high"|"max"; +# low|medium map to high and xhigh maps to max for compatibility. +# Model-specific listings may document additional controls; GLM-5.2 documents +# multiple thinking effort levels on SiliconFlow CN. +# Source: +# https://docs.siliconflow.cn/cn/api-reference/chat-completions/chat-completions +# https://cloud.siliconflow.cn/open/models?target=zai-org%2FGLM-5.2 +doc = "https://cloud.siliconflow.com/models" diff --git a/sdk/nexent/core/model_service/profiles/providers/tokenpony/provider.toml b/sdk/nexent/core/model_service/profiles/providers/tokenpony/provider.toml new file mode 100644 index 0000000000..da1188d989 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/tokenpony/provider.toml @@ -0,0 +1,8 @@ +# Nexent-authored provider metadata; not copied from models.dev. +# Source: https://docs.tokenpony.cn/ (checked 2026-10-16). +# This declares the OpenAI-compatible endpoint, not per-model qualification. +name = "TokenPony 小马算力" +env = ["TOKENPONY_API_KEY"] +npm = "@ai-sdk/openai-compatible" +api = "https://api.tokenpony.cn/v1" +doc = "https://docs.tokenpony.cn/" diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/deepseek-v4-flash.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/deepseek-v4-flash.toml new file mode 100644 index 0000000000..706a72594b --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/deepseek-v4-flash.toml @@ -0,0 +1,23 @@ +# Coding Plan model name from https://www.volcengine.com/docs/82379/1928261 +# (accessed 2026-08-26). Subscription tier: no per-token price. +# Toggle: thinking.type = enabled|disabled, POST /api/coding/v3/chat/completions +# (verified: enabled -> reasoning_tokens 57, disabled -> 0). +# No graded effort: the same body accepts reasoning_effort, but acceptance is not +# effect. Over n=5 runs per level on a prompt hard enough to consume the budget, +# none and minimal were hard zeros (the off switch) while the graded values fully +# overlapped -- low 659/873/216/487/845, medium 271/1548/556/296/1245, +# high 121/123/132/349/262, xhigh 928/753/268/480/110, max 811/435/570/910/623. +# 'high' sits below 'low', so the levels are not separable and only the toggle is +# declared, matching the toggle the lab (providers/deepseek) exposes. +base_model = "deepseek/deepseek-v4-flash" + +[[reasoning_options]] +type = "toggle" + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0 +output = 0 +cache_read = 0 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/deepseek-v4-pro.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/deepseek-v4-pro.toml new file mode 100644 index 0000000000..68e7561f92 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/deepseek-v4-pro.toml @@ -0,0 +1,22 @@ +# Coding Plan model name from https://www.volcengine.com/docs/82379/1928261 +# (accessed 2026-08-26). Subscription tier: no per-token price. +# Toggle: thinking.type = enabled|disabled, POST /api/coding/v3/chat/completions +# (verified: enabled -> reasoning_tokens 79, disabled -> 0). +# No graded effort: the same body accepts reasoning_effort, but acceptance is not +# effect. Over n=5 runs per level on a prompt hard enough to consume the budget, +# none and minimal were hard zeros (the off switch) while the graded values fully +# overlapped -- low 200/517/616/638/313, high 466/673/604/1075/415, +# max 389/204/240/1535/688. The levels are not separable, so only the toggle is +# declared, matching the toggle the lab (providers/deepseek) exposes. +base_model = "deepseek/deepseek-v4-pro" + +[[reasoning_options]] +type = "toggle" + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0 +output = 0 +cache_read = 0 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/glm-5.3-flash.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/glm-5.3-flash.toml new file mode 100644 index 0000000000..45dd0b497e --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/glm-5.3-flash.toml @@ -0,0 +1,17 @@ +# Coding Plan model name from https://www.volcengine.com/docs/82379/1928261 +# (accessed 2026-09-11): the model list includes both glm-5.3 (glm-latest) +# and glm-5.3-flash. Subscription tier: no per-token price. +# Effort levels mirror providers/zai-coding-plan/glm-5.3-flash.toml. +base_model = "zhipuai/glm-5.3-flash" + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0 +output = 0 +cache_read = 0 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/glm-5.3.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/glm-5.3.toml new file mode 100644 index 0000000000..8d75263dbe --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/glm-5.3.toml @@ -0,0 +1,30 @@ +# Coding Plan model name from https://www.volcengine.com/docs/82379/1928261 +# (accessed 2026-08-26). Subscription tier: no per-token price. +# Effort: POST /api/coding/v3/chat/completions accepts low|medium|high|xhigh|max, +# but only the three effective levels are declared. This model reports +# reasoning_tokens = 0 on every level, so the measure is reasoning_content length; +# over n=5 runs per level on a prompt hard enough to consume the thinking budget: +# low 0 0 0 0 0 medium 0 0 0 0 0 -> medium aliases low +# high 169 149 0 188 0 +# xhigh 2840 1943 3338 4655 6279 max 2363 4567 2962 4606 2327 +# -> xhigh aliases max +# Three distinct tiers, matching the lab (providers/zai) and the coding-plan peers. +# Volcengine's mapping table (82379/1449737) covers glm-5-2-260617 but not 5.3, and +# that row disagrees with what 5.3 does here (it maps low/medium to high, whereas +# 5.3 returns no reasoning at all on both), so these levels are set from the +# measurements above rather than from that table. +# No toggle: both reasoning_effort=none|minimal and thinking.type=disabled are +# rejected with 'not supported by this model'. +base_model = "zhipuai/glm-5.3" + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0 +output = 0 +cache_read = 0 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/kimi-k3.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/kimi-k3.toml new file mode 100644 index 0000000000..7b7fb0c9e7 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/kimi-k3.toml @@ -0,0 +1,28 @@ +# Coding Plan model name from https://www.volcengine.com/docs/82379/1928261 +# (accessed 2026-09-11), listed alongside kimi-k2.7-code. Subscription tier: +# no per-token price. +# reasoning_options mirror the lab (providers/moonshotai/models/kimi-k3.toml) +# and the kimi-code-plan-cn relay peer. Wire fields on this provider's OpenAI +# path (provider.toml): POST /api/coding/v3/chat/completions takes +# thinking.type = enabled|disabled plus reasoning_effort = low|high|max +# (`adaptive` is the Moonshot lab surface, not accepted here); the separate +# Anthropic path (/api/coding/v1/messages) uses output_config.effort. +# Endpoint-specific effort behavior on this relay was not measured; the +# baseline is copied per relay convention — adjust if wire probing shows +# levels collapsing as they do on kimi-k2.7-code (see that file). +base_model = "moonshotai/kimi-k3" + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0 +output = 0 +cache_read = 0 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/minimax-m3.