feat: add GPT-5.5 and Claude Opus 4.7 models

CHANGELOG.md

@@ -19,10 +19,10 @@ All notable changes to the Tachikoma project will be documented in this file.
 - Azure provider unit tests using URLProtocol stubs to verify path, query, and auth header construction.
 
 ### Changed
-- Added OpenAI's GPT-5.1 family (flagship/mini/nano) throughout the model enums, selectors, provider factories, capability registry, pricing tables, docs, and test suites. GPT aliases (`gpt`, `gpt-5`, `gpt-4o`) now normalize to `.openai(.gpt51)` so downstream apps inherit the new default seamlessly.
+- Added OpenAI's GPT-5.1 family (flagship/mini/nano) throughout the model enums, selectors, provider factories, capability registry, pricing tables, docs, and test suites. GPT aliases (`gpt`, `gpt-5`) now normalize to supported GPT-5 models so downstream apps inherit the new default seamlessly.
 - Expanded xAI Grok support to the full November 2025 catalog (`grok-4-fast-*`, `grok-code-fast-1`, `grok-2-*`, `grok-vision-beta`, etc.), updated the CLI shortcuts so `grok` now maps to `grok-4-fast-reasoning`, and refreshed selectors, provider parsers, capability tables, and docs snippets to match the official API lineup.
 - Google/Gemini support now targets the Gemini 2.5 family exclusively (`gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite`), with updated model selectors, parsers, docs, and pricing tables; older 1.5/2.0 IDs are no longer recognized.
-- Removed deprecated OpenAI reasoning models (`o1`, `o1-mini`, `o3`, `o3-mini`) in favour of the GPT‑5 family plus `o4-mini`, updating enums, provider factories, capability tables, prompts, and documentation metadata accordingly.
+- Removed deprecated OpenAI reasoning models (`o1`, `o1-mini`, `o3`, `o3-mini`, `o4-mini`) in favour of the GPT‑5 family, updating enums, provider factories, capability tables, prompts, and documentation metadata accordingly.
 - Google/Gemini integration now uses the documented `x-goog-api-key` header with `alt=sse` streaming, adds fallbacks for `GOOGLE_API_KEY` / `GOOGLE_APPLICATION_CREDENTIALS`, and hardens the SSE decoder so live tests succeed consistently.
 - Pruned Anthropic model support to the Claude 4.x line (Opus 4, Sonnet 4 / 4.5, Haiku 4.5) to match current API availability and reduce maintenance burden.
 - `TachikomaConfiguration` now loads credentials first and lets environment variables override them so operators can supersede stored settings without editing credentials files.
@@ -71,8 +71,8 @@ All notable changes to the Tachikoma project will be documented in this file.
 
 #### Provider Support
 - **OpenAI Provider**: Complete integration with dual API support
-  - Chat Completions API for standard models (GPT-4o, GPT-4.1)
-  - Responses API for reasoning models (o3, o4 series)
+  - Chat Completions API for standard custom models
+  - Responses API for GPT-5 models
   - Automatic API selection based on model capabilities
   - Parameter filtering for reasoning models
   - Full streaming support for both APIs
@@ -80,7 +80,7 @@ All notable changes to the Tachikoma project will be documented in this file.
 
 - **Anthropic Provider**: Native Claude API integration
   - Support for Claude 4 (Opus, Sonnet) with thinking modes
-  - Claude 3.5/3.7 series compatibility
+  - Claude 4.x series compatibility
   - Content block handling for multimodal inputs
   - System prompt separation
   - Server-Sent Events streaming

Examples/AI-CLI/README.md

@@ -25,7 +25,7 @@ ai-cli "What is the capital of France?"
 ai-cli --model claude "Explain quantum computing"
 
 # Stream the response
-ai-cli --stream --model gpt-4o "Write a short story"
+ai-cli --stream --model gpt-5.5 "Write a short story"
 ```
 
 ## Parameters
@@ -35,7 +35,7 @@ ai-cli --stream --model gpt-4o "Write a short story"
 | `-m, --model <MODEL>` | Specify the AI model to use |
 | `--api <chat\|responses>` | For OpenAI models: select API type (default: responses for GPT-5) |
 | `-s, --stream` | Stream the response in real-time |
-| `--thinking` | Show reasoning process (O3, O4, GPT-5 - note: API currently doesn't expose actual reasoning) |
+| `--thinking` | Show GPT-5 reasoning process (note: API currently doesn't expose actual reasoning) |
 | `--verbose, -v` | Show detailed debug output |
 | `--config` | Show current configuration and API key status |
 | `--help, -h` | Show help message |
@@ -60,15 +60,11 @@ Add to your shell profile (`~/.zshrc`, `~/.bashrc`) for persistence.
 ## Supported Models
 
 ### OpenAI
-- **GPT-5 Series**: `gpt-5`, `gpt-5-mini`, `gpt-5-nano`
-- **O-Series**: `o3`, `o3-mini`, `o3-pro`, `o4-mini`
-- **GPT-4**: `gpt-4.1`, `gpt-4.1-mini`, `gpt-4o`, `gpt-4o-mini`, `gpt-4-turbo`
-- **Legacy**: `gpt-3.5-turbo`
+- **GPT-5 Series**: `gpt-5.5`, `gpt-5.2`, `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`
 
 ### Anthropic
 - **Claude 4**: `claude-opus-4-1-20250805`, `claude-sonnet-4-20250514`
-- **Claude 3.7**: `claude-3-7-sonnet`
-- **Claude 3.5**: `claude-3-5-opus`, `claude-3-5-sonnet`, `claude-3-5-haiku`
+- **Claude 4.x**: `claude-opus-4-7`, `claude-sonnet-4.5`, `claude-haiku-4.5`
 
 ### Google
 - **Gemini 2.5**: `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite`
@@ -81,7 +77,7 @@ Add to your shell profile (`~/.zshrc`, `~/.bashrc`) for persistence.
 
 ### Model Shortcuts
 - `claude` → claude-opus-4-1-20250805
-- `gpt` → gpt-4.1
+- `gpt` → gpt-5.5
 - `gemini` → gemini-2.5-flash
 - `grok` → grok-4-fast-reasoning
 - `llama` → llama3.3
@@ -98,7 +94,7 @@ ai-cli --stream --model gpt-5 "Explain the theory of relativity"
 
 # API selection for OpenAI
 ai-cli --model gpt-5 --api chat "Use Chat Completions API"
-ai-cli --model o3 --api responses "Use Responses API"
+ai-cli --model gpt-5-thinking --api responses "Use Responses API"
 
 # Debug mode
 ai-cli --verbose --model opus "Debug this request"

Examples/AI-CLI/Sources/AI-CLI.swift

@@ -190,7 +190,7 @@ struct AICLI {
             -m, --model <MODEL>     Specify the AI model to use
             --api <API>            For OpenAI models: 'chat' or 'responses' (default: responses for GPT-5)
             -s, --stream           Stream the response (partial support)
-            --thinking             Show reasoning/thinking process (O3, O4, GPT-5 via Responses API)
+            --thinking             Show reasoning/thinking process (GPT-5 via Responses API)
             --verbose, -v          Show detailed debug output
             --config               Show current configuration and exit
             -h, --help             Show this help message
@@ -202,7 +202,7 @@ struct AICLI {
 
             # Use specific models
             ai-cli --model claude "Explain quantum computing"
-            ai-cli --model gpt-4o "Describe this image"
+            ai-cli --model gpt-5.5 "Describe this image"
             ai-cli --model grok "Tell me a joke"
             ai-cli --model llama3.3 "Help me debug this code"
 
@@ -222,14 +222,12 @@ struct AICLI {
         OpenAI:
           • gpt-5, gpt-5-pro, gpt-5-mini, gpt-5-nano (GPT-5 series, August 2025)
           • gpt-5-thinking, gpt-5-thinking-mini, gpt-5-thinking-nano
-          • gpt-4.1, gpt-4.1-mini, o4-mini (GPT-4.1 / reasoning)
-          • gpt-4o, gpt-4o-mini (Multimodal)
-          • gpt-4-turbo (Legacy)
+          • gpt-5.5, gpt-5.2, gpt-5.1 (flagship)
+          • gpt-5.5, gpt-5-mini (multimodal)
 
         Anthropic:
           • claude-opus-4-1-20250805, claude-sonnet-4-20250514 (Claude 4)
-          • claude-3-7-sonnet (Claude 3.7)
-          • claude-3-5-opus, claude-3-5-sonnet, claude-3-5-haiku (Claude 3.5)
+          • claude-opus-4-7, claude-sonnet-4.5, claude-haiku-4.5 (Claude 4.x)
 
         Google:
           • gemini-2.5-pro (reasoning, thinking support)
@@ -257,7 +255,7 @@ struct AICLI {
 
         SHORTCUTS:
           • claude, opus → claude-opus-4-1-20250805
-          • gpt, gpt4 → gpt-4.1
+          • gpt → gpt-5.5
           • grok → grok-4-fast-reasoning
           • gemini → gemini-2.5-flash
           • llama, llama3 → llama3.3
@@ -473,7 +471,7 @@ struct AICLI {
 
         let supportsThinking = self.isReasoningModel(model) && actualApiMode != .chat
         if config.showThinking, !supportsThinking {
-            print("⚠️  Note: --thinking only works with O3, O4, and GPT-5 models via Responses API")
+            print("⚠️  Note: --thinking only works with GPT-5 models via Responses API")
         }
 
         if case let .openai(openaiModel) = model, actualApiMode == .chat {
@@ -585,8 +583,7 @@ struct AICLI {
     static func isReasoningModel(_ model: LanguageModel) -> Bool {
         guard case let .openai(openaiModel) = model else { return false }
         switch openaiModel {
-        case .o4Mini,
-             .gpt5,
+        case .gpt5,
              .gpt5Pro,
              .gpt5Mini,
              .gpt5Nano,
@@ -672,7 +669,7 @@ struct AICLI {
                     }
                 }
 
-                // If no summary, try content array (for O3/O4)
+                // If no summary, try content array.
                 if reasoningText == nil || reasoningText?.isEmpty == true {
                     if let contentArray = output["content"] as? [[String: Any]] {
                         let reasoningParts = contentArray.compactMap { item -> String? in
@@ -801,15 +798,11 @@ struct AICLI {
         switch model {
         case let .openai(openaiModel):
             switch openaiModel {
-            case .gpt5: return nil // Pricing TBD
-            case .gpt5Mini: return nil // Pricing TBD
-            case .gpt5Nano: return nil // Pricing TBD
-            case .gpt4o:
+            case .gpt55, .gpt52, .gpt51, .gpt5:
                 inputCostPer1k = 0.005
-                outputCostPer1k = 0.015
-            case .gpt4oMini:
-                inputCostPer1k = 0.000_15
-                outputCostPer1k = 0.0006
+                outputCostPer1k = 0.020
+            case .gpt5Mini, .gpt5Nano:
+                return nil // Pricing TBD
             default: return nil
             }
         case let .anthropic(anthropicModel):

Examples/Advanced/TachikomaExamples.swift

@@ -56,8 +56,8 @@ extension TachikomaExamples {
         print("Creating provider-specific models with type safety:")
 
         // OpenAI models
-        let gpt4o = Model.openai(.gpt4o)
-        let gpt41 = Model.openai(.gpt41)
+        let gpt55 = Model.openai(.gpt55)
+        let gpt52 = Model.openai(.gpt52)
         let gpt5Mini = Model.openai(.gpt5Mini)
 
         // Anthropic models
@@ -75,11 +75,11 @@ extension TachikomaExamples {
 
         // Custom endpoints
         let openRouter = Model.openRouter(modelId: "anthropic/claude-sonnet-4.5")
-        let customAPI = Model.openaiCompatible(modelId: "gpt-4", baseURL: "https://api.azure.com")
+        let customAPI = Model.openaiCompatible(modelId: "gpt-5.5", baseURL: "https://api.azure.com")
 
         let models = [
-            gpt4o,
-            gpt41,
+            gpt55,
+            gpt52,
             gpt5Mini,
             opus4,
             sonnet4,
@@ -100,7 +100,7 @@ extension TachikomaExamples {
 
         // Model capabilities
         print("\nModel capabilities:")
-        print("  • Vision support: \(gpt4o.supportsVision)")
+        print("  • Vision support: \(gpt55.supportsVision)")
         print("  • Tool support: \(opus4.supportsTools)")
         print("  • Streaming support: \(sonnet4.supportsStreaming)")
     }
@@ -192,7 +192,7 @@ extension TachikomaExamples {
         // Record some usage
         try? await tracker.recordUsage(
             operation: .generation,
-            model: "gpt-4o".lowercased(),
+            model: "gpt-5.5".lowercased(),
             inputTokens: 100,
             outputTokens: 50,
             cost: 0.003,
@@ -273,7 +273,7 @@ extension TachikomaExamples {
 
         // Test different provider creations (these will fail without API keys, which is expected)
         let models: [(String, Model)] = [
-            ("OpenAI GPT-4o", .openai(.gpt4o)),
+            ("OpenAI GPT-5.5", .openai(.gpt55)),
             ("Anthropic Opus 4", .anthropic(.opus4)),
             ("Grok 4 Fast", .grok(.grok4FastReasoning)),
             ("Ollama Llama 3.3", .ollama(.llama3_3)),
@@ -319,7 +319,7 @@ extension String {
 extension Model {
     var supportsVision: Bool {
         switch self {
-        case .openai(.gpt4o), .grok(.grok2Vision), .ollama(.llava):
+        case .openai(.gpt55), .grok(.grok2Vision), .ollama(.llava):
             true
         default:
             false

Examples/Agent-CLI/README.md

@@ -50,7 +50,7 @@ In interactive mode:
 ```bash
 agent-cli --model gpt-5 "Explain quantum computing"
 agent-cli --model claude "Write a haiku"
-agent-cli --model o3 --thinking "Solve this step by step: ..."
