Update default model to gpt-5.6-terra, require Ruby 3.3

.github/workflows/ci.yml

@@ -19,7 +19,7 @@ jobs:
       - name: Ruby Setup
         uses: ruby/setup-ruby@v1
         with:
-          ruby-version: "3.2.1"
+          ruby-version: "3.3"
           bundler-cache: true
 
       - name: Build

CHANGELOG.md

@@ -5,6 +5,20 @@ 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).
 
+## [Unreleased]
+
+### Changed
+
+- **Default model**: Changed from `gpt-5.2` to `gpt-5.6-terra`, the cost-balanced tier of OpenAI's GPT-5.6 family. Pricing is roughly the same as `gpt-5.2` ($2/$12 vs $1.75/$14 per 1M input/output tokens) with a newer model and a 1M-token context window. `generate_schema!` uses the same model.
+
+- **Default `reasoning_effort` is now `"none"` instead of `nil`**: GPT-5.5 and newer models fall back to `"medium"` reasoning when the `reasoning` parameter is omitted, so keeping the old `nil` default would have silently made every default call slower and more expensive. The gem now sends `reasoning: { effort: "none" }` by default, which matches the previous behavior on `gpt-5.2`. Setting `reasoning_effort = nil` still omits the parameter entirely, which you will need for models that do not accept it (e.g., `gpt-4o`).
+
+- **No reasoning summary requested when `reasoning_effort` is `"none"`**: `summary: "auto"` is only sent alongside a non-`"none"` effort.
+
+- **Requires Ruby 3.3 or newer** (was 3.2). Ruby 3.2 reached end-of-life on March 31, 2026, and newer versions of the `openai` gem require 3.3+. CI now runs on Ruby 3.3. (Breaking for Ruby 3.2 users.)
+
+- **Bumped `openai` runtime dependency from `~> 0.59` to `~> 0.80`**: Picks up the GPT-5.6 model slugs and a summer's worth of upstream SDK fixes. No gem code changes were required for the bump.
+
 ## [0.6.1] - 2026-04-22
 
 ### Added

Gemfile.lock

@@ -6,7 +6,7 @@ PATH
       base64 (~> 0.1, > 0.1.1)
       json (~> 2.0)
       marcel (~> 1.0)
-      openai (~> 0.59)
+      openai (~> 0.80)
       tty-spinner (~> 0.9.3)
 
 GEM
@@ -66,10 +66,11 @@ GEM
     lint_roller (1.1.0)
     logger (1.7.0)
     marcel (1.1.0)
-    openai (0.59.0)
+    openai (0.80.0)
       base64
       cgi
-      connection_pool
+      connection_pool (>= 2.2.3)
+      logger
     ostruct (0.6.3)
     parallel (1.27.0)
     parser (3.3.10.1)

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-5.2", ... }
+#       :response => { id: "resp_abc...", model: "gpt-5.6-terra", ... }
 #    }
 
 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-5.2", ... }
+#           :response => { id: "resp_abc...", model: "gpt-5.6-terra", ... }
 #        }
 #    ]
 ```
@@ -131,11 +131,11 @@ chat.add("Here's what I think...", role: "assistant")
 
 ### Model
 
-The gem defaults to `gpt-5.2`. You can change it:
+The gem defaults to `gpt-5.6-terra`. You can change it:
 
 ```ruby
 chat = AI::Chat.new
-chat.model = "gpt-4o"
+chat.model = "gpt-5.6-luna"
 ```
 
 ### API Key
@@ -403,13 +403,15 @@ Control how much reasoning the model does before responding:
 
 ```ruby
 chat = AI::Chat.new
-chat.reasoning_effort = "high"  # "low", "medium", or "high"
+chat.reasoning_effort = "high"  # "none", "low", "medium", "high", "xhigh", or "max"
 
 chat.user("Explain the tradeoffs between microservices and monoliths.")
 chat.generate!
 ```
 
-By default, `reasoning_effort` is `nil` (no reasoning parameter is sent). For `gpt-5.2`, this is equivalent to no reasoning.
+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.
+
+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.
 
 ### Verbosity
 
@@ -474,7 +476,7 @@ chat.generate!
 
 response = chat.last[:response]
 response[:id]     # => "resp_abc123..."
-response[:model]  # => "gpt-5.2"
+response[:model]  # => "gpt-5.6-terra"
 response[:usage]  # => { input_tokens: 5, output_tokens: 7, total_tokens: 12 }
 ```
 

ai-chat.gemspec

@@ -18,8 +18,8 @@ Gem::Specification.new do |spec|
     "source_code_uri" => "https://github.com/firstdraft/ai-chat"
   }
 