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/minimax-m3.toml new file mode 100644 index 0000000000..60ca6e93c0 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/models/minimax-m3.toml @@ -0,0 +1,21 @@ +# Coding Plan model name from https://www.volcengine.com/docs/82379/1928261 +# (accessed 2026-08-26). Subscription tier: no per-token price. +# Toggle: thinking.type = enabled|disabled, POST /api/coding/v3/chat/completions +# Toggle only, matching first-party MiniMax and minimax-coding-plan. +# The same body also accepts reasoning_effort, but acceptance is not effect: +# over n=3 runs per level on a hard arithmetic prompt, reasoning_content length was +# minimal 1161/1033/805, high 1195/741/840, max 957/709/1088 -- fully overlapping, +# so no graded control is declared. 'none' is likewise accepted without disabling +# thinking (reasoning_tokens stays non-zero), so thinking.type is the real switch. +base_model = "minimax/MiniMax-M3" + +[[reasoning_options]] +type = "toggle" + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0 +output = 0 +cache_read = 0 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/provider.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/provider.toml new file mode 100644 index 0000000000..7420f6ed56 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine-coding-plan/provider.toml @@ -0,0 +1,42 @@ +name = "Volcengine Ark Coding Plan" +env = ["ARK_CODING_PLAN_API_KEY"] +npm = "@ai-sdk/openai-compatible" +# Coding Plan is a subscription tier with its own base URL. Volcengine warns that +# calling the pay-as-you-go https://ark.cn-beijing.volces.com/api/v3 with a Coding +# Plan key does NOT draw down the subscription and bills separately, so the two +# providers must stay distinct. +# The env var is deliberately not ARK_API_KEY: /api/coding/v3 does not check which +# plan a key belongs to, so a pay-as-you-go key reaches it and bills per token. +# Sharing ARK_API_KEY would light up both providers at once and let a user pick a +# Coding Plan entry while silently paying pay-as-you-go rates. +# Reasoning HTTP format (accessed 2026-08-26): +# OpenAI Chat: POST /api/coding/v3/chat/completions; `thinking.type` = enabled or +# disabled (`auto` is rejected on most models); `reasoning_effort` is accepted on +# every model but only some values have effect, so each file declares just the +# levels it honours, measured on this endpoint (n=4-6 per level, 2026-08-27): +# doubao-seed-2.1-turbo / 2.0-lite / evolving -> none|low|high +# (medium overlaps low, xhigh and max overlap high rather than exceeding it) +# glm-5.3 -> low|high|max +# (rejects none and minimal outright; medium aliases low, xhigh aliases max) +# deepseek-v4-flash / -pro -> toggle only +# (graded values fully overlap; on Flash 'high' lands below 'low') +# kimi-k2.7-code -> no controls +# minimax-m3 -> toggle only +# (tolerates none but still spends reasoning tokens on it, so it is not an +# off switch there; thinking.type is) +# Plaintext reasoning is in `choices[].message.reasoning_content` +# and the encrypted block in `choices[].message.encrypted_content`. +# Anthropic: base https://ark.cn-beijing.volces.com/api/coding, POST /v1/messages +# with `thinking.type` and `output_config.effort`. +# Model names here are the short console labels (doubao-seed-2.1-turbo). This +# endpoint is the more permissive of the two: it accepts the dated IDs as well +# (verified 2026-08-27 - doubao-seed-2-1-turbo-260628, doubao-seed-2-0-lite-260428 +# and glm-5-3-260930 all resolve here), whereas the pay-as-you-go /api/v3 accepts +# only the dated form and 404s on a short label. The short labels are used because +# they are what the Coding Plan console shows. `ark-code-latest` follows whichever +# model is selected in the console. +# Sources: +# https://www.volcengine.com/docs/82379/1928261 +# https://www.volcengine.com/docs/82379/2546386 +doc = "https://www.volcengine.com/docs/82379/1928261" +api = "https://ark.cn-beijing.volces.com/api/coding/v3" diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine/models/deepseek-v4-flash-ga-260731.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/deepseek-v4-flash-ga-260731.toml new file mode 100644 index 0000000000..85e6279908 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/deepseek-v4-flash-ga-260731.toml @@ -0,0 +1,21 @@ +# Model ID verified against POST /api/v3/chat/completions (2026-08-27): `deepseek-v4-flash-ga-260731` +# CNY→USD rate: 6.737012, 2026-08-26, source: https://open.er-api.com/v6/latest/USD +# CNY list price source: https://www.volcengine.com/docs/82379/1544106 +# Toggle: thinking.type = enabled|disabled +# Effort: reasoning_effort = minimal|low|medium|high +base_model = "deepseek/deepseek-v4-flash-0731" + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["minimal", "low", "medium", "high"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.4453 +output = 1.3359 +cache_read = 0.01484 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine/models/deepseek-v4-pro-ga-260813.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/deepseek-v4-pro-ga-260813.toml new file mode 100644 index 0000000000..369ff4ef40 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/deepseek-v4-pro-ga-260813.toml @@ -0,0 +1,21 @@ +# Model ID verified against POST /api/v3/chat/completions (2026-08-27): `deepseek-v4-pro-ga-260813` +# CNY→USD rate: 6.737012, 2026-08-26, source: https://open.er-api.com/v6/latest/USD +# CNY list price source: https://www.volcengine.com/docs/82379/1544106 +# Toggle: thinking.type = enabled|disabled +# Effort: reasoning_effort = minimal|low|medium|high +base_model = "deepseek/deepseek-v4-pro-0813" + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["minimal", "low", "medium", "high"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1.3359 +output = 4.00771 +cache_read = 0.04453 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine/models/doubao-seed-2-1-pro-260628.