+agent-cli --model gpt-5-thinking --thinking "Solve this step by step: ..."
 ```
 
 ### With MCP Tools
@@ -171,13 +171,11 @@ agent-cli --mcp-server "db -- npx @modelcontextprotocol/server-postgres postgres
 ## Supported Models
 
 ### OpenAI
-- GPT-5 series: `gpt-5`, `gpt-5-mini`, `gpt-5-nano`
-- O-series: `o3`, `o3-mini`, `o3-pro`, `o4-mini`
-- GPT-4: `gpt-4.1`, `gpt-4o`, `gpt-4-turbo`
+- GPT-5 series: `gpt-5.5`, `gpt-5.2`, `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`
 
 ### Anthropic
 - Claude 4: `opus-4`, `sonnet-4`
-- Claude 3.5: `claude`, `sonnet`, `haiku`
+- Claude 4.x: `claude`, `sonnet`, `haiku`
 
 ### Others
 - Google: `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite`
@@ -209,7 +207,7 @@ agent-cli --mcp-server "db -- npx @modelcontextprotocol/server-sqlite ./app.db"
 
 ### Complex Reasoning
 ```bash
-agent-cli --model o3 --thinking \
+agent-cli --model gpt-5-thinking --thinking \
           "Plan a distributed system architecture for a social media platform"
 ```
 

Examples/Agent-CLI/Sources/Agent-CLI/Agent.swift

@@ -163,8 +163,7 @@ final class Agent {
         switch model {
         case let .openai(openaiModel):
             switch openaiModel {
-            case .o4Mini,
-                 .gpt5,
+            case .gpt5,
                  .gpt5Pro,
                  .gpt5Mini,
                  .gpt5Nano,

Examples/Agent-CLI/Sources/Agent-CLI/AgentCLI.swift

@@ -27,7 +27,7 @@ struct AgentCLI: AsyncParsableCommand {
     @Argument(help: "Query or task for the agent")
     var query: String?
 
-    @Option(name: .shortAndLong, help: "AI model to use (e.g., gpt-5, claude, o3)")
+    @Option(name: .shortAndLong, help: "AI model to use (e.g., gpt-5.5, gpt-5-thinking, claude)")
     var model: String = "gpt-5"
 
     @Flag(name: .shortAndLong, help: "Interactive conversation mode")

Examples/Demos/TachikomaDemo.sh

@@ -49,7 +49,7 @@ echo "  • Stream token handling"
 echo ""
 
 echo "🔑 API Integration Examples (require valid API keys):"
-echo "  • OpenAI GPT-4o, GPT-4.1, o3 generation"
+echo "  • OpenAI GPT-5 generation"
 echo "  • Anthropic Claude Opus 4, Sonnet 4 generation"
 echo "  • Grok 4 and Grok 2 Vision models"
 echo "  • Ollama local models (llama3.3, llava)"
@@ -70,7 +70,7 @@ echo "🚀 How to Use Tachikoma:"
 echo "========================"
 echo ""
 echo "1. Basic Generation:"
-echo '   let answer = try await generate("What is 2+2?", using: .openai(.gpt4o))'
+echo '   let answer = try await generate("What is 2+2?", using: .openai(.gpt55))'
 echo ""
 echo "2. With Tools:"
 echo '   @ToolKit'
@@ -88,4 +88,4 @@ echo '   let response = try await conversation.continue(using: .anthropic(.opus4
 echo ""
 
 echo "🕷️  Tachikoma - Intelligent • Adaptable • Reliable"
-echo "   All examples completed successfully!"
+echo "   All examples completed successfully!"

Examples/HarmonyFeatures.swift

@@ -9,7 +9,7 @@ func demonstrateMultiChannelResponse() async throws {
     print("=== Multi-Channel Response Demo ===\n")
 
     let result = try await generateText(
-        model: .openai(.gpt4o),
+        model: .openai(.gpt55),
         messages: [
             .user("Explain how recursion works in programming"),
         ],
@@ -36,7 +36,7 @@ func demonstrateReasoningEffort() async throws {
     // High effort for complex problem
     print("High effort response:")
     let complexResult = try await generateText(
-        model: .openai(.gpt4o),
+        model: .openai(.gpt55),
         messages: [
             .user("Design a distributed system for real-time collaboration"),
         ],
@@ -50,7 +50,7 @@ func demonstrateReasoningEffort() async throws {
     // Low effort for simple query
     print("\nLow effort response:")
     let simpleResult = try await generateText(
-        model: .openai(.gpt4o),
+        model: .openai(.gpt55),
         messages: [
             .user("What is the capital of Japan?"),
         ],
@@ -82,7 +82,7 @@ func demonstrateRetryHandler() async throws {
                 print("Attempting API call...")
                 // Simulate a call that might fail
                 return try await generateText(
-                    model: .openai(.gpt4o),
+                    model: .openai(.gpt55),
                     messages: [.user("Hello")],
                 )
             },
@@ -144,7 +144,7 @@ func demonstrateEnhancedTools() async throws {
     }
 
     let result = try await generateText(
-        model: .openai(.gpt4o),
+        model: .openai(.gpt55),
         messages: [
             .user("What's 25 * 4 and what's the weather in Tokyo?"),
         ],
@@ -278,7 +278,7 @@ func demonstrateIntegratedFeatures() async throws {
     do {
         let result = try await retryHandler.execute {
             try await generateText(
-                model: .openai(.gpt4o),
+                model: .openai(.gpt55),
                 messages: [
                     .system("You are a helpful assistant. Use channels to organize your response."),
                     .user("Analyze the benefits of functional programming"),

Examples/RealtimeAPIDemo.swift

@@ -16,7 +16,7 @@ public struct RealtimeAPIDemo {
         // 1. Create configuration
         print("\n1️⃣ Creating Configuration...")
         let config = SessionConfiguration(
-            model: "gpt-4o-realtime-preview",
+            model: "gpt-realtime",
             voice: .nova,
             instructions: "You are a helpful voice assistant",
             inputAudioFormat: .pcm16,
@@ -108,7 +108,7 @@ public struct RealtimeAPIDemo {
         Example:
         let conversation = try RealtimeConversation(configuration: config)
         try await conversation.start(
-            model: .gpt4oRealtime,
+            model: .custom("gpt-realtime"),
             voice: .nova
         )
         ════════════════════════════════════════════════
@@ -118,7 +118,7 @@ public struct RealtimeAPIDemo {
     /// Create a sample configuration for testing
     public static func createSampleConfiguration() -> SessionConfiguration {
         SessionConfiguration.voiceConversation(
-            model: "gpt-4o-realtime-preview",
+            model: "gpt-realtime",
             voice: .nova,
         )
     }
@@ -127,7 +127,7 @@ public struct RealtimeAPIDemo {
     public static func validateTypes() -> Bool {
         // Test configuration creation
         let config = self.createSampleConfiguration()
-        guard config.model == "gpt-4o-realtime-preview" else { return false }
+        guard config.model == "gpt-realtime" else { return false }
 
         // Test VAD configuration
         let vad = RealtimeTurnDetection.serverVAD

Examples/RealtimeExample.swift

@@ -49,7 +49,7 @@ class RealtimeVoiceAssistant {
 
         // Start the conversation
         self.conversation = try await startRealtimeConversation(
-            model: .gpt4oRealtime,
+            model: .custom("gpt-realtime"),
             voice: .nova,
             instructions: """
             You are a helpful voice assistant. Keep responses concise and natural.
@@ -151,7 +151,7 @@ class RealtimeDemo {
 
         // Server VAD automatically detects when user starts/stops speaking
         let conversation = try await startRealtimeConversation(
-            model: .gpt4oRealtime,
+            model: .custom("gpt-realtime"),
             voice: .alloy,
             instructions: "You are a voice assistant with server-side voice activity detection",
         )
@@ -171,7 +171,7 @@ class RealtimeDemo {
 
         // Configure for both text and audio responses
         let conversation = try await startRealtimeConversation(
-            model: .gpt4oRealtime,
+            model: .custom("gpt-realtime"),
             voice: .shimmer,
         )
 
@@ -222,7 +222,7 @@ class RealtimeDemo {
         }
 
         let conversation = try await startRealtimeConversation(
-            model: .gpt4oRealtime,
+            model: .custom("gpt-realtime"),
             voice: .echo,
             instructions: "You are a smart home assistant. Use the available tools to control devices.",
             tools: [smartHomeTool],
@@ -315,7 +315,7 @@ struct RealtimeVoiceView: View {
     private func setupConversation() async {
         do {
             conversation = try await startRealtimeConversation(
-                model: .gpt4oRealtime,
+                model: .custom("gpt-realtime"),
                 voice: .nova,
             )
 

Examples/RealtimeQuickTest.swift

@@ -13,7 +13,7 @@ func testRealtimeConfiguration() async throws {
     // Test 1: Session Configuration
     print("\n1️⃣ Testing Session Configuration:")
     let voiceConfig = SessionConfiguration.voiceConversation(
-        model: "gpt-4o-realtime-preview",
+        model: "gpt-realtime",
         voice: .nova,
     )
     print("   ✅ Model: \(voiceConfig.model)")

Examples/RealtimeUsageExamples.swift

@@ -24,7 +24,7 @@ class BasicVoiceAssistant: ObservableObject {
 
         // Start conversation with voice
         try await self.conversation?.start(
-            model: .gpt4oRealtime,
+            model: .custom("gpt-realtime"),
             voice: .nova,
             instructions: "You are a helpful voice assistant. Keep responses concise.",
         )
@@ -307,7 +307,7 @@ class AudioStreamingExample {
     func setupAudioStreaming(apiKey: String) async throws {
         // Configure for audio streaming
         var config = EnhancedSessionConfiguration(
-            model: "gpt-4o-realtime-preview",
+            model: "gpt-realtime",
             voice: .echo,
             inputAudioFormat: .pcm16,
             outputAudioFormat: .pcm16,
@@ -371,7 +371,7 @@ class MultiTurnConversation {
     func runConversation(apiKey: String) async throws {
         // Configure for multi-turn dialogue
         let config = EnhancedSessionConfiguration(
-            model: "gpt-4o-realtime-preview",
+            model: "gpt-realtime",
             voice: .fable,
             instructions: """
             You are a knowledgeable assistant engaged in a multi-turn conversation.
