Default to gpt-6-luna with low reasoning effort; release 0.8.0

CHANGELOG.md

@@ -5,6 +5,14 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.8.0] - 2026-10-19
+
+### Changed
+
+- **Default model**: Changed from `gpt-5.6-terra` to `gpt-6-luna`, the most efficient tier of OpenAI's GPT-6 family. It costs far less ($0.10/$0.50 vs $2/$12 per 1M input/output tokens) and answered a sample prompt about twice as fast in our testing. `generate_schema!` uses the same model.
+
+- **Default `reasoning_effort` is now `"low"` instead of `"none"`**: `gpt-6.1-sol` and `gpt-6-astra` reject `"none"` with a 400 error, so switching `model` to one of them used to fail unless you also changed `reasoning_effort`. `"low"` is accepted by all of OpenAI's current flagship models. Since the effort is no longer `"none"`, a reasoning summary (`summary: "auto"`) is now requested by default too. Set `reasoning_effort = "none"` to get the previous behavior on models that support it.
+
 ## [0.7.0] - 2026-08-24
 
 ### Changed

Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    ai-chat (0.7.0)
+    ai-chat (0.8.0)
       amazing_print (~> 2.0)
       base64 (~> 0.1, > 0.1.1)
       json (~> 2.0)

README.md

@@ -56,7 +56,7 @@ ap response
 # => {
 #           :role => "assistant",
 #        :content => "Matz is nice and so we are nice.",
-#       :response => { id: "resp_abc...", model: "gpt-5.6-terra", ... }
+#       :response => { id: "resp_abc...", model: "gpt-6-luna", ... }
 #    }
 
 ap chat.messages
@@ -68,7 +68,7 @@ ap chat.messages
 #        {
 #               :role => "assistant",
 #            :content => "Matz is nice and so we are nice.",
-#           :response => { id: "resp_abc...", model: "gpt-5.6-terra", ... }
+#           :response => { id: "resp_abc...", model: "gpt-6-luna", ... }
 #        }
 #    ]
 ```
@@ -131,11 +131,11 @@ chat.add("Here's what I think...", role: "assistant")
 
 ### Model
 
-The gem defaults to `gpt-5.6-terra`. You can change it:
+The gem defaults to `gpt-6-luna`. You can change it:
 
 ```ruby
 chat = AI::Chat.new
-chat.model = "gpt-5.6-luna"
+chat.model = "gpt-6.1-sol"
 ```
 
 ### API Key
@@ -409,9 +409,9 @@ chat.user("Explain the tradeoffs between microservices and monoliths.")
 chat.generate!
 ```
 
-By default, `reasoning_effort` is `"none"`, so the model answers directly without spending reasoning tokens. This matters because GPT-5.5 and newer models (including the default `gpt-5.6-terra`) fall back to `"medium"` reasoning when the parameter is omitted, which is slower and more expensive.
+By default, `reasoning_effort` is `"low"`, so the model does a little reasoning before it answers. `"low"` is accepted by all of OpenAI's current flagship models (`gpt-6-luna`, `gpt-6.1-sol`, and `gpt-6-astra`), while some of them reject `"none"`, so you can switch models without also changing the effort. The gem always sends an effort because GPT-5.5 and newer models fall back to `"medium"` when the parameter is omitted, which is slower and more expensive.
 
-Set `reasoning_effort = nil` to omit the reasoning parameter entirely — you will need this if you switch to a model that does not support reasoning (e.g., `gpt-4o`) or one that predates the `"none"` option.
+Set `reasoning_effort = "none"` for the fastest replies on models that support it (such as `gpt-6-luna`). Set `reasoning_effort = nil` to omit the reasoning parameter entirely — you will need this if you switch to a model that does not support reasoning (e.g., `gpt-4o`).
 
 ### Verbosity
 
@@ -476,7 +476,7 @@ chat.generate!
 