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/doubao-seed-2-1-pro-260628.toml new file mode 100644 index 0000000000..2fd899c3e7 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/doubao-seed-2-1-pro-260628.toml @@ -0,0 +1,19 @@ +# Model ID verified against POST /api/v3/chat/completions (2026-08-26): `doubao-seed-2-1-pro-260628` +# CNY→USD rate: 6.737012, 2026-08-26, source: https://open.er-api.com/v6/latest/USD +# CNY list price source: https://www.volcengine.com/docs/82379/1544106 +# Effort: reasoning_effort = none|minimal|low|medium|high|xhigh|max +# none/minimal disable thinking, so no separate toggle: sending effort alone +# already enables it, and thinking.type=disabled + a graded effort is rejected. +base_model = "bytedance-seed/seed-2.1-pro" + +[[reasoning_options]] +type = "effort" +values = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.8906 +output = 4.45301 +cache_read = 0.17812 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine/models/glm-5-2-260617.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/glm-5-2-260617.toml new file mode 100644 index 0000000000..64005aefe7 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/glm-5-2-260617.toml @@ -0,0 +1,21 @@ +# Model ID verified against POST /api/v3/chat/completions (2026-08-27): `glm-5-2-260617` +# CNY→USD rate: 6.737012, 2026-08-26, source: https://open.er-api.com/v6/latest/USD +# CNY list price source: https://www.volcengine.com/docs/82379/1544106 +# Toggle: thinking.type = enabled|disabled +# Effort: reasoning_effort = minimal|low|medium|high +base_model = "zhipuai/glm-5.2" + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["minimal", "low", "medium", "high"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1.18747 +output = 4.15615 +cache_read = 0.29687 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine/models/glm-5-3-flash-260828.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/glm-5-3-flash-260828.toml new file mode 100644 index 0000000000..98cb87ee9b --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine/models/glm-5-3-flash-260828.toml @@ -0,0 +1,18 @@ +# Model ID verified against POST /api/v3/chat/completions (2026-09-16): `glm-5-3-flash-260828` +# CNY→USD rate: 6.737012, 2026-08-26, source: https://open.er-api.com/v6/latest/USD +# CNY list price source: https://docs.volcengine.com/docs/82379/1449737 (深度思考) + 智谱官方价目表 +# GLM-5.3-Flash always reasons (thinking cannot be disabled on this model ID) +# Effort: reasoning_effort = low|high|max +base_model = "zhipuai/glm-5.3-flash" + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.11875 +output = 0.41563 +cache_read = 0.03414 diff --git a/sdk/nexent/core/model_service/profiles/providers/volcengine/provider.toml b/sdk/nexent/core/model_service/profiles/providers/volcengine/provider.toml new file mode 100644 index 0000000000..6ca805b6ab --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/providers/volcengine/provider.toml @@ -0,0 +1,23 @@ +name = "Volcengine Ark" +env = ["ARK_API_KEY"] +npm = "@ai-sdk/openai-compatible" +# Reasoning HTTP format (accessed 2026-08-26): +# OpenAI Chat: POST /api/v3/chat/completions; `thinking.type` = enabled|disabled +# (`auto` is documented but rejected with "Unsupported thinking type for the +# current model" on every model here except doubao-seed-1-6-flash), +# `reasoning_effort` = none|minimal|low|medium|high|xhigh|max (none and minimal +# disable thinking; doubao-seed-1-6-251015 accepts only the middle four). +# Plaintext reasoning is in +# `choices[].message.reasoning_content`, encrypted block is in +# `choices[].message.encrypted_content` (returned by default). +# Responses: POST /api/v3/responses; `reasoning.effort`; encrypted block requires +# explicit `include = ["reasoning.encrypted_content"]`. +# Anthropic: POST /api/v3/compatible/v1/messages; `thinking.type`, +# `output_config.effort` = low|high|max; plaintext thinking is in +# `content[thinking].thinking`, signature in `content[thinking].signature` (returned by default). +# Sources: +# https://www.volcengine.com/docs/82379/1298459 +# https://www.volcengine.com/docs/82379/1330310 +# https://www.volcengine.com/docs/82379/2636748 +doc = "https://www.volcengine.com/docs/82379/1330310" +api = "https://ark.cn-beijing.volces.com/api/v3" diff --git a/sdk/nexent/core/model_service/profiles/rules/manifest.toml b/sdk/nexent/core/model_service/profiles/rules/manifest.toml new file mode 100644 index 0000000000..6282a575e3 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/manifest.toml @@ -0,0 +1,133 @@ +schema_version = 1 + +[[rules]] +path = "rules/models/alibaba/qwen3.5-plus.toml" +content_sha256 = "4d224c51db0164b25333be6292764747060d53e1d69ff6feb2f12e32495876cd" + +[[rules]] +path = "rules/providers/alibaba-cn/provider.toml" +content_sha256 = "1defa8dbcc173c84db79c9fad31c17bc570033951bcbbe7f40dfcfa8080343ab" + +[[rules]] +path = "rules/providers/alibaba-cn/qwen3.5-plus.toml" +content_sha256 = "d9d1144d7d6ca02686aceefd4a8a1658838b0e24121d8778e461e20566b51315" + +[[rules]] +path = "rules/providers/alibaba-cn/models/deepseek-v4-flash.toml" +content_sha256 = "73bed1583302aeea226835bd28f9d86dc2026871c7f6ad08c393cea4fae5eee7" + +[[rules]] +path = "rules/providers/alibaba-cn/models/deepseek-v4-pro.toml" +content_sha256 = "22be7a8432b3808bbf634a545cbd63cb41049bf142b9e012c411cd4449bd551a" + +[[rules]] +path = "rules/providers/alibaba-cn/models/deepseek-v4.1-flash.toml" +content_sha256 = "217e98b9d94989bc5faf7f41605f0b96c5446fd4a8fbc27baf1aabc220955b1b" + +[[rules]] +path = "rules/providers/alibaba-cn/models/glm-5.1.toml" +content_sha256 = "02561502c0416231cc4dfaad96ab2832530c379ff955bbf956879e8516a452d2" + +[[rules]] +path = "rules/providers/alibaba-cn/models/glm-5.3.toml" +content_sha256 = "e55f5afc1d19a057fa2fa8dd0b68e99383c9f2c70165c78fcd0c341404431d16" + +[[rules]] +path = "rules/providers/alibaba-cn/models/glm-5.toml" +content_sha256 = "058ec2c46307217c3c64e9160af6f9ad34c73589691f795105b739db4d94bbc3" + +[[rules]] +path = "rules/providers/alibaba-cn/models/kimi-k3.toml" +content_sha256 = "531223214346f6b6266c8358b10ba25002e05691b5079553a250e65acdccca85" + +[[rules]] +path = "rules/providers/alibaba-cn/models/qwen3.5-397b-a17b.toml" +content_sha256 = "f2ad3aace634ee9fdd6c2f0e0a2cd632c4b7f1b1f03d664f76da197ee415eac9" + +[[rules]] +path = "rules/providers/alibaba-cn/models/qwen3.5-flash.toml" +content_sha256 = "d9dba14e40e69835eb94c604afaf5a4a60ef4e9a69ac0330817f266523904f8a" + +[[rules]] +path = "rules/providers/alibaba-cn/models/qwen3.7-max.toml" +content_sha256 = "f134180b58589496e24f7b1bff31f3a7135229ef689d08fcf365849655aab909" + +[[rules]] +path = "rules/providers/alibaba-cn/models/qwen3.7-plus.toml" +content_sha256 = "b0ae6784eaa4928043972713e006771a3700b169663f2817674227768010a1b5" + +[[rules]] +path = "rules/providers/alibaba-cn/models/qwen3.8-flash.toml" +content_sha256 = "fc0ab4a85f709e3d823c7ba166c5136ed3470dd2369faa032c3e578500429c60" + +[[rules]] +path = "rules/providers/alibaba-cn/models/qwen3.8-max.toml" +content_sha256 = "cdb054ab498b5a30bc4b4ab210700526f8bb07b0d54f87242c1b202dadcba22b" + +[[rules]] +path = "rules/providers/deepseek/models/deepseek-flash.toml" +content_sha256 = "127a9d916ae2ddf8d676a31cb8d92f6768df501983b2325c472ecd2c72c160c5" + +[[rules]] +path = "rules/providers/deepseek/models/deepseek-v4-flash.toml" +content_sha256 = "e1ab98d3fcdebf7413370919f78447b2d409567f8fef20f843e4c4771aecc4cb" + +[[rules]] +path = "rules/providers/deepseek/models/deepseek-v4-pro.toml" +content_sha256 = "2f0d21578a59816c3b47868ee8515ec9f085f0bcf2c42c8795ad391555dc9078" + +[[rules]] +path = "rules/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml" +content_sha256 = "b88c977337d85b93215dc35430df6f44f605a3b6ba0f63aae04ec22d2be0ac05" + +[[rules]] +path = "rules/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml" +content_sha256 = "f143db517f6c563a72d453e08f9ce2b6210a835f41b09d342a4ce9a4a29c86cf" + +[[rules]] +path = "rules/providers/siliconflow-cn/models/zai-org/GLM-5.2.toml" +content_sha256 = "86135162014e0b5628229ce675bd60c10518ef90d429b5c9f5e1db8536e3b2e7" + +[[rules]] +path = "rules/providers/volcengine-coding-plan/models/deepseek-v4-flash.toml" +content_sha256 = "b0dd71bbd783be50713ea7d063bf088fb55b3740eceba98ec4b5c86542fee367" + +[[rules]] +path = "rules/providers/volcengine-coding-plan/models/deepseek-v4-pro.toml" +content_sha256 = "768cb798c55b2d8708de8efddce49bcc1c974b835d76b54c50c99810c7fcdc90" + +[[rules]] +path = "rules/providers/volcengine-coding-plan/models/glm-5.3-flash.toml" +content_sha256 = "7ef730aafd506cc418f4243b3906de73ce8921137d7f938cb1a5a9c3985f8956" + +[[rules]] +path = "rules/providers/volcengine-coding-plan/models/glm-5.3.toml" +content_sha256 = "e11cd3a7618e3fdcd0de50e3f47fb1347fa5e453bb79767a3efd987d6cf71c3d" + +[[rules]] +path = "rules/providers/volcengine-coding-plan/models/kimi-k3.toml" +content_sha256 = "4ac865e8296d3cd562c0a56e8942390e3447e3a0304c447e6306a4cf88df6591" + +[[rules]] +path = "rules/providers/volcengine-coding-plan/models/minimax-m3.toml" +content_sha256 = "a5ed68800751785096385a16331bc4593dbd15b67a4fc3c500f3a22b7c8ec7d8" + +[[rules]] +path = "rules/providers/volcengine/models/deepseek-v4-flash-ga-260731.toml" +content_sha256 = "417b732ec22c5a27096374a0f8e8c0e7d78229634ee34d65bf21af0601da37ea" + +[[rules]] +path = "rules/providers/volcengine/models/deepseek-v4-pro-ga-260813.toml" +content_sha256 = "68a73cdee1c005b82881f8075cb742ad7bdd2a5f2e122c28ccbb5e803114e68c" + +[[rules]] +path = "rules/providers/volcengine/models/doubao-seed-2-1-pro-260628.toml" +content_sha256 = "f45f11dc48706e35268e5babd2cf4d1f3dd28eef5c399d7f7cbb424604bdd16a" + +[[rules]] +path = "rules/providers/volcengine/models/glm-5-2-260617.toml" +content_sha256 = "bc38551ccad4513986a0fe5ec701a17bcd4e19837316baac476edd5c38086dfe" + +[[rules]] +path = "rules/providers/volcengine/models/glm-5-3-flash-260828.toml" +content_sha256 = "4b9924594b13541b7db9df89904531e6ca5e32e1eaaf6667a0453a6fde2c11a0" diff --git a/sdk/nexent/core/model_service/profiles/rules/models/alibaba/qwen3.5-plus.toml b/sdk/nexent/core/model_service/profiles/rules/models/alibaba/qwen3.5-plus.toml new file mode 100644 index 0000000000..f9a8d8f88e --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/models/alibaba/qwen3.5-plus.toml @@ -0,0 +1,12 @@ +schema_version = 1 +rule_id = "qwen35_plus_instruction_v1" +version = 1 +target_kind = "model" +target_key = "alibaba/qwen3.5-plus" +reason = "Use Nexent's initial traditional instruction strategy" +evidence = "Nexent model access SPEC: traditional instruction roles, strategy system_only_v1" +aliases = ["qwen3.5-plus"] + +[instruction_mapping] +ref = "system_only_v1" +version = 1 diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4-flash.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4-flash.toml new file mode 100644 index 0000000000..bc5b2ef7f5 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4-flash.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "alibaba_cn_deepseek_v4_flash_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/deepseek-v4-flash" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/deepseek-v4-flash.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_deepseek_v4_flash_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/deepseek-v4-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["high", "max"]} +source = {"ref" = "alibaba_cn_deepseek_v4_flash_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/deepseek-v4-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4-pro.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4-pro.toml new file mode 100644 index 0000000000..83d4a7b0ea --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4-pro.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "alibaba_cn_deepseek_v4_pro_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/deepseek-v4-pro" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/deepseek-v4-pro.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_deepseek_v4_pro_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/deepseek-v4-pro.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["high", "max"]} +source = {"ref" = "alibaba_cn_deepseek_v4_pro_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/deepseek-v4-pro.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4.1-flash.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4.1-flash.toml new file mode 100644 index 0000000000..0a1837e657 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/deepseek-v4.1-flash.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "alibaba_cn_deepseek_v4_1_flash_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/deepseek-v4.1-flash" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/deepseek-v4.1-flash.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_deepseek_v4_1_flash_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/deepseek-v4.1-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "alibaba_cn_deepseek_v4_1_flash_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/deepseek-v4.1-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.1.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.1.toml new file mode 100644 index 0000000000..88f2ea7449 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.1.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "alibaba_cn_glm_5_1_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/glm-5.1" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/glm-5.1.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_glm_5_1_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/glm-5.1.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "integer", "maximum" = 131072} +source = {"ref" = "alibaba_cn_glm_5_1_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/glm-5.1.