@@ -547,7 +547,7 @@ class VoiceAssistantViewController: UIViewController {
                 self.conversation = try RealtimeConversation(configuration: config)
 
                 try await self.conversation?.start(
-                    model: .gpt4oRealtime,
+                    model: .custom("gpt-realtime"),
                     voice: .shimmer,
                 )
 

Examples/RealtimeVoiceAssistant.swift

@@ -29,7 +29,7 @@ class RealtimeVoiceAssistant {
 
         // Start with voice configuration
         try await conversation.start(
-            model: .gpt4oRealtime,
+            model: .custom("gpt-realtime"),
             voice: .nova,
             instructions: "You are a helpful, witty, and friendly AI assistant. Keep responses concise.",
         )
@@ -77,7 +77,7 @@ class RealtimeVoiceAssistant {
 
         // Advanced configuration with all features
         let config = SessionConfiguration(
-            model: "gpt-4o-realtime-preview",
+            model: "gpt-realtime",
             voice: .nova,
             instructions: """
             You are an expert AI assistant with deep knowledge across many domains.
@@ -147,7 +147,7 @@ class RealtimeVoiceAssistant {
 
         // Configuration with tools
         let config = SessionConfiguration.withTools(
-            model: "gpt-4o-realtime-preview",
+            model: "gpt-realtime",
             voice: .nova,
             tools: [
                 // Weather tool

Package.resolved

@@ -1,15 +1,6 @@
 {
-  "originHash" : "af7882d9f5dc7b555ad8239a69ab2881f08c29a54e8bc5c4a70f3482727b8abc",
+  "originHash" : "12a454cd38a6ae2519d652cc0872f7f18feb64690ce83d1507bae6db71c1841c",
   "pins" : [
-    {
-      "identity" : "commander",
-      "kind" : "remoteSourceControl",
-      "location" : "https://github.com/steipete/Commander.git",
-      "state" : {
-        "revision" : "ae2ce746b386ff94b26648cfe5625cfa8d02639b",
-        "version" : "0.2.2"
-      }
-    },
     {
       "identity" : "eventsource",
       "kind" : "remoteSourceControl",

README.md

@@ -99,7 +99,7 @@ print(result.text)
 
 Common picks:
 - Anthropic: `claude-opus-4-5` (`LanguageModel.default`)
-- OpenAI: `gpt-5.2` (flagship), `gpt-5` (coding/agents), `o4-mini` (reasoning), `gpt-4o` (vision)
+- OpenAI: `gpt-5.5` (flagship), `gpt-5` (coding/agents), `gpt-5-thinking` (reasoning)
 - Google: `gemini-3-flash`
 - Grok: `grok-4-fast-reasoning`
 - Local: `ollama/llama3.3`

Sources/Tachikoma/Auth/AuthManager.swift

@@ -575,7 +575,7 @@ struct TKProviderValidator {
             request.setValue("application/json", forHTTPHeaderField: "Content-Type")
             request.setValue("2023-06-01", forHTTPHeaderField: "anthropic-version")
             request.httpBody = try? JSONSerialization.data(withJSONObject: [
-                "model": "claude-3-haiku-20241022",
+                "model": "claude-haiku-4.5",
                 "max_tokens": 1,
                 "messages": [
                     ["role": "user", "content": "ping"],

Sources/Tachikoma/Core/Embeddings.swift

@@ -113,10 +113,10 @@ public enum EmbeddingModel: Sendable {
         // Convert to LanguageModel for usage tracking
         switch self {
         case .openai:
-            .openai(.gpt4o) // Placeholder for tracking
+            .openai(.gpt55) // Placeholder for tracking
         case .cohere, .voyage, .custom:
             // Return a dummy model for tracking purposes
-            .openai(.gpt4o)
+            .openai(.gpt55)
         }
     }
 }

Sources/Tachikoma/Core/Generation.swift

@@ -670,7 +670,7 @@ public func analyze(
         model
     } else {
         // Use a vision-capable model by default
-        .openai(.gpt4o)
+        .openai(.gpt55)
     }
 
     // Ensure the model supports vision

Sources/Tachikoma/Core/ModelCapabilities.swift

@@ -237,6 +237,7 @@ public final class ModelCapabilityRegistry: @unchecked Sendable {
             excludedParameters: ["temperature", "topP", "frequencyPenalty", "presencePenalty"],
         )
 
+        self.capabilities["openai:gpt-5.5"] = gpt5Capabilities
         self.capabilities["openai:gpt-5.1"] = gpt5Capabilities
         self.capabilities["openai:gpt-5.2"] = gpt5Capabilities
         self.capabilities["openai:gpt-5"] = gpt5Capabilities
@@ -248,36 +249,6 @@ public final class ModelCapabilityRegistry: @unchecked Sendable {
         self.capabilities["openai:gpt-5-thinking-nano"] = gpt5Capabilities
         self.capabilities["openai:gpt-5-chat-latest"] = gpt5Capabilities
 
-        // O4/GPT-5 reasoning models (fixed temperature, reasoning effort)
-        let reasoningCapabilities = ModelParameterCapabilities(
-            supportsTemperature: false,
-            supportsTopP: false,
-            supportedProviderOptions: .init(
-                supportsReasoningEffort: true,
-                supportsPreviousResponseId: true,
-            ),
-            forcedTemperature: 1.0,
-            excludedParameters: ["temperature", "topP"],
-        )
-
-        self.capabilities["openai:o4"] = reasoningCapabilities
-        self.capabilities["openai:o4-mini"] = reasoningCapabilities
-
-        // Standard GPT-4 models
-        let gpt4Capabilities = ModelParameterCapabilities(
-            supportedProviderOptions: .init(
-                supportsParallelToolCalls: true,
-                supportsResponseFormat: true,
-                supportsLogprobs: true,
-            ),
-        )
-
-        self.capabilities["openai:gpt-4o"] = gpt4Capabilities
-        self.capabilities["openai:gpt-4o-mini"] = gpt4Capabilities
-        self.capabilities["openai:gpt-4.1"] = gpt4Capabilities
-        self.capabilities["openai:gpt-4.1-mini"] = gpt4Capabilities
-        self.capabilities["openai:gpt-4-turbo"] = gpt4Capabilities
-
         // Claude 4 models with thinking
         let claude4Capabilities = ModelParameterCapabilities(
             supportedProviderOptions: .init(
@@ -286,6 +257,7 @@ public final class ModelCapabilityRegistry: @unchecked Sendable {
             ),
         )
 
+        self.capabilities["anthropic:claude-opus-4-7"] = claude4Capabilities
         self.capabilities["anthropic:claude-opus-4-5"] = claude4Capabilities
         self.capabilities["anthropic:claude-opus-4-1-20250805"] = claude4Capabilities
         self.capabilities["anthropic:claude-sonnet-4-20250514"] = claude4Capabilities

Sources/Tachikoma/Core/ProviderOptions.swift

@@ -56,7 +56,7 @@ public struct OpenAIOptions: Sendable, Codable {
     /// Verbosity level for GPT-5 models
     public var verbosity: Verbosity?
 
-    /// Reasoning effort for O3/O4 models
+    /// Reasoning effort for GPT-5 models
     public var reasoningEffort: ReasoningEffort?
 
     /// Previous response ID for Responses API chaining

Sources/Tachikoma/Core/Tachikoma.swift

@@ -27,7 +27,7 @@ import Foundation
 //
 // ```swift
 // // Simple generation
-// let answer = try await generate("What is 2+2?", using: .openai(.gpt4o))
+// let answer = try await generate("What is 2+2?", using: .openai(.gpt55))
 //
 // // Conversation management
 // let conversation = Conversation()
@@ -114,7 +114,7 @@ public enum API {
     /// Model selection system
     public enum Models {
         /// Type-safe model selection
-        /// - `.openai(.gpt4o)`, `.anthropic(.opus4)`, `.grok(.grok4)`, `.ollama(.llama3_3)`
+        /// - `.openai(.gpt55)`, `.anthropic(.opus47)`, `.grok(.grok4)`, `.ollama(.llama3_3)`
         public static let typed = "Provider-specific model enums"
 
         /// Custom endpoints
@@ -196,16 +196,16 @@ public enum API {
 /// Migration guide from legacy API to modern API
 @available(macOS 13.0, iOS 16.0, watchOS 9.0, tvOS 16.0, *)
 public enum MigrationGuide {
-    /// Legacy: `Tachikoma.shared.getModel("gpt-4").getResponse(request)`
-    /// Modern: `generate("Hello", using: .openai(.gpt4o))`
+    /// Legacy: `Tachikoma.shared.getModel("gpt-5.5").getResponse(request)`
+    /// Modern: `generate("Hello", using: .openai(.gpt55))`
     public static let simpleGeneration = """
     // OLD (deprecated)
-    let model = try await Tachikoma.shared.getModel("gpt-4")
+    let model = try await Tachikoma.shared.getModel("gpt-5.5")
     let request = ModelRequest(messages: [.user(content: .text("Hello"))], settings: .default)
     let response = try await model.getResponse(request: request)
 
     // NEW (modern)
-    let response = try await generate("Hello", using: .openai(.gpt4o))
+    let response = try await generate("Hello", using: .openai(.gpt55))
     """
 
     /// Legacy: Complex ModelRequest/ModelResponse handling

Sources/Tachikoma/Core/Types.swift

@@ -186,8 +186,9 @@ public enum OpenAIAPIMode: String, Sendable, CaseIterable {
     public static func defaultMode(for model: LanguageModel.OpenAI) -> OpenAIAPIMode {
         // Determine default API mode for a given model
         switch model {
-        case .o4Mini, .gpt5, .gpt5Pro, .gpt5Mini, .gpt5Nano, .gpt51, .gpt52:
-            .responses // Reasoning models and GPT-5 default to Responses API
+        case .gpt5, .gpt5Pro, .gpt5Mini, .gpt5Nano, .gpt51, .gpt52, .gpt55,
+             .gpt5Thinking, .gpt5ThinkingMini, .gpt5ThinkingNano, .gpt5ChatLatest:
+            .responses // GPT-5 defaults to Responses API
         default:
             .chat // All other models use Chat Completions API
         }
@@ -614,7 +615,7 @@ public enum ResponseChannel: String, Sendable, Codable, CaseIterable {
     case final // Final answer to the user
 }
 
-/// Reasoning effort level for models that support it (o3, opus-4, etc.)
+/// Reasoning effort level for models that support it (GPT-5 thinking, opus-4, etc.)