 response = chat.last[:response]
 response[:id]     # => "resp_abc123..."
-response[:model]  # => "gpt-5.6-terra"
+response[:model]  # => "gpt-6-luna"
 response[:usage]  # => { input_tokens: 5, output_tokens: 7, total_tokens: 12 }
 ```
 

ai-chat.gemspec

@@ -2,7 +2,7 @@
 
 Gem::Specification.new do |spec|
   spec.name = "ai-chat"
-  spec.version = "0.7.0"
+  spec.version = "0.8.0"
   spec.authors = ["Raghu Betina", "Jelani Woods"]
   spec.email = ["raghu@firstdraft.com", "jelani@firstdraft.com"]
   spec.homepage = "https://github.com/firstdraft/ai-chat"

examples/17_verbosity.rb

@@ -45,8 +45,8 @@
 puts "-" * 30
 chat4 = AI::Chat.new
 chat4.user("How high do planes typically fly?")
-  begin
-    chat4.verbosity = :extreme
+begin
+  chat4.verbosity = :extreme
   puts "✗ Failed to raise ArgumentError for invalid verbosity: #{chat4.verbosity}"
 rescue ArgumentError => e
   puts "✓ Raises Argument error correctly: #{e.message}"
@@ -66,9 +66,9 @@
 chat5.system("Extract the profile information from the user message. Describe it like a 90's robot.")
 chat5.user("My name is Bryan. I like to skateboard and be cool. I'm seventeen and a quarter. You can reach me at bryan@example.com.")
 puts chat5.last[:content]
-puts 
+puts
 begin
-  response = chat5.generate![:content]
+  chat5.generate!
   puts "✓ Generates structured output response with verbosity"
 rescue => e
   puts "✗ Failed to generate structured output response with verbosity: #{e.message}"
@@ -96,7 +96,7 @@
 chat7.verbosity = :low
 chat7.user("How high do planes typically fly?")
 begin
-  response = chat7.generate![:content]
+  chat7.generate!
   puts "✗ Failed to raise error when setting :low verbosity on unsupported model."
 rescue OpenAI::Errors::BadRequestError => e
   puts "✓ Successfully raises error on for unsupported verbosity for older model: #{e.message}"

lib/ai/chat.rb

@@ -32,8 +32,8 @@ def initialize(api_key: nil, api_key_env_var: nil, proxy: nil)
       @proxy = proxy.nil? ? ENV[PROXY_ENV]&.downcase == "true" : !!proxy
       @api_key = resolve_api_key
       @messages = []
-      @reasoning_effort = "none"
-      @model = "gpt-5.6-terra"
+      @reasoning_effort = "low"
+      @model = "gpt-6-luna"
       client_options = {api_key: @api_key}
       client_options[:base_url] = BASE_PROXY_URL if @proxy
       @client = OpenAI::Client.new(**client_options)
@@ -46,7 +46,7 @@ def initialize(api_key: nil, api_key_env_var: nil, proxy: nil)
 
     def self.generate_schema!(description, location: "schema.json", api_key: nil, api_key_env_var: nil, proxy: nil)
       proxy = proxy.nil? ? ENV[PROXY_ENV]&.downcase == "true" : !!proxy
-      api_key = api_key || ENV.fetch(api_key_env_var || (proxy ? PROXY_KEY_ENV : OPENAI_KEY_ENV))
+      api_key ||= ENV.fetch(api_key_env_var || (proxy ? PROXY_KEY_ENV : OPENAI_KEY_ENV))
       prompt_path = File.expand_path("../prompts/schema_generator.md", __dir__)
       system_prompt = File.read(prompt_path)
 
@@ -57,7 +57,7 @@ def self.generate_schema!(description, location: "schema.json", api_key: nil, ap
 
       client = OpenAI::Client.new(**options)
       response = client.responses.create(
-        model: "gpt-5.6-terra",
+        model: "gpt-6-luna",
         input: [
           {role: :system, content: system_prompt},
           {role: :user, content: description}
@@ -175,7 +175,7 @@ def proxy=(value)
       @api_key = new_key
       @api_key_validated = false
       @client = new_client
-    rescue => error
+    rescue
       @proxy = previous_proxy
       raise
     end

spec/unit/chat_spec.rb

@@ -769,19 +769,28 @@ def schema_client_double
     allow(responses).to receive(:create).and_return(double("response"))
   end
 
-  it "defaults to gpt-5.6-terra" do
-    expect(chat.model).to eq("gpt-5.6-terra")
+  it "defaults to gpt-6-luna" do
+    expect(chat.model).to eq("gpt-6-luna")
   end
 
-  it "defaults reasoning_effort to \"none\"" do
-    expect(chat.reasoning_effort).to eq("none")
+  it "defaults reasoning_effort to \"low\"" do
+    expect(chat.reasoning_effort).to eq("low")
   end
 
-  it "sends reasoning effort none without a summary by default" do
+  it "sends reasoning effort low with a summary by default" do
     chat.send(:create_response)
 
     expect(responses).to have_received(:create).with(
-      hash_including(model: "gpt-5.6-terra", reasoning: {effort: "none"})
+      hash_including(model: "gpt-6-luna", reasoning: {effort: "low", summary: "auto"})
+    )
+  end
+
+  it "sends reasoning effort none without a summary" do
+    chat.reasoning_effort = "none"
+    chat.send(:create_response)
+
+    expect(responses).to have_received(:create).with(
+      hash_including(reasoning: {effort: "none"})
     )
   end