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking_budget", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.3.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.3.toml new file mode 100644 index 0000000000..d95f9bc38f --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.3.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "alibaba_cn_glm_5_3_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/glm-5.3" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/glm-5.3.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "alibaba_cn_glm_5_3_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/glm-5.3.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.toml new file mode 100644 index 0000000000..523b3626a9 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/glm-5.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "alibaba_cn_glm_5_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/glm-5" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/glm-5.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_glm_5_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/glm-5.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "integer", "maximum" = 32768} +source = {"ref" = "alibaba_cn_glm_5_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/glm-5.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking_budget", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/kimi-k3.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/kimi-k3.toml new file mode 100644 index 0000000000..ee53a52e0d --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/kimi-k3.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "alibaba_cn_kimi_k3_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/kimi-k3" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/kimi-k3.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "alibaba_cn_kimi_k3_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/kimi-k3.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.5-397b-a17b.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.5-397b-a17b.toml new file mode 100644 index 0000000000..adffb416cb --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.5-397b-a17b.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "alibaba_cn_qwen3_5_397b_a17b_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/qwen3.5-397b-a17b" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.5-397b-a17b.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_qwen3_5_397b_a17b_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.5-397b-a17b.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "integer", "maximum" = 81920} +source = {"ref" = "alibaba_cn_qwen3_5_397b_a17b_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.5-397b-a17b.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking_budget", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.5-flash.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.5-flash.toml new file mode 100644 index 0000000000..205e1fbc96 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.5-flash.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "alibaba_cn_qwen3_5_flash_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/qwen3.5-flash" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.5-flash.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_qwen3_5_flash_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.5-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "integer", "maximum" = 81920} +source = {"ref" = "alibaba_cn_qwen3_5_flash_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.5-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking_budget", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.7-max.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.7-max.toml new file mode 100644 index 0000000000..f86e6fb21a --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.7-max.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "alibaba_cn_qwen3_7_max_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/qwen3.7-max" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.7-max.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_qwen3_7_max_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.7-max.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "integer", "maximum" = 262144} +source = {"ref" = "alibaba_cn_qwen3_7_max_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.7-max.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking_budget", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.7-plus.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.7-plus.toml new file mode 100644 index 0000000000..21d2b7204f --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.7-plus.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "alibaba_cn_qwen3_7_plus_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/qwen3.7-plus" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.7-plus.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_qwen3_7_plus_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.7-plus.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "integer", "maximum" = 262144} +source = {"ref" = "alibaba_cn_qwen3_7_plus_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.7-plus.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking_budget", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.8-flash.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.8-flash.toml new file mode 100644 index 0000000000..9edfb817c0 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.8-flash.toml @@ -0,0 +1,45 @@ +schema_version = 1 +rule_id = "alibaba_cn_qwen3_8_flash_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/qwen3.8-flash" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.8-flash.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_qwen3_8_flash_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.8-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "medium", "xhigh"]} +source = {"ref" = "alibaba_cn_qwen3_8_flash_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.8-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "integer", "maximum" = 262144} +source = {"ref" = "alibaba_cn_qwen3_8_flash_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.8-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking_budget", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.8-max.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.8-max.toml new file mode 100644 index 0000000000..6e63ee110a --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/models/qwen3.8-max.toml @@ -0,0 +1,45 @@ +schema_version = 1 +rule_id = "alibaba_cn_qwen3_8_max_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "alibaba-cn/qwen3.8-max" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.8-max.