 @available(macOS 13.0, iOS 16.0, watchOS 9.0, tvOS 16.0, *)
 public enum ReasoningEffort: String, Sendable, Codable, CaseIterable {
     case low

Sources/Tachikoma/Models/Model.swift

@@ -31,8 +31,8 @@ public enum LanguageModel: Sendable, CustomStringConvertible, Hashable {
     // MARK: - Provider Sub-Enums
 
     public enum OpenAI: Sendable, Hashable, CaseIterable {
-        /// Latest models (2025)
-        case o4Mini
+        /// GPT-5.5 Series
+        case gpt55 // Flagship GPT-5.5
 
         /// GPT-5.2 Series
         case gpt52 // Flagship GPT-5.2
@@ -50,25 +50,12 @@ public enum LanguageModel: Sendable, CustomStringConvertible, Hashable {
         case gpt5ThinkingNano
         case gpt5ChatLatest // Non-reasoning default chat deployment
 
-        // GPT-4.1 Series
-        case gpt41
-        case gpt41Mini
-
-        // GPT-4o Series (Multimodal)
-        case gpt4o
-        case gpt4oMini
-        case gpt4oRealtime // Realtime API support
-
-        // Legacy support
-        case gpt4Turbo
-        case gpt35Turbo
-
         /// Fine-tuned models
         case custom(String)
 
         public static var allCases: [OpenAI] {
             [
-                .o4Mini,
+                .gpt55,
                 .gpt52,
                 .gpt51,
                 .gpt5,
@@ -79,20 +66,13 @@ public enum LanguageModel: Sendable, CustomStringConvertible, Hashable {
                 .gpt5ThinkingMini,
                 .gpt5ThinkingNano,
                 .gpt5ChatLatest,
-                .gpt41,
-                .gpt41Mini,
-                .gpt4o,
-                .gpt4oMini,
-                .gpt4oRealtime,
-                .gpt4Turbo,
-                .gpt35Turbo,
             ]
         }
 
         public var modelId: String {
             switch self {
             case let .custom(id): id
-            case .o4Mini: "o4-mini"
+            case .gpt55: "gpt-5.5"
             case .gpt52: "gpt-5.2"
             case .gpt51: "gpt-5.1"
             case .gpt5: "gpt-5"
@@ -103,83 +83,80 @@ public enum LanguageModel: Sendable, CustomStringConvertible, Hashable {
             case .gpt5ThinkingMini: "gpt-5-thinking-mini"
             case .gpt5ThinkingNano: "gpt-5-thinking-nano"
             case .gpt5ChatLatest: "gpt-5-chat-latest"
-            case .gpt41: "gpt-4.1"
-            case .gpt41Mini: "gpt-4.1-mini"
-            case .gpt4o: "gpt-4o"
-            case .gpt4oMini: "gpt-4o-mini"
-            case .gpt4oRealtime: "gpt-4o-realtime-preview"
-            case .gpt4Turbo: "gpt-4-turbo"
-            case .gpt35Turbo: "gpt-3.5-turbo"
             }
         }
 
         public var supportsVision: Bool {
             switch self {
-            case .gpt52,
+            case .gpt55,
+                 .gpt52,
                  .gpt51,
                  .gpt5, .gpt5Pro, .gpt5Mini, .gpt5Nano, .gpt5Thinking, .gpt5ThinkingMini, .gpt5ThinkingNano,
                  .gpt5ChatLatest: true // GPT-5+ supports multimodal
-            case .gpt4o, .gpt4oMini, .gpt4oRealtime: true
             default: false
             }
         }
 
         public var supportsTools: Bool {
             switch self {
-            case .o4Mini: true
-            case .gpt52,
+            case .gpt55,
+                 .gpt52,
                  .gpt51,
                  .gpt5, .gpt5Pro, .gpt5Mini, .gpt5Nano, .gpt5Thinking, .gpt5ThinkingMini, .gpt5ThinkingNano,
                  .gpt5ChatLatest: true // GPT-5+ excels at tool calling
-            case .gpt41, .gpt41Mini, .gpt4o, .gpt4oMini, .gpt4oRealtime, .gpt4Turbo: true
-            case .gpt35Turbo: true
             case .custom: true // Assume custom models support tools
             }
         }
 
         public var supportsAudioInput: Bool {
             switch self {
-            case .gpt52,
+            case .gpt55,
+                 .gpt52,
                  .gpt51,
                  .gpt5, .gpt5Pro, .gpt5Mini, .gpt5Nano, .gpt5Thinking, .gpt5ThinkingMini, .gpt5ThinkingNano,
                  .gpt5ChatLatest: true // GPT-5+ is fully multimodal
-            case .gpt4o, .gpt4oMini, .gpt4oRealtime: true // GPT-4o models support native audio input
             default: false
             }
         }
 
         public var supportsAudioOutput: Bool {
             switch self {
-            case .gpt4oRealtime: true // Realtime API supports native audio output
+            case let .custom(id): id.contains("realtime")
             default: false
             }
         }
 
         public var supportsRealtime: Bool {
             switch self {
-            case .gpt4oRealtime: true
+            case let .custom(id): id.contains("realtime")
             default: false
             }
         }
 
         public var contextLength: Int {
             switch self {
-            case .o4Mini: 128_000
-            case .gpt52,
+            case .gpt55,
+                 .gpt52,
                  .gpt51,
                  .gpt5, .gpt5Pro, .gpt5Mini, .gpt5Nano, .gpt5Thinking, .gpt5ThinkingMini, .gpt5ThinkingNano,
                  .gpt5ChatLatest: 400_000 // 272k input + 128k output
-            case .gpt41, .gpt41Mini: 1_000_000
-            case .gpt4o, .gpt4oMini, .gpt4oRealtime: 128_000
-            case .gpt4Turbo: 128_000
-            case .gpt35Turbo: 16000
             case .custom: 128_000 // Default assumption
             }
         }
+
+        public var isUnsupportedLegacyFamily: Bool {
+            let normalized = self.modelId.lowercased()
+            let compact = normalized.replacingOccurrences(of: "-", with: "")
+                .replacingOccurrences(of: ".", with: "")
+            return normalized.hasPrefix("gpt-4") || compact.hasPrefix("gpt4") ||
+                normalized.hasPrefix("gpt-3") || compact.hasPrefix("gpt3") ||
+                normalized.hasPrefix("o3") || normalized.hasPrefix("o4")
+        }
     }
 
     public enum Anthropic: Sendable, Hashable, CaseIterable {
-        // Claude 4.x / 4.5 Series (2025)
+        // Claude 4.x / 4.5+ Series
+        case opus47
         case opus45
         case opus4
         case opus4Thinking
@@ -193,6 +170,7 @@ public enum LanguageModel: Sendable, CustomStringConvertible, Hashable {
 
         public static var allCases: [Anthropic] {
             [
+                .opus47,
                 .opus45,
                 .opus4,
                 .opus4Thinking,
@@ -206,6 +184,7 @@ public enum LanguageModel: Sendable, CustomStringConvertible, Hashable {
         public var modelId: String {
             switch self {
             case let .custom(id): id
+            case .opus47: "claude-opus-4-7"
             case .opus45: "claude-opus-4-5"
             case .opus4: "claude-opus-4-1-20250805"
             case .opus4Thinking: "claude-opus-4-1-20250805-thinking"
@@ -218,7 +197,8 @@ public enum LanguageModel: Sendable, CustomStringConvertible, Hashable {
 
         public var supportsVision: Bool {
             switch self {
-            case .opus45, .opus4, .opus4Thinking, .sonnet4, .sonnet4Thinking, .sonnet45, .haiku45: true
+            case .opus47, .opus45, .opus4, .opus4Thinking, .sonnet4, .sonnet4Thinking, .sonnet45, .haiku45:
+                true
             case .custom: true // Assume custom models support vision
             }
         }
@@ -239,6 +219,7 @@ public enum LanguageModel: Sendable, CustomStringConvertible, Hashable {
 
         public var contextLength: Int {
             switch self {
+            case .opus47: 1_000_000
             case .opus45, .opus4, .opus4Thinking, .sonnet4, .sonnet4Thinking, .sonnet45, .haiku45: 500_000
             case .custom: 200_000 // Default assumption
             }
@@ -869,15 +850,12 @@ public enum LanguageModel: Sendable, CustomStringConvertible, Hashable {
 
     // MARK: - Default Model
 
-    public static let `default`: LanguageModel = .anthropic(.opus45)
+    public static let `default`: LanguageModel = .anthropic(.opus47)
 
     // MARK: - Convenience Static Properties
 
-    /// Default Claude model (opus45)
-    public static let claude: LanguageModel = .anthropic(.opus45)
-
-    /// Default GPT-4o model
-    public static let gpt4o: LanguageModel = .openai(.gpt4o)
+    /// Default Claude model (opus47)
+    public static let claude: LanguageModel = .anthropic(.opus47)
 
     /// Default Grok model (Grok-4-0709)
     public static let grok4: LanguageModel = .grok(.grok4)
@@ -1137,6 +1115,10 @@ extension LanguageModel {
 
         // MARK: OpenAI models
 
+        if compact.contains("gpt4") || compact.contains("gpt3") || compact.contains("o3") || compact.contains("o4") {
+            return nil
+        }
+
         if dashed == "gpt-5-pro" || compact == "gpt5pro" {
             return .openai(.gpt5Pro)
         }
@@ -1151,6 +1133,13 @@ extension LanguageModel {
             return .openai(.gpt5Thinking)
         }
 
+        if dotted.contains("gpt-5-5") || compact.contains("gpt55") {
+            // GPT-5.5 currently has no mini/nano variants; map those suffixes to GPT-5 mini/nano.
+            if dotted.contains("nano") || compact.contains("nano") { return .openai(.gpt5Nano) }
+            if dotted.contains("mini") || compact.contains("mini") { return .openai(.gpt5Mini) }
+            return .openai(.gpt55)
+        }
+
         if dotted.contains("gpt-5-2") || compact.contains("gpt52") {
             // GPT-5.2 currently has no mini/nano variants; map those suffixes to GPT-5 mini/nano.
             if dotted.contains("nano") || compact.contains("nano") { return .openai(.gpt5Nano) }
@@ -1181,40 +1170,23 @@ extension LanguageModel {
             return .openai(.gpt5)
         }
 
-        if dotted.contains("gpt-4o-realtime") || compact.contains("gpt4orealtime") {
-            return .openai(.gpt4oRealtime)
-        }
+        // MARK: Anthropic models
 
-        if dotted.contains("gpt-4o-mini") || compact.contains("gpt4omini") {
-            return .openai(.gpt4oMini)
-        }
-
-        if dotted.contains("gpt-4o") || compact.contains("gpt4o") {
-            return .openai(.gpt4o)
-        }
-
-        if dotted.contains("gpt-4.1-mini") || compact.contains("gpt41mini") {
-            return .openai(.gpt41Mini)
-        }
-
-        if dotted.contains("gpt-4.1") || compact.contains("gpt41") {
-            return .openai(.gpt41)
+        if dotted.contains("claude-3") || compact.contains("claude3") {
+            return nil
         }
 
         if
-            dashed == "o3" || compact == "o3" || dashed == "o3-pro" || dashed == "o3-mini" || compact == "o3mini" ||
-            compact == "o3pro"
+            dotted.contains("claude-opus-4-7") ||
+            dotted.contains("claude-opus-4.7") ||
+            compact.contains("claudeopus47") ||
+            dotted.contains("opus-4-7") ||
+            dotted.contains("opus-4.7") ||
+            compact.contains("opus47")
         {
-            // o3 family is deprecated; steer callers to GPT-5.1 Mini
-            return .openai(.gpt5Mini)
+            return .anthropic(.opus47)
         }
 