-  spec.required_ruby_version = ">= 3.2"
-  spec.add_runtime_dependency "openai", "~> 0.59"
+  spec.required_ruby_version = ">= 3.3"
+  spec.add_runtime_dependency "openai", "~> 0.80"
   spec.add_runtime_dependency "marcel", "~> 1.0"
   spec.add_runtime_dependency "base64", "~> 0.1", "> 0.1.1"
   spec.add_runtime_dependency "json", "~> 2.0"

examples/15_proxy.rb

@@ -10,7 +10,7 @@
 
 unless ENV["AICHAT_PROXY_KEY"]
   puts "Skipping proxy tests - set AICHAT_PROXY_KEY environment variable to run these examples"
-  exit 0
+  return # not exit: all.rb require_relatives this file, and exit would skip the remaining examples
 end
 
 puts "\n=== AI::Chat Proxy Examples ==="

examples/17_verbosity.rb

@@ -79,6 +79,7 @@
 puts "-" * 30
 chat6 = AI::Chat.new
 chat6.model = "gpt-4.1-nano"
+chat6.reasoning_effort = nil # gpt-4.1 models reject the reasoning parameter
 chat6.verbosity = :medium
 chat6.user("How high do planes typically fly?")
 response = chat6.generate![:content]
@@ -91,6 +92,7 @@
 puts "-" * 30
 chat7 = AI::Chat.new
 chat7.model = "gpt-4.1-nano"
+chat7.reasoning_effort = nil
 chat7.verbosity = :low
 chat7.user("How high do planes typically fly?")
 begin

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 = nil
-      @model = "gpt-5.2"
+      @reasoning_effort = "none"
+      @model = "gpt-5.6-terra"
       client_options = {api_key: @api_key}
       client_options[:base_url] = BASE_PROXY_URL if @proxy
       @client = OpenAI::Client.new(**client_options)
@@ -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-5.2",
+        model: "gpt-5.6-terra",
         input: [
           {role: :system, content: system_prompt},
           {role: :user, content: description}
@@ -307,6 +307,13 @@ def extract_filename(obj)
       end
     end
 
+    # Reasoning summaries are only meaningful when the model actually reasons.
+    def reasoning_parameters
+      params = {effort: reasoning_effort}
+      params[:summary] = "auto" unless reasoning_effort.to_s == "none"
+      params
+    end
+
     def create_conversation
       conversation = client.conversations.create
       self.conversation_id = conversation.id
@@ -321,7 +328,7 @@ def create_response
       parameters[:background] = background if background
       parameters[:tools] = tools unless tools.empty?
       parameters[:text] = schema if schema
-      parameters[:reasoning] = {effort: reasoning_effort, summary: "auto"} if reasoning_effort
+      parameters[:reasoning] = reasoning_parameters if reasoning_effort
 
       create_conversation unless conversation_id
       parameters[:conversation] = conversation_id

spec/unit/chat_spec.rb

@@ -758,3 +758,48 @@ def schema_client_double
     end
   end
 end
+
+RSpec.describe AI::Chat, "request parameters" do
+  let(:chat) { AI::Chat.new(api_key: "test-key") }
+  let(:responses) { double("responses") }
+
+  before do
+    chat.conversation_id = "conv_123"
+    allow(chat).to receive(:client).and_return(double("client", responses: responses))
+    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")
+  end
+
+  it "defaults reasoning_effort to \"none\"" do
+    expect(chat.reasoning_effort).to eq("none")
+  end
+
+  it "sends reasoning effort none without 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"})
+    )
+  end
+
+  it "requests a reasoning summary when reasoning_effort is set" do
+    chat.reasoning_effort = "high"
+    chat.send(:create_response)
+
+    expect(responses).to have_received(:create).with(
+      hash_including(reasoning: {effort: "high", summary: "auto"})
+    )
+  end
+
+  it "omits the reasoning parameter when reasoning_effort is nil" do
+    chat.reasoning_effort = nil
+    chat.send(:create_response)
+
+    expect(responses).to have_received(:create) do |params|
+      expect(params).not_to have_key(:reasoning)
+    end
+  end
+end