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "alibaba_cn_qwen3_8_max_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.8-max.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "medium", "xhigh"]} +source = {"ref" = "alibaba_cn_qwen3_8_max_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.8-max.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "integer", "minimum" = 0, "maximum" = 262144} +source = {"ref" = "alibaba_cn_qwen3_8_max_controls_v1", "version" = 1, "evidence" = "https://help.aliyun.com/zh/model-studio/deep-thinking; pinned models.dev providers/alibaba-cn/models/qwen3.8-max.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking_budget", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/provider.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/provider.toml new file mode 100644 index 0000000000..997ee81276 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/provider.toml @@ -0,0 +1,8 @@ +schema_version = 1 +rule_id = "alibaba_cn_endpoint_hints_v1" +version = 1 +target_kind = "provider" +target_key = "alibaba-cn" +reason = "Retain the copied China endpoint as an explicit selection hint" +evidence = "models.dev pinned alibaba-cn/provider.toml api field; upstream e69e1844210e5cfe7c7e7f68d57328bc7451a51c" +endpoint_hints = ["https://dashscope.aliyuncs.com/compatible-mode/v1"] diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/qwen3.5-plus.toml b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/qwen3.5-plus.toml new file mode 100644 index 0000000000..e6de8c1cde --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/alibaba-cn/qwen3.5-plus.toml @@ -0,0 +1,117 @@ +schema_version = 1 +rule_id = "alibaba_cn_qwen35_plus_thinking_v1" +version = 3 +target_kind = "provider_model" +target_key = "alibaba-cn/qwen3.5-plus" +model_profile_key = "alibaba/qwen3.5-plus" +reason = "Declare the exact OpenAI-compatible thinking control without guessing from a capability flag" +evidence = "https://www.alibabacloud.com/help/zh/model-studio/deep-thinking (checked 2026-10-13)" + +[[capability_overrides]] +path = "/reasoning/controls" +value = [{mode = "toggle", parameter_key = "enable_thinking", values = [false, true]}, {mode = "budget_tokens", parameter_key = "reasoning_budget_tokens", max = 81920}] +reason = "Preserve the copied toggle and budget controls; register their distinct API bindings" +evidence = "https://www.alibabacloud.com/help/zh/model-studio/deep-thinking#supported-models (checked 2026-10-13)" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +default_value = true + +[parameters.schema] +type = "boolean" + +[parameters.source] +ref = "alibaba_cn_qwen35_plus_thinking_v1" +version = 1 +evidence = "https://www.alibabacloud.com/help/zh/model-studio/deep-thinking" + +[parameters.api_binding] +path = "/enable_thinking" +transform_ref = "identity_v1" +transform_version = 1 + + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" + +[parameters.schema] +type = "integer" +maximum = 81920 + +[parameters.source] +ref = "alibaba_cn_qwen35_plus_thinking_v1" +version = 1 +evidence = "Pinned models.dev providers/alibaba-cn/models/qwen3.5-plus.toml: reasoning_options budget_tokens max=81920; lower bound remains unknown" + +[parameters.condition] +key = "enable_thinking" +op = "eq" +value = true + +[parameters.api_binding] +path = "/thinking_budget" +transform_ref = "identity_v1" +transform_version = 1 + +[[parameters]] +key = "stop" +visible = false +editable = true +required = false +allowed_scopes = ["outlet"] +ui_group = "runtime" +owner = "combination" + +[parameters.schema] +type = ["string", "array"] +minLength = 1 + +[parameters.schema.items] +type = "string" +minLength = 1 + +[parameters.source] +ref = "alibaba_cn_qwen35_plus_thinking_v1" +version = 2 +evidence = "https://www.alibabacloud.com/help/zh/model-studio/qwen-api-via-openai-chat-completions (checked 2026-10-14); string stop subset consumed by CodeAgent and planning" + +[parameters.api_binding] +path = "/stop" +transform_ref = "identity_v1" +transform_version = 1 + +[[parameters]] +key = "max_output_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "generation" +owner = "combination" + +[parameters.schema] +type = "integer" +minimum = 1 +maximum = 65536 + +[parameters.source] +ref = "alibaba_cn_qwen35_plus_thinking_v1" +version = 3 +evidence = "Pinned alibaba-cn/qwen3.5-plus limit.output=65536; https://www.alibabacloud.com/help/zh/model-studio/qwen-api-via-openai-chat-completions (checked 2026-10-14): max_tokens limits answer tokens for this model, excluding reasoning" + +[parameters.api_binding] +path = "/max_tokens" +transform_ref = "identity_v1" +transform_version = 1 diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-flash.toml b/sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-flash.toml new file mode 100644 index 0000000000..3b352a9ca7 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-flash.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "deepseek_deepseek_flash_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "deepseek/deepseek-flash" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://api-docs.deepseek.com/api/create-chat-completion/; pinned models.dev providers/deepseek/models/deepseek-flash.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "deepseek_deepseek_flash_controls_v1", "version" = 1, "evidence" = "https://api-docs.deepseek.com/api/create-chat-completion/; pinned models.dev providers/deepseek/models/deepseek-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "deepseek_deepseek_flash_controls_v1", "version" = 1, "evidence" = "https://api-docs.deepseek.com/api/create-chat-completion/; pinned models.dev providers/deepseek/models/deepseek-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-v4-flash.toml b/sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-v4-flash.toml new file mode 100644 index 0000000000..1631b5b897 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-v4-flash.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "deepseek_deepseek_v4_flash_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "deepseek/deepseek-v4-flash" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://api-docs.deepseek.com/api/create-chat-completion/; pinned models.dev providers/deepseek/models/deepseek-v4-flash.