-        if dashed == "o4-mini" || compact == "o4mini" {
-            return .openai(.o4Mini)
-        }
-
-        // MARK: Anthropic models
-
         if
             dotted.contains("claude-opus-4-5") ||
             dotted.contains("claude-opus-4.5") ||
@@ -1268,7 +1240,7 @@ extension LanguageModel {
 
         let canonicalForms = [normalized, dashed, compact]
         if canonicalForms.contains(where: { genericClaudeIdentifiers.contains($0) }) {
-            return .anthropic(.sonnet45)
+            return .anthropic(.opus47)
         }
 
         // MARK: Google models
@@ -1385,10 +1357,6 @@ extension LanguageModel {
             return .openai(.gpt5Mini)
         }
 
-        if compact.contains("o4") {
-            return .openai(.o4Mini)
-        }
-
         return nil
     }
 }

Sources/Tachikoma/Models/ModelInterface.swift

@@ -88,7 +88,7 @@ public struct ModelSettings: Sendable, Codable {
 
     /// Legacy initializer for backward compatibility
     public init(
-        modelName: String = "gpt-4",
+        modelName: String = "gpt-5.5",
         maxTokens: Int? = nil,
         temperature: Double? = nil,
         topP: Double? = nil,

Sources/Tachikoma/Models/ModelSelection.swift

@@ -35,6 +35,10 @@ public struct ModelSelector {
             return .grok(grokModel)
         }
 
+        if Self.isUnsupportedLegacyOpenAIModel(normalized) || Self.isUnsupportedLegacyAnthropicModel(normalized) {
+            throw ModelValidationError.unsupportedModel(modelString)
+        }
+
         // Ollama shortcuts and models
         if let ollamaModel = parseOllamaModel(normalized) {
             return .ollama(ollamaModel)
@@ -46,7 +50,7 @@ public struct ModelSelector {
         }
 
         // Custom model ID - try to infer provider
-        if normalized.contains("gpt") || normalized.contains("o3") || normalized.contains("o4") {
+        if normalized.contains("gpt-5") || normalized.contains("gpt5") {
             return .openai(.custom(normalized))
         }
 
@@ -66,6 +70,13 @@ public struct ModelSelector {
 
     private static func parseOpenAIModel(_ input: String) -> Model.OpenAI? {
         switch input {
+        // GPT-5.5 models
+        case "gpt-5.5", "gpt5.5", "gpt-5-5", "gpt5-5", "gpt55":
+            return .gpt55
+        case "gpt-5.5-mini", "gpt5.5-mini", "gpt55-mini", "gpt55mini", "gpt-5-5-mini", "gpt5-5-mini":
+            return .gpt5Mini
+        case "gpt-5.5-nano", "gpt5.5-nano", "gpt55-nano", "gpt55nano", "gpt-5-5-nano", "gpt5-5-nano":
+            return .gpt5Nano
         // GPT-5.2 models
         case "gpt-5.2", "gpt5.2", "gpt-5-2", "gpt5-2", "gpt52":
             return .gpt52
@@ -97,27 +108,17 @@ public struct ModelSelector {
             return .gpt5ThinkingNano
         case "gpt-5-chat-latest", "gpt5-chat-latest":
             return .gpt5ChatLatest
-        // Direct matches
-        case "gpt-4o", "gpt4o":
-            return .gpt4o
-        case "gpt-4o-mini", "gpt4o-mini", "gpt4omini":
-            return .gpt4oMini
-        case "gpt-4.1", "gpt4.1", "gpt41":
-            return .gpt41
-        case "gpt-4.1-mini", "gpt4.1-mini", "gpt41mini":
-            return .gpt41Mini
-        case "o4-mini", "o4mini":
-            return .o4Mini
         // Shortcuts
         case "gpt":
-            return .gpt51 // Default to flagship GPT-5.1
-        case "gpt4", "gpt-4":
-            return .gpt4o // Default to latest GPT-4 variant
+            return .gpt55 // Default to flagship GPT-5.5
         case "openai":
-            return .gpt51 // Default to GPT-5.1
+            return .gpt55 // Default to GPT-5.5
         default:
             // Check if it's an OpenAI model ID
-            if input.hasPrefix("gpt") || input.hasPrefix("o4") {
+            if Self.isUnsupportedLegacyOpenAIModel(input) {
+                return nil
+            }
+            if input.hasPrefix("gpt-5") || input.hasPrefix("gpt5") {
                 return .custom(input)
             }
             return nil
@@ -131,6 +132,8 @@ public struct ModelSelector {
             return .opus4
         case "claude-opus-4-20250514-thinking":
             return .opus4Thinking
+        case "claude-opus-4-7", "claude-opus-4.7", "opus-4-7", "opus-4.7", "opus47":
+            return .opus47
         case "claude-opus-4-5", "claude-opus-4.5", "opus-4-5", "opus-4.5", "opus45":
             return .opus45
         case "claude-sonnet-4-20250514":
@@ -141,17 +144,20 @@ public struct ModelSelector {
             return .sonnet45
         // Shortcuts
         case "claude":
-            return .sonnet45 // Default plain Claude alias to latest Sonnet
+            return .opus47
         case "claude-opus", "opus":
-            return .opus45
+            return .opus47
         case "claude-sonnet", "sonnet":
             return .sonnet4
         case "claude-haiku", "haiku":
             return .haiku45
         case "anthropic":
-            return .opus45 // Default Anthropic model
+            return .opus47 // Default Anthropic model
         default:
             // Check if it's a Claude model ID
+            if Self.isUnsupportedLegacyAnthropicModel(input) {
+                return nil
+            }
             if input.hasPrefix("claude") {
                 return .custom(input)
             }
@@ -159,6 +165,20 @@ public struct ModelSelector {
         }
     }
 
+    private static func isUnsupportedLegacyOpenAIModel(_ input: String) -> Bool {
+        let normalized = input.lowercased()
+        let compact = normalized.replacingOccurrences(of: "-", with: "").replacingOccurrences(of: ".", with: "")
+        return normalized.hasPrefix("gpt-4") || compact.hasPrefix("gpt4") ||
+            normalized.hasPrefix("gpt-3") || compact.hasPrefix("gpt3") ||
+            normalized.hasPrefix("o3") || normalized.hasPrefix("o4")
+    }
+
+    private static func isUnsupportedLegacyAnthropicModel(_ input: String) -> Bool {
+        let normalized = input.lowercased()
+        let compact = normalized.replacingOccurrences(of: "-", with: "").replacingOccurrences(of: ".", with: "")
+        return normalized.hasPrefix("claude-3") || compact.hasPrefix("claude3")
+    }
+
     private static func parseGoogleModel(_ input: String) -> Model.Google? {
         switch input {
         case "gemini-3-flash", "gemini-3-flash-preview", "gemini3flash", "gemini-3flash":
@@ -388,14 +408,14 @@ public func getAllAvailableModels() -> String {
     )
 
     output += "\nShortcuts:\n"
-    output += "  • claude, claude-opus, opus → claude-opus-4-20250514\n"
-    output += "  • gpt, gpt4 → gpt-4.1\n"
+    output += "  • claude, claude-opus, opus → claude-opus-4-7\n"
+    output += "  • gpt → gpt-5.5\n"
     output += "  • gemini → gemini-3-flash\n"
     output += "  • grok → grok-4-fast-reasoning\n"
     output += "  • llama, llama3 → llama3.3\n"
 
     output += "\nCustom Models:\n"
-    output += "  • OpenRouter: anthropic/claude-3.5-sonnet\n"
+    output += "  • OpenRouter: anthropic/claude-opus-4-7\n"
     output += "  • Custom OpenAI: custom-gpt-model\n"
     output += "  • Local Ollama: any-model:tag\n"
 
@@ -423,15 +443,15 @@ extension ModelSelector {
         // Get recommended models for specific use cases
         switch useCase {
         case .coding:
-            [.claude, .gpt4o, .google(.gemini25Pro)]
+            [.claude, .openai(.gpt55), .google(.gemini25Pro)]
         case .vision:
-            [.claude, .gpt4o, .google(.gemini3Flash)]
+            [.claude, .openai(.gpt55), .google(.gemini3Flash)]
         case .reasoning:
             [.openai(.gpt5Mini), .claude, .google(.gemini25Pro)]
         case .local:
             [.llama, .ollama(.mistralNemo), .ollama(.commandRPlus)]
         case .general:
-            [.claude, .gpt4o, .google(.gemini3Flash), .grok(.grok4FastReasoning), .llama]
+            [.claude, .openai(.gpt55), .google(.gemini3Flash), .grok(.grok4FastReasoning), .llama]
         }
     }
 }
@@ -449,6 +469,7 @@ public enum UseCase {
 public enum ModelValidationError: Error, LocalizedError {
     case visionNotSupported(String)
     case toolsNotSupported(String)
+    case unsupportedModel(String)
 
     public var errorDescription: String? {
         switch self {
@@ -456,6 +477,8 @@ public enum ModelValidationError: Error, LocalizedError {
             "Model '\(modelId)' does not support vision inputs"
         case let .toolsNotSupported(modelId):
             "Model '\(modelId)' does not support tool calling"
+        case let .unsupportedModel(modelId):
+            "Model '\(modelId)' is no longer supported"
         }
     }
 }

Sources/Tachikoma/Providers/OpenAI/OpenAIProvider.swift

@@ -20,6 +20,10 @@ public final class OpenAIProvider: ModelProvider {
         configuration: TachikomaConfiguration,
         session: URLSession = .shared,
     ) throws {
+        guard !model.isUnsupportedLegacyFamily else {
+            throw TachikomaError.unsupportedOperation("OpenAI model '\(model.modelId)' is no longer supported")
+        }
+
         self.model = model
         self.modelId = model.modelId
         self.baseURL = configuration.getBaseURL(for: .openai) ?? "https://api.openai.com/v1"

Sources/Tachikoma/Providers/OpenAI/OpenAIResponsesProvider.swift

@@ -3,7 +3,7 @@ import Foundation
 import FoundationNetworking
 #endif
 
-/// Provider for OpenAI Responses API (GPT-5, o3, o4)
+/// Provider for OpenAI Responses API (GPT-5)
 @available(macOS 13.0, iOS 16.0, watchOS 9.0, tvOS 16.0, *)
 public final class OpenAIResponsesProvider: ModelProvider {
     public let modelId: String
@@ -29,6 +29,10 @@ public final class OpenAIResponsesProvider: ModelProvider {
         configuration: TachikomaConfiguration,
         session: URLSession = .shared,
     ) throws {
+        guard !model.isUnsupportedLegacyFamily else {
+            throw TachikomaError.unsupportedOperation("OpenAI model '\(model.modelId)' is no longer supported")
+        }
+
         self.model = model
         self.modelId = model.modelId
         self.configuration = configuration
@@ -918,7 +922,7 @@ public final class OpenAIResponsesProvider: ModelProvider {
 
     private static func isReasoningModel(_ model: LanguageModel.OpenAI) -> Bool {
         switch model {
-        case .o4Mini:
+        case .gpt5Thinking, .gpt5ThinkingMini, .gpt5ThinkingNano:
             true
         default:
             false
@@ -927,7 +931,8 @@ public final class OpenAIResponsesProvider: ModelProvider {
 
     private static func isGPT5Model(_ model: LanguageModel.OpenAI) -> Bool {
         switch model {
-        case .gpt52,
+        case .gpt55,
+             .gpt52,
              .gpt51,
              .gpt5,
              .gpt5Pro,

Sources/Tachikoma/Providers/OpenAI/OpenAIResponsesTypes.swift

@@ -28,7 +28,7 @@ struct OpenAIResponsesRequest: Encodable {
     let serviceTier: String?
     let include: [String]?
 
-    /// Reasoning configuration (for o3/o4/GPT-5)
+    /// Reasoning configuration (for GPT-5 thinking models)
     let reasoning: ReasoningConfig?
 
     /// Truncation for long inputs

Sources/Tachikoma/Providers/ProviderFactory.swift

@@ -15,9 +15,9 @@ public struct ProviderFactory {
         // Create a provider for the specified language model
         switch model {
         case let .openai(openaiModel):
-            // Use Responses API for reasoning models (o4) and GPT-5 family
+            // Use Responses API for the GPT-5 family
             switch openaiModel {
-            case .o4Mini,
+            case .gpt55,
                  .gpt52,
                  .gpt51,
                  .gpt5,

Sources/Tachikoma/Providers/ProviderParser.swift

@@ -8,10 +8,10 @@ public enum ProviderParser {
         /// The provider name (e.g., "openai", "anthropic", "ollama")
         public let provider: String
 
-        /// The model name (e.g., "gpt-4", "claude-3", "llava:latest")
+        /// The model name (e.g., "gpt-5.5", "claude-opus-4-7", "llava:latest")
         public let model: String
 
-        /// The full string representation (e.g., "openai/gpt-4")
+        /// The full string representation (e.g., "openai/gpt-5.5")
         public var fullString: String {
             "\(self.provider)/\(self.model)"
         }
@@ -23,7 +23,7 @@ public enum ProviderParser {
     }
 
     /// Parse a provider string in the format "provider/model"
-    /// - Parameter providerString: String like "openai/gpt-4" or "ollama/llava:latest"
+    /// - Parameter providerString: String like "openai/gpt-5.5" or "ollama/llava:latest"
     /// - Returns: Parsed configuration or nil if invalid format
     public static func parse(_ providerString: String) -> ProviderConfig? {
         // Parse a provider string in the format "provider/model"
@@ -44,7 +44,7 @@ public enum ProviderParser {
     }
 
     /// Parse a comma-separated list of providers
-    /// - Parameter providersString: String like "openai/gpt-4,anthropic/claude-3,ollama/llava:latest"
+    /// - Parameter providersString: String like "openai/gpt-5.5,anthropic/claude-opus-4-7,ollama/llava:latest"
     /// - Returns: Array of parsed configurations
     public static func parseList(_ providersString: String) -> [ProviderConfig] {
         // Parse a comma-separated list of providers
@@ -54,7 +54,7 @@ public enum ProviderParser {
     }
 
     /// Get the first provider from a comma-separated list
-    /// - Parameter providersString: String like "openai/gpt-4,anthropic/claude-3"
+    /// - Parameter providersString: String like "openai/gpt-5.5,anthropic/claude-opus-4-7"
     /// - Returns: First parsed configuration or nil if none valid
     public static func parseFirst(_ providersString: String) -> ProviderConfig? {
         // Get the first provider from a comma-separated list
@@ -202,8 +202,22 @@ public enum ProviderParser {
     // MARK: - Private Helpers
 
     private static func parseOpenAIModel(_ modelString: String) -> LanguageModel? {
-        switch modelString.lowercased() {
-        case "o4-mini": .openai(.o4Mini)
+        let normalized = modelString.lowercased()
+        let compact = normalized.replacingOccurrences(of: "-", with: "").replacingOccurrences(of: ".", with: "")
+        guard
+            !normalized.hasPrefix("gpt-4"), !compact.hasPrefix("gpt4"),
+            !normalized.hasPrefix("gpt-3"), !compact.hasPrefix("gpt3"),
+            !normalized.hasPrefix("o3"), !normalized.hasPrefix("o4")
+        else {
+            return nil
+        }
+
+        return switch normalized {
+        case "gpt-5.5", "gpt5.5", "gpt-5-5", "gpt5-5", "gpt55": .openai(.gpt55)
+        case "gpt-5.5-mini", "gpt5.5-mini", "gpt-5-5-mini", "gpt5-5-mini", "gpt55-mini", "gpt55mini":
+            .openai(.gpt5Mini)
+        case "gpt-5.5-nano", "gpt5.5-nano", "gpt-5-5-nano", "gpt5-5-nano", "gpt55-nano", "gpt55nano":
+            .openai(.gpt5Nano)
         case "gpt-5.2", "gpt5.2", "gpt-5-2", "gpt5-2", "gpt52": .openai(.gpt52)
         case "gpt-5.2-mini", "gpt5.2-mini", "gpt-5-2-mini", "gpt5-2-mini", "gpt52-mini", "gpt52mini":
             .openai(.gpt5Mini)
@@ -222,12 +236,6 @@ public enum ProviderParser {
         case "gpt-5-thinking-mini", "gpt5-thinking-mini", "gpt5thinkingmini": .openai(.gpt5ThinkingMini)
         case "gpt-5-thinking-nano", "gpt5-thinking-nano", "gpt5thinkingnano": .openai(.gpt5ThinkingNano)
         case "gpt-5-chat-latest", "gpt5-chat-latest": .openai(.gpt5ChatLatest)
-        case "gpt-4.1", "gpt4.1": .openai(.gpt41)
-        case "gpt-4.1-mini", "gpt4.1-mini": .openai(.gpt41Mini)
-        case "gpt-4o", "gpt4o": .openai(.gpt4o)
-        case "gpt-4o-mini", "gpt4o-mini": .openai(.gpt4oMini)
-        case "gpt-4-turbo", "gpt4-turbo": .openai(.gpt4Turbo)
-        case "gpt-3.5-turbo", "gpt35-turbo": .openai(.gpt35Turbo)
         default:
             // Handle custom/fine-tuned models
             .openai(.custom(modelString))
@@ -235,7 +243,15 @@ public enum ProviderParser {
     }
 
     private static func parseAnthropicModel(_ modelString: String) -> LanguageModel? {
-        switch modelString.lowercased() {
+        let normalized = modelString.lowercased()
+        let compact = normalized.replacingOccurrences(of: "-", with: "").replacingOccurrences(of: ".", with: "")
+        guard !normalized.hasPrefix("claude-3"), !compact.hasPrefix("claude3") else {
+            return nil
+        }
+
+        return switch normalized {
+        case "claude-opus-4-7", "claude-opus-4.7", "claude-opus-4-7-latest", "opus-4-7", "opus-4.7", "opus47":
+            .anthropic(.opus47)
         case "claude-opus-4-5", "claude-opus-4.5", "claude-opus-4-5-latest", "opus-4-5", "opus-4.5", "opus45":
             .anthropic(.opus45)
         case "claude-opus-4-1-20250805", "claude-opus-4-20250514", "claude-opus-4", "opus-4": .anthropic(.opus4)

Sources/Tachikoma/Utilities/UsageTracking.swift

@@ -528,7 +528,7 @@ public struct ModelCostCalculator: Sendable {
         // OpenAI Pricing (as of 2025)
         case let .openai(openaiModel):
             switch openaiModel {
-            case .o4Mini: (1.50, 6.00)
+            case .gpt55: (5.00, 20.00) // GPT-5.5 pricing estimate
             case .gpt52: (5.00, 20.00) // GPT-5.2 pricing estimate
             case .gpt51: (5.00, 20.00) // GPT-5.1 pricing estimate
             case .gpt5: (5.00, 20.00) // GPT-5 pricing estimate
@@ -539,18 +539,12 @@ public struct ModelCostCalculator: Sendable {
             case .gpt5ThinkingMini: (4.00, 16.00)
             case .gpt5ThinkingNano: (1.50, 6.00)
             case .gpt5ChatLatest: (2.50, 10.00)
-            case .gpt41: (2.50, 10.00)
-            case .gpt41Mini: (0.15, 0.60)
-            case .gpt4o: (2.50, 10.00)
-            case .gpt4oMini: (0.15, 0.60)
-            case .gpt4oRealtime: (5.00, 20.00) // Realtime API pricing estimate
-            case .gpt4Turbo: (10.00, 30.00)
-            case .gpt35Turbo: (0.50, 1.50)
             case .custom: (2.50, 10.00) // Default estimate
             }
         // Anthropic Pricing (as of 2025)
         case let .anthropic(anthropicModel):
             switch anthropicModel {
+            case .opus47: (15.00, 75.00)
             case .opus45: (5.00, 25.00)
             case .opus4, .opus4Thinking: (15.00, 75.00)
             case .sonnet4, .sonnet4Thinking: (3.00, 15.00)

Sources/TachikomaAudio/Realtime/Configuration/RealtimeConfiguration.swift

@@ -128,7 +128,7 @@ public struct InputAudioTranscription: Sendable, Codable {
 /// Session configuration with all options
 @available(macOS 13.0, iOS 16.0, watchOS 9.0, tvOS 16.0, *)
 public struct SessionConfiguration: Sendable, Codable {
-    /// Model to use (e.g., "gpt-4o-realtime-preview")
+    /// Model to use (e.g., "gpt-realtime")
     public var model: String
 
     /// Voice for audio responses
@@ -213,7 +213,7 @@ public struct SessionConfiguration: Sendable, Codable {
     }
 
     public init(
-        model: String = "gpt-4o-realtime-preview",
+        model: String = "gpt-realtime",
         voice: RealtimeVoice = .alloy,
         instructions: String? = nil,
         inputAudioFormat: RealtimeAudioFormat = .pcm16,
@@ -242,7 +242,7 @@ public struct SessionConfiguration: Sendable, Codable {
 
     /// Create a default configuration for voice conversations
     public static func voiceConversation(
-        model: String = "gpt-4o-realtime-preview",
+        model: String = "gpt-realtime",
         voice: RealtimeVoice = .alloy,
     )
         -> SessionConfiguration
@@ -258,7 +258,7 @@ public struct SessionConfiguration: Sendable, Codable {
 
     /// Create a configuration for text-only interactions
     public static func textOnly(
-        model: String = "gpt-4o-realtime-preview",
+        model: String = "gpt-realtime",
     )
         -> SessionConfiguration
     {
@@ -273,7 +273,7 @@ public struct SessionConfiguration: Sendable, Codable {
 
     /// Create a configuration with tools
     public static func withTools(
-        model: String = "gpt-4o-realtime-preview",
+        model: String = "gpt-realtime",
         voice: RealtimeVoice = .alloy,
         tools: [RealtimeTool],
     )

Sources/TachikomaAudio/Realtime/RealtimeConversation.swift

@@ -150,7 +150,7 @@ public final class RealtimeConversation: ObservableObject {
 
         // Create session with configuration
         let sessionConfig = SessionConfiguration(
-            model: "gpt-4o-realtime-preview",
+            model: "gpt-realtime",
             voice: .alloy,
             instructions: nil,
             tools: nil,
@@ -167,7 +167,7 @@ public final class RealtimeConversation: ObservableObject {
 
     /// Start the conversation
     public func start(
-        model: LanguageModel.OpenAI = .gpt4oRealtime,
+        model: LanguageModel.OpenAI = .custom("gpt-realtime"),
         voice: RealtimeVoice = .alloy,
         instructions: String? = nil,
         tools: [RealtimeTool]? = nil,
@@ -495,7 +495,7 @@ public final class RealtimeConversation: ObservableObject {
 
 @available(macOS 13.0, iOS 16.0, watchOS 9.0, tvOS 16.0, *)
 public func startRealtimeConversation(
-    model: LanguageModel.OpenAI = .gpt4oRealtime,
+    model: LanguageModel.OpenAI = .custom("gpt-realtime"),
     voice: RealtimeVoice = .alloy,
     instructions: String? = nil,
     tools: [AgentTool]? = nil,
@@ -548,7 +548,7 @@ public final class RealtimeConversation {
     }
 
     public func start(
-        model _: LanguageModel.OpenAI = .gpt4oRealtime,
+        model _: LanguageModel.OpenAI = .custom("gpt-realtime"),
         voice _: RealtimeVoice = .alloy,
         instructions _: String? = nil,
         tools _: [RealtimeTool]? = nil,
@@ -560,7 +560,7 @@ public final class RealtimeConversation {
 // swiftformat:disable wrapMultilineStatementBraces wrapReturnType indent
 @available(macOS 13.0, iOS 16.0, watchOS 9.0, tvOS 16.0, *)
 public func startRealtimeConversation(
-    model _: LanguageModel.OpenAI = .gpt4oRealtime,
+    model _: LanguageModel.OpenAI = .custom("gpt-realtime"),
     voice _: RealtimeVoice = .alloy,
     instructions _: String? = nil,
     tools _: [AgentTool]? = nil,

Sources/TachikomaAudio/Realtime/RealtimeEvents.swift

@@ -188,7 +188,7 @@ public struct RealtimeSessionConfig: Codable, Sendable {
     }
 
     public init(
-        model: String = "gpt-4o-realtime-preview",
+        model: String = "gpt-realtime",
         voice: RealtimeVoice = .alloy,
         instructions: String? = nil,
         tools: [RealtimeTool]? = nil,

Sources/TachikomaAudio/Types/AudioModels.swift

@@ -16,20 +16,16 @@ public enum TranscriptionModel: Sendable, CustomStringConvertible {
 
     public enum OpenAI: String, CaseIterable, Sendable {
         case whisper1 = "whisper-1"
-        case gpt4oTranscribe = "gpt-4o-transcribe"
-        case gpt4oMiniTranscribe = "gpt-4o-mini-transcribe"
 
         public var supportsTimestamps: Bool {
             switch self {
             case .whisper1: true
-            case .gpt4oTranscribe, .gpt4oMiniTranscribe: false
             }
         }
 
         public var supportsLanguageDetection: Bool {
             switch self {
             case .whisper1: true
-            case .gpt4oTranscribe, .gpt4oMiniTranscribe: false
             }
         }
     }
@@ -155,12 +151,10 @@ public enum SpeechModel: Sendable, CustomStringConvertible {
     public enum OpenAI: String, CaseIterable, Sendable {
         case tts1 = "tts-1"
         case tts1HD = "tts-1-hd"
-        case gpt4oMiniTTS = "gpt-4o-mini-tts"
 
         public var supportsVoiceInstructions: Bool {
             switch self {
             case .tts1, .tts1HD: false
-            case .gpt4oMiniTTS: true
             }
         }
 
@@ -226,5 +220,5 @@ public enum SpeechModel: Sendable, CustomStringConvertible {
     public static let `default`: SpeechModel = .openai(.tts1)
     public static let highQuality: SpeechModel = .openai(.tts1HD)
     public static let fast: SpeechModel = .openai(.tts1)
-    public static let expressive: SpeechModel = .openai(.gpt4oMiniTTS)
+    public static let expressive: SpeechModel = .openai(.tts1HD)
 }

Sources/TachikomaMCP/README.md

@@ -12,7 +12,7 @@ import Tachikoma
 
 // Use Tachikoma normally without MCP
 let result = try await generateText(
-    model: .openai(.gpt4o),
+    model: .openai(.gpt55),
     messages: messages,
     tools: staticTools
 )
@@ -26,7 +26,7 @@ import TachikomaMCP
 // Now you can use MCP tools alongside static tools
 let mcpTools = try await MCPToolDiscovery.withFilesystem()
 let result = try await generateText(
-    model: .openai(.gpt4o),
+    model: .openai(.gpt55),
     messages: messages,
     tools: staticTools + mcpTools
 )

Tests/TachikomaTests/Core/AsyncSequenceTests.swift

@@ -17,7 +17,7 @@ struct AsyncSequenceTests {
 
         let result = StreamTextResult(
             stream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 
@@ -86,7 +86,7 @@ struct AsyncSequenceTests {
 
         let result = StreamObjectResult(
             objectStream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
             schema: TestData.self,
         )
@@ -117,7 +117,7 @@ struct AsyncSequenceTests {
 
         let result = StreamTextResult(
             stream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 
@@ -149,7 +149,7 @@ struct AsyncSequenceTests {
 
         let result = StreamTextResult(
             stream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 
@@ -183,7 +183,7 @@ struct AsyncSequenceTests {
 
         let result = StreamTextResult(
             stream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 
@@ -218,7 +218,7 @@ struct AsyncSequenceTests {
 
         let result = StreamTextResult(
             stream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 
@@ -262,7 +262,7 @@ struct AsyncSequenceTests {
 
         let result = StreamObjectResult(
             objectStream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
             schema: TestItem.self,
         )

Tests/TachikomaTests/Core/ConfigurationArchitectureTests.swift

@@ -55,7 +55,7 @@ struct ConfigurationArchitectureTests {
         // Example 1: Zero configuration
         _ = {
             Task {
-                _ = try await generate("What is 2+2?", using: .openai(.gpt4o))
+                _ = try await generate("What is 2+2?", using: .openai(.gpt55))
             }
         }
 
@@ -66,7 +66,7 @@ struct ConfigurationArchitectureTests {
             )
 
             Task {
-                _ = try await generate("Hello", using: .openai(.gpt4o))
+                _ = try await generate("Hello", using: .openai(.gpt55))
             }
         }
 
@@ -79,7 +79,7 @@ struct ConfigurationArchitectureTests {
 
                 _ = try await generate(
                     "Test prompt",
-                    using: .openai(.gpt4o),
+                    using: .openai(.gpt55),
                     configuration: testConfig,
                 )
             }

Tests/TachikomaTests/Core/GenerationTests.swift

@@ -10,7 +10,7 @@ struct GenerationTests {
         try await TestHelpers.withTestConfiguration(apiKeys: ["openai": "test-key"]) { config in
             let result = try await generate(
                 "What is 2+2?",
-                using: .openai(.gpt4o),
+                using: .openai(.gpt55),
                 maxTokens: 100,
                 configuration: config,
             )
@@ -18,7 +18,7 @@ struct GenerationTests {
             self.assertOpenAIResult(
                 result,
                 prompt: "What is 2+2?",
-                modelId: "gpt-4o",
+                modelId: "gpt-5.5",
                 configuration: config,
             )
         }
@@ -56,7 +56,7 @@ struct GenerationTests {
         try await TestHelpers.withTestConfiguration(apiKeys: ["openai": "test-key"]) { config in
             let result = try await generate(
                 "Tell me a joke",
-                using: .openai(.gpt4oMini),
+                using: .openai(.gpt5Mini),
                 system: "You are a comedian",
                 temperature: 0.8,
                 configuration: config,
@@ -77,7 +77,7 @@ struct GenerationTests {
         try await TestHelpers.withTestConfiguration(apiKeys: ["openai": "test-key"]) { config in
             let stream = try await stream(
                 "Count to 5",
-                using: .openai(.gpt4o),
+                using: .openai(.gpt55),
                 maxTokens: 50,
                 configuration: config,
             )
@@ -141,7 +141,7 @@ struct GenerationTests {
             let result = try await analyze(
                 image: .base64(testImageBase64),
                 prompt: "What do you see?",
-                using: .openai(.gpt4o),
+                using: .openai(.gpt55),
                 configuration: config,
             )
 
@@ -156,12 +156,12 @@ struct GenerationTests {
     @Test
     func `Analyze Function - Non-Vision Model Error`() async {
         _ = await TestHelpers.withTestConfiguration(apiKeys: ["openai": "test-key"]) { config in
-            // GPT-4.1 doesn't support vision
+            // Custom OpenAI models default to text-only capabilities
             await #expect(throws: TachikomaError.self) {
                 try await analyze(
                     image: .base64("test-image"),
                     prompt: "Describe this",
-                    using: .openai(.gpt41),
+                    using: .openai(.custom("text-only-openai")),
                     configuration: config,
                 )
             }
@@ -179,7 +179,7 @@ struct GenerationTests {
                 configuration: config,
             )
 
-            // Should default to GPT-4o for vision tasks
+            // Should default to GPT-5.5 for vision tasks
             self.assertOpenAIResult(
                 result,
                 prompt: "Analyze this image",
@@ -194,7 +194,7 @@ struct GenerationTests {
     func `Generate Function - Missing API Key`() async {
         _ = await TestHelpers.withEmptyTestConfiguration { config in
             await #expect(throws: TachikomaError.self) {
-                try await generate("Test", using: .openai(.gpt4o), configuration: config)
+                try await generate("Test", using: .openai(.gpt55), configuration: config)
             }
         }
     }
@@ -208,7 +208,7 @@ struct GenerationTests {
             // With mock provider (test-key), this should work even with invalid URL
             // Real implementations would fail with network error
             do {
-                let result = try await generate("Test", using: .openai(.gpt4o), configuration: config)
+                let result = try await generate("Test", using: .openai(.gpt55), configuration: config)
                 #expect(!result.isEmpty)
             } catch {
                 // If using real provider, invalid URL will cause network error
@@ -226,7 +226,7 @@ struct GenerationTests {
             // Test generation without tools
             let result = try await generate(
                 "Hello",
-                using: .openai(.gpt4o),
+                using: .openai(.gpt55),
                 configuration: config,
             )
 

Tests/TachikomaTests/Core/LanguageModelCoverageTests.swift

@@ -68,8 +68,8 @@ struct LanguageModelCoverageTests {
     @Test
     func `LanguageModel top level switches`() {
         let baseModels: [LanguageModel] = [
-            .openai(.gpt51),
-            .anthropic(.opus45),
+            .openai(.gpt55),
+            .anthropic(.opus47),
             .google(.gemini25Flash),
             .mistral(.large2),
             .groq(.mixtral8x7b),

Tests/TachikomaTests/Core/MinimalModernAPITests.swift

@@ -9,8 +9,8 @@ struct MinimalModernAPITests {
     @Test
     func `Model enum construction`() {
         // Test that model enums can be constructed
-        let openaiModel = Model.openai(.gpt4o)
-        let anthropicModel = Model.anthropic(.opus45)
+        let openaiModel = Model.openai(.gpt55)
+        let anthropicModel = Model.anthropic(.opus47)
         _ = Model.grok(.grok4)
         _ = Model.ollama(.llama33)
 
@@ -35,10 +35,10 @@ struct MinimalModernAPITests {
         let defaultModel = Model.default
         // Should compile without errors
         switch defaultModel {
-        case .anthropic(.opus45):
+        case .anthropic(.opus47):
             break // Expected default
         default:
-            Issue.record("Expected default to be Anthropic Opus 4.5")
+            Issue.record("Expected default to be Anthropic Opus 4.7")
         }
     }
 

Tests/TachikomaTests/Core/ModelCapabilitiesTests.swift

@@ -7,6 +7,7 @@ enum ModelCapabilitiesTests {
         @Test
         func `GPT-5 models exclude temperature and topP`() {
             let models: [LanguageModel] = [
+                .openai(.gpt55),
                 .openai(.gpt52),
                 .openai(.gpt51),
                 .openai(.gpt5),
@@ -35,26 +36,9 @@ enum ModelCapabilitiesTests {
         }
 
         @Test
-        func `Reasoning models have forced temperature`() {
-            let model = LanguageModel.openai(.o4Mini)
-            let capabilities = ModelCapabilityRegistry.shared.capabilities(for: model)
-
-            #expect(!capabilities.supportsTemperature)
-            #expect(!capabilities.supportsTopP)
-            #expect(capabilities.forcedTemperature == 1.0)
-            #expect(capabilities.excludedParameters.contains("temperature"))
-            #expect(capabilities.excludedParameters.contains("topP"))
-            #expect(capabilities.supportedProviderOptions.supportsReasoningEffort)
-            #expect(capabilities.supportedProviderOptions.supportsPreviousResponseId)
-        }
-
-        @Test
-        func `GPT-4 models support standard parameters`() {
+        func `Custom OpenAI models support standard parameters`() {
             let models: [LanguageModel] = [
-                .openai(.gpt4o),
-                .openai(.gpt4oMini),
-                .openai(.gpt41),
-                .openai(.gpt4Turbo),
+                .openai(.custom("custom-openai")),
             ]
 
             for model in models {
@@ -65,15 +49,13 @@ enum ModelCapabilitiesTests {
                 #expect(capabilities.supportsMaxTokens)
                 #expect(capabilities.supportsFrequencyPenalty)
                 #expect(capabilities.supportsPresencePenalty)
-                #expect(capabilities.supportedProviderOptions.supportsParallelToolCalls)
-                #expect(capabilities.supportedProviderOptions.supportsResponseFormat)
-                #expect(capabilities.supportedProviderOptions.supportsLogprobs)
             }
         }
 
         @Test
         func `Claude models support thinking`() {
             let models: [LanguageModel] = [
+                .anthropic(.opus47),
                 .anthropic(.opus4),
                 .anthropic(.sonnet4),
                 .anthropic(.sonnet45),
@@ -151,7 +133,7 @@ enum ModelCapabilitiesTests {
 
     struct SettingsValidationTests {
         @Test
-        func `Validate settings for GPT-5.1`() {
+        func `Validate settings for GPT-5.5`() {
             let settings = GenerationSettings(
                 maxTokens: 1000,
                 temperature: 0.7,
@@ -166,7 +148,7 @@ enum ModelCapabilitiesTests {
                 ),
             )
 
-            let validated = settings.validated(for: .openai(.gpt51))
+            let validated = settings.validated(for: .openai(.gpt55))
 
             #expect(validated.maxTokens == 1000)
             #expect(validated.temperature == nil) // Excluded
@@ -178,7 +160,7 @@ enum ModelCapabilitiesTests {
         }
 
         @Test
-        func `Validate settings for O3 with forced temperature`() {
+        func `Validate settings for GPT-5 strips unsupported options`() {
             let settings = GenerationSettings(
                 temperature: 0.5,
                 topP: 0.8,
@@ -190,16 +172,16 @@ enum ModelCapabilitiesTests {
                 ),
             )
 
-            let validated = settings.validated(for: LanguageModel.openai(.o4Mini))
+            let validated = settings.validated(for: LanguageModel.openai(.gpt55))
 
-            #expect(validated.temperature == 1.0) // Forced to 1.0
+            #expect(validated.temperature == nil) // Excluded
             #expect(validated.topP == nil) // Excluded
-            #expect(validated.providerOptions.openai?.reasoningEffort == .high) // Kept
-            #expect(validated.providerOptions.openai?.verbosity == nil) // Removed
+            #expect(validated.providerOptions.openai?.reasoningEffort == nil) // Removed
+            #expect(validated.providerOptions.openai?.verbosity == .medium) // Kept
         }
 
         @Test
-        func `Validate settings for GPT-4`() {
+        func `Validate settings for custom OpenAI model`() {
             let settings = GenerationSettings(
                 maxTokens: 2000,
                 temperature: 0.8,
@@ -216,17 +198,13 @@ enum ModelCapabilitiesTests {
                 ),
             )
 
-            let validated = settings.validated(for: .openai(.gpt4o))
+            let validated = settings.validated(for: .openai(.custom("custom-openai")))
 
             #expect(validated.maxTokens == 2000)
             #expect(validated.temperature == 0.8)
             #expect(validated.topP == 0.95)
             #expect(validated.frequencyPenalty == 0.2)
             #expect(validated.presencePenalty == 0.1)
-            #expect(validated.providerOptions.openai?.parallelToolCalls == true)
-            #expect(validated.providerOptions.openai?.responseFormat == .json)
-            #expect(validated.providerOptions.openai?.logprobs == true)
-            #expect(validated.providerOptions.openai?.topLogprobs == 3)
         }
 
         @Test
@@ -308,7 +286,7 @@ enum ModelCapabilitiesTests {
         func `Concurrent capability access`() async {
             let models: [LanguageModel] = [
                 .openai(.gpt51),
-                .openai(.gpt4o),
+                .openai(.gpt55),
                 .anthropic(.opus4),
                 .google(.gemini25Flash),
             ]

Tests/TachikomaTests/Core/ModelParsingTests.swift

@@ -9,9 +9,9 @@ struct ModelParsingTests {
     }
 
     @Test
-    func `parse GPT-5.1 base model`() {
-        let parsed = LanguageModel.parse(from: "gpt-5.1")
-        #expect(parsed == .openai(.gpt51))
+    func `parse GPT-5.5 base model`() {
+        let parsed = LanguageModel.parse(from: "gpt-5.5")
+        #expect(parsed == .openai(.gpt55))
     }
 
     @Test
@@ -26,6 +26,12 @@ struct ModelParsingTests {
         #expect(parsed == .openai(.gpt5Nano))
     }
 
+    @Test
+    func `parse Claude Opus 4.7 model id`() {
+        let parsed = LanguageModel.parse(from: "claude-opus-4-7")
+        #expect(parsed == .anthropic(.opus47))
+    }
+
     @Test
     func `parse Claude Sonnet 4.5 snapshot id`() {
         let parsed = LanguageModel.parse(from: "claude-sonnet-4-5-20250929")
@@ -35,7 +41,7 @@ struct ModelParsingTests {
     @Test
     func `parse shorthand Claude alias`() {
         let parsed = LanguageModel.parse(from: "claude")
-        #expect(parsed == .anthropic(.sonnet45))
+        #expect(parsed == .anthropic(.opus47))
     }
 
     @Test
@@ -49,4 +55,24 @@ struct ModelParsingTests {
         let parsed = LanguageModel.parse(from: "gemini")
         #expect(parsed == .google(.gemini3Flash))
     }
+
+    @Test
+    func `ModelSelector rejects legacy OpenAI before Ollama fallback`() throws {
+        if #available(macOS 13.0, iOS 16.0, tvOS 16.0, watchOS 9.0, *) {
+            for model in ["gpt-4o", "gpt-4.1", "gpt-3.5-turbo", "o4-mini", "o3-mini"] {
+                #expect(throws: ModelValidationError.self) {
+                    _ = try ModelSelector.parseModel(model)
+                }
+            }
+        }
+    }
+
+    @Test
+    func `ModelSelector rejects Claude 3 before Ollama fallback`() throws {
+        if #available(macOS 13.0, iOS 16.0, tvOS 16.0, watchOS 9.0, *) {
+            #expect(throws: ModelValidationError.self) {
+                _ = try ModelSelector.parseModel("claude-3-sonnet")
+            }
+        }
+    }
 }

Tests/TachikomaTests/Core/StopConditionsIntegrationTests.swift

@@ -327,7 +327,7 @@ private struct MockTextProvider: ModelProvider {
 
 private struct MockOpenAIProvider: ModelProvider {
     var modelId: String {
-        "gpt-4"
+        "gpt-5.5"
     }
 
     var baseURL: String? {

Tests/TachikomaTests/Core/StreamObjectTests.swift

@@ -89,12 +89,12 @@ struct StreamObjectTests {
 
         let result = StreamObjectResult(
             objectStream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
             schema: TestPerson.self,
         )
 
-        #expect(result.model == .openai(.gpt4o))
+        #expect(result.model == .openai(.gpt55))
         #expect(result.schema == TestPerson.self)
     }
 
@@ -157,7 +157,7 @@ struct StreamObjectTests {
 
         let result = StreamObjectResult(
             objectStream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
             schema: TestPerson.self,
         )
@@ -193,7 +193,7 @@ struct StreamObjectTests {
 
         let result = StreamObjectResult(
             objectStream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
             schema: TestPerson.self,
         )
@@ -221,7 +221,7 @@ struct StreamObjectTests {
 
         let result = StreamObjectResult(
             objectStream: testStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
             schema: TestPerson.self,
         )

Tests/TachikomaTests/Core/StreamTransformTests.swift

@@ -231,7 +231,7 @@ struct StreamTransformTests {
 
         let result = StreamTextResult(
             stream: stream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 
@@ -264,7 +264,7 @@ struct StreamTransformTests {
 
         let result = StreamTextResult(
             stream: stream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 
@@ -291,7 +291,7 @@ struct StreamTransformTests {
 
         let result = StreamTextResult(
             stream: stream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 

Tests/TachikomaTests/Core/TachikomaConfigurationTests.swift

@@ -367,35 +367,35 @@ struct TachikomaConfigurationTests {
                 Task {
                     // Uses default (.current)
                     _ = try await generateText(
-                        model: .openai(.gpt4o),
+                        model: .openai(.gpt55),
                         messages: [.user("Test")],
                     )
 
                     // Uses explicit
                     _ = try await generateText(
-                        model: .openai(.gpt4o),
+                        model: .openai(.gpt55),
                         messages: [.user("Test")],
                         configuration: explicitConfig,
                     )
 
                     // Stream functions
                     _ = try await streamText(
-                        model: .openai(.gpt4o),
+                        model: .openai(.gpt55),
                         messages: [.user("Test")],
                     )
 
                     _ = try await streamText(
-                        model: .openai(.gpt4o),
+                        model: .openai(.gpt55),
                         messages: [.user("Test")],
                         configuration: explicitConfig,
                     )
 
                     // Convenience functions
-                    _ = try await generate("Test", using: .openai(.gpt4o))
-                    _ = try await generate("Test", using: .openai(.gpt4o), configuration: explicitConfig)
+                    _ = try await generate("Test", using: .openai(.gpt55))
+                    _ = try await generate("Test", using: .openai(.gpt55), configuration: explicitConfig)
 
-                    _ = try await stream("Test", using: .openai(.gpt4o))
-                    _ = try await stream("Test", using: .openai(.gpt4o), configuration: explicitConfig)
+                    _ = try await stream("Test", using: .openai(.gpt55))
+                    _ = try await stream("Test", using: .openai(.gpt55), configuration: explicitConfig)
                 }
             }
 
@@ -559,9 +559,9 @@ struct TachikomaConfigurationTests {
                 return DummyProvider()
             }
 
-            let provider = try config.makeProvider(for: .openai(.gpt4o))
+            let provider = try config.makeProvider(for: .openai(.gpt55))
             #expect(provider is DummyProvider)
-            #expect(capturedModel == .openai(.gpt4o))
+            #expect(capturedModel == .openai(.gpt55))
         }
 
         @Test
@@ -569,8 +569,8 @@ struct TachikomaConfigurationTests {
             let config = TachikomaConfiguration(loadFromEnvironment: false)
             config.setAPIKey("mock-key", for: .openai)
 
-            let provider = try config.makeProvider(for: .openai(.gpt4o))
-            #expect(provider is OpenAIProvider)
+            let provider = try config.makeProvider(for: .openai(.gpt55))
+            #expect(provider is OpenAIResponsesProvider)
         }
     }
 }

Tests/TachikomaTests/HarmonyFeatures/ResponseCacheTests.swift

@@ -317,7 +317,7 @@ struct ResponseCacheTests {
 
         // Create a mock provider
         let mockProvider = ResponseCacheMockProvider(
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             response: ProviderResponse(text: "Cached response", usage: nil, finishReason: .stop),
         )
 
@@ -336,7 +336,7 @@ struct ResponseCacheTests {
         // Use a simple counter that can be modified in the closure
         let callCount = Box(value: 0)
         var mockProvider = ResponseCacheMockProvider(
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             response: ProviderResponse(text: "Response", usage: nil, finishReason: .stop),
         )
         mockProvider.onGenerateText = { _ in
@@ -368,7 +368,7 @@ struct ResponseCacheTests {
 
         let callCount = Box(value: 0)
         var mockProvider = ResponseCacheMockProvider(
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             response: ProviderResponse(text: "Test", usage: nil, finishReason: .stop),
         )
         mockProvider.onStreamText = { _ in

Tests/TachikomaTests/IntegrationTests.swift

@@ -29,7 +29,7 @@ struct IntegrationTests {
 
         let streamResult = StreamTextResult(
             stream: textStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 

Tests/TachikomaTests/Providers/AnthropicInterleavedDefaultsTests.swift

@@ -25,7 +25,7 @@ struct AnthropicInterleavedDefaultsTests {
     @Test
     func `Provider request includes beta header and thinking payload`() throws {
         let config = TachikomaConfiguration(apiKeys: ["anthropic": "test-key"])
-        let provider = try AnthropicProvider(model: .opus45, configuration: config)
+        let provider = try AnthropicProvider(model: .opus47, configuration: config)
 
         let settings = GenerationSettings(
             maxTokens: 64,
@@ -49,7 +49,7 @@ struct AnthropicInterleavedDefaultsTests {
 
         let body = try #require(urlRequest.httpBody)
         let json = try #require(try JSONSerialization.jsonObject(with: body) as? [String: Any])
-        #expect(json["model"] as? String == "claude-opus-4-5")
+        #expect(json["model"] as? String == "claude-opus-4-7")
         #expect(json["stream"] as? Bool == true)
 
         let thinking = try #require(json["thinking"] as? [String: Any])

Tests/TachikomaTests/Providers/Compatible/AzureOpenAIProviderTests.swift

@@ -20,7 +20,7 @@ private final class AzureTestURLProtocol: URLProtocol {
     static let responseBody: Data = """
     {
       "id": "chatcmpl-azure",
-      "model": "gpt-4o",
+      "model": "gpt-5.5",
       "choices": [
         {
           "index": 0,
@@ -85,7 +85,7 @@ struct AzureOpenAIProviderTests {
         await AzureTestURLProtocol.reset()
 
         let provider = try AzureOpenAIProvider(
-            deploymentId: "gpt-4o",
+            deploymentId: "gpt-5.5",
             resource: "my-aoai",
             apiVersion: "2025-04-01-preview",
             endpoint: nil,
@@ -99,7 +99,7 @@ struct AzureOpenAIProviderTests {
         #expect(response.text == "hello azure")
 
         let sentRequest = await AzureTestURLProtocol.fetchLastRequest()
-        #expect(sentRequest?.url?.path == "/openai/deployments/gpt-4o/chat/completions")
+        #expect(sentRequest?.url?.path == "/openai/deployments/gpt-5.5/chat/completions")
 
         if let components = sentRequest?.url.flatMap({ URLComponents(url: $0, resolvingAgainstBaseURL: false) }) {
             let apiVersion = components.queryItems?.first { $0.name == "api-version" }?.value
@@ -122,7 +122,7 @@ struct AzureOpenAIProviderTests {
         await AzureTestURLProtocol.reset()
 
         let provider = try AzureOpenAIProvider(
-            deploymentId: "gpt-4o-mini",
+            deploymentId: "gpt-5-mini",
             resource: nil,
             apiVersion: "2025-04-01-preview",
             endpoint: nil,

Tests/TachikomaTests/Providers/Integration/ProviderIntegrationTests.swift

@@ -49,7 +49,7 @@ struct ProviderIntegrationTests {
 
     @Test(.enabled(if: Self.hasOpenAIKey))
     func `OpenAI Provider - Real API Call`() async throws {
-        let model = Model.openai(.gpt4oMini)
+        let model = Model.openai(.gpt5Mini)
         let config = TachikomaConfiguration()
         do {
             _ = try ProviderFactory.createProvider(for: model, configuration: config)
@@ -72,7 +72,7 @@ struct ProviderIntegrationTests {
 
     @Test(.enabled(if: Self.hasOpenAIKey))
     func `OpenAI Provider - Tool Calling`() async throws {
-        let model = Model.openai(.gpt4oMini)
+        let model = Model.openai(.gpt5Mini)
         let config = TachikomaConfiguration()
 
         do {
@@ -117,7 +117,7 @@ struct ProviderIntegrationTests {
 
     @Test(.enabled(if: Self.hasOpenAIKey))
     func `OpenAI Provider - Streaming`() async throws {
-        let model = Model.openai(.gpt4oMini)
+        let model = Model.openai(.gpt5Mini)
         let config = TachikomaConfiguration()
 
         do {
@@ -323,7 +323,7 @@ struct ProviderIntegrationTests {
 
     @Test(.enabled(if: Self.hasOpenAIKey))
     func `Multi-Modal Provider - Vision Support`() async throws {
-        let model = Model.openai(.gpt4o)
+        let model = Model.openai(.gpt55)
         let config = TachikomaConfiguration()
         let provider = try ProviderFactory.createProvider(for: model, configuration: config)
 

Tests/TachikomaTests/Providers/Integration/ProviderSystemTests.swift

@@ -9,10 +9,10 @@ struct ProviderSystemTests {
     @Test
     func `Provider Factory - OpenAI Provider Creation`() async throws {
         try await TestHelpers.withTestConfiguration(apiKeys: ["openai": "test-key"]) { config in
-            let model = Model.openai(.gpt4o)
+            let model = Model.openai(.gpt55)
             let provider = try ProviderFactory.createProvider(for: model, configuration: config)
 
-            #expect(provider.modelId == "gpt-4o")
+            #expect(provider.modelId == "gpt-5.