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "deepseek_deepseek_v4_flash_controls_v1", "version" = 1, "evidence" = "https://api-docs.deepseek.com/api/create-chat-completion/; pinned models.dev providers/deepseek/models/deepseek-v4-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "deepseek_deepseek_v4_flash_controls_v1", "version" = 1, "evidence" = "https://api-docs.deepseek.com/api/create-chat-completion/; pinned models.dev providers/deepseek/models/deepseek-v4-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-v4-pro.toml b/sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-v4-pro.toml new file mode 100644 index 0000000000..5b5bf6f944 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/deepseek/models/deepseek-v4-pro.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "deepseek_deepseek_v4_pro_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "deepseek/deepseek-v4-pro" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://api-docs.deepseek.com/api/create-chat-completion/; pinned models.dev providers/deepseek/models/deepseek-v4-pro.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "deepseek_deepseek_v4_pro_controls_v1", "version" = 1, "evidence" = "https://api-docs.deepseek.com/api/create-chat-completion/; pinned models.dev providers/deepseek/models/deepseek-v4-pro.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "deepseek_deepseek_v4_pro_controls_v1", "version" = 1, "evidence" = "https://api-docs.deepseek.com/api/create-chat-completion/; pinned models.dev providers/deepseek/models/deepseek-v4-pro.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml b/sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml new file mode 100644 index 0000000000..c905882789 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "siliconflow_cn_pro_zai_org_glm_5_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "siliconflow-cn/Pro/zai-org/GLM-5" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://docs.siliconflow.cn/docs/api/chat-completions-post; pinned models.dev providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "siliconflow_cn_pro_zai_org_glm_5_controls_v1", "version" = 1, "evidence" = "https://docs.siliconflow.cn/docs/api/chat-completions-post; pinned models.dev providers/siliconflow-cn/models/Pro/zai-org/GLM-5.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/enable_thinking", "transform_ref" = "identity_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml b/sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml new file mode 100644 index 0000000000..9bcdcbf857 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "siliconflow_cn_deepseek_ai_deepseek_v4_flash_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "siliconflow-cn/deepseek-ai/DeepSeek-V4-Flash" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://docs.siliconflow.cn/docs/api/chat-completions-post; pinned models.dev providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "reasoning_budget_tokens" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "integer", "minimum" = 128, "maximum" = 32768} +source = {"ref" = "siliconflow_cn_deepseek_ai_deepseek_v4_flash_controls_v1", "version" = 1, "evidence" = "https://docs.siliconflow.cn/docs/api/chat-completions-post; pinned models.dev providers/siliconflow-cn/models/deepseek-ai/DeepSeek-V4-Flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking_budget", "transform_ref" = "identity_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/zai-org/GLM-5.2.toml b/sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/zai-org/GLM-5.2.toml new file mode 100644 index 0000000000..349736f918 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/siliconflow-cn/models/zai-org/GLM-5.2.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "siliconflow_cn_zai_org_glm_5_2_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "siliconflow-cn/zai-org/GLM-5.2" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://docs.siliconflow.cn/docs/api/chat-completions-post; pinned models.dev providers/siliconflow-cn/models/zai-org/GLM-5.2.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["high", "max"]} +source = {"ref" = "siliconflow_cn_zai_org_glm_5_2_controls_v1", "version" = 1, "evidence" = "https://docs.siliconflow.cn/docs/api/chat-completions-post; pinned models.dev providers/siliconflow-cn/models/zai-org/GLM-5.2.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/deepseek-v4-flash.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/deepseek-v4-flash.toml new file mode 100644 index 0000000000..2b37b3c990 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/deepseek-v4-flash.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "volcengine_coding_plan_deepseek_v4_flash_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine-coding-plan/deepseek-v4-flash" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/deepseek-v4-flash.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "volcengine_coding_plan_deepseek_v4_flash_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/deepseek-v4-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/deepseek-v4-pro.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/deepseek-v4-pro.toml new file mode 100644 index 0000000000..e1c4732ca0 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/deepseek-v4-pro.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "volcengine_coding_plan_deepseek_v4_pro_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine-coding-plan/deepseek-v4-pro" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/deepseek-v4-pro.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "volcengine_coding_plan_deepseek_v4_pro_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/deepseek-v4-pro.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/glm-5.3-flash.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/glm-5.3-flash.toml new file mode 100644 index 0000000000..1d69265ecb --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/glm-5.3-flash.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "volcengine_coding_plan_glm_5_3_flash_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine-coding-plan/glm-5.3-flash" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/glm-5.3-flash.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "volcengine_coding_plan_glm_5_3_flash_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/glm-5.3-flash.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/glm-5.3.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/glm-5.3.toml new file mode 100644 index 0000000000..0202cd2772 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/glm-5.3.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "volcengine_coding_plan_glm_5_3_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine-coding-plan/glm-5.3" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/glm-5.3.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "volcengine_coding_plan_glm_5_3_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/glm-5.3.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/kimi-k3.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/kimi-k3.toml new file mode 100644 index 0000000000..a8139c1788 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/kimi-k3.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "volcengine_coding_plan_kimi_k3_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine-coding-plan/kimi-k3" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/kimi-k3.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "volcengine_coding_plan_kimi_k3_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/kimi-k3.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "volcengine_coding_plan_kimi_k3_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/kimi-k3.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/minimax-m3.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/minimax-m3.toml new file mode 100644 index 0000000000..a6739da3b8 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine-coding-plan/models/minimax-m3.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "volcengine_coding_plan_minimax_m3_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine-coding-plan/minimax-m3" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/minimax-m3.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "volcengine_coding_plan_minimax_m3_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1928261; pinned models.dev providers/volcengine-coding-plan/models/minimax-m3.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/deepseek-v4-flash-ga-260731.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/deepseek-v4-flash-ga-260731.toml new file mode 100644 index 0000000000..bc4fcb052f --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/deepseek-v4-flash-ga-260731.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "volcengine_deepseek_v4_flash_ga_260731_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine/deepseek-v4-flash-ga-260731" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/deepseek-v4-flash-ga-260731.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "volcengine_deepseek_v4_flash_ga_260731_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/deepseek-v4-flash-ga-260731.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["minimal", "low", "medium", "high"]} +source = {"ref" = "volcengine_deepseek_v4_flash_ga_260731_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/deepseek-v4-flash-ga-260731.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/deepseek-v4-pro-ga-260813.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/deepseek-v4-pro-ga-260813.toml new file mode 100644 index 0000000000..ba60c60e32 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/deepseek-v4-pro-ga-260813.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "volcengine_deepseek_v4_pro_ga_260813_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine/deepseek-v4-pro-ga-260813" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/deepseek-v4-pro-ga-260813.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "volcengine_deepseek_v4_pro_ga_260813_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/deepseek-v4-pro-ga-260813.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["minimal", "low", "medium", "high"]} +source = {"ref" = "volcengine_deepseek_v4_pro_ga_260813_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/deepseek-v4-pro-ga-260813.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/doubao-seed-2-1-pro-260628.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/doubao-seed-2-1-pro-260628.toml new file mode 100644 index 0000000000..d96b5fd549 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/doubao-seed-2-1-pro-260628.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "volcengine_doubao_seed_2_1_pro_260628_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine/doubao-seed-2-1-pro-260628" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/doubao-seed-2-1-pro-260628.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["none", "minimal", "low", "medium", "high", "xhigh", "max"]} +source = {"ref" = "volcengine_doubao_seed_2_1_pro_260628_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/doubao-seed-2-1-pro-260628.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/glm-5-2-260617.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/glm-5-2-260617.toml new file mode 100644 index 0000000000..36abb3cc04 --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/glm-5-2-260617.toml @@ -0,0 +1,32 @@ +schema_version = 1 +rule_id = "volcengine_glm_5_2_260617_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine/glm-5-2-260617" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/glm-5-2-260617.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "enable_thinking" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "boolean"} +source = {"ref" = "volcengine_glm_5_2_260617_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/glm-5-2-260617.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/thinking/type", "transform_ref" = "thinking_toggle_v1", "transform_version" = 1} + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["minimal", "low", "medium", "high"]} +source = {"ref" = "volcengine_glm_5_2_260617_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/glm-5-2-260617.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1} +condition = {key = "enable_thinking", op = "eq", value = true} diff --git a/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/glm-5-3-flash-260828.toml b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/glm-5-3-flash-260828.toml new file mode 100644 index 0000000000..cc520ba19d --- /dev/null +++ b/sdk/nexent/core/model_service/profiles/rules/providers/volcengine/models/glm-5-3-flash-260828.toml @@ -0,0 +1,19 @@ +schema_version = 1 +rule_id = "volcengine_glm_5_3_flash_260828_controls_v1" +version = 1 +target_kind = "provider_model" +target_key = "volcengine/glm-5-3-flash-260828" +reason = "Bind only the exact upstream-declared reasoning controls for OpenAI Chat" +evidence = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/glm-5-3-flash-260828.toml and provider.toml; reviewed 2026-10-16" + +[[parameters]] +key = "reasoning_effort" +visible = true +editable = true +required = false +allowed_scopes = ["outlet", "conversation"] +ui_group = "reasoning" +owner = "combination" +schema = {"type" = "string", "enum" = ["low", "high", "max"]} +source = {"ref" = "volcengine_glm_5_3_flash_260828_controls_v1", "version" = 1, "evidence" = "https://www.volcengine.com/docs/82379/1330310; pinned models.dev providers/volcengine/models/glm-5-3-flash-260828.toml and provider.toml; reviewed 2026-10-16"} +api_binding = {"path" = "/reasoning_effort", "transform_ref" = "identity_v1", "transform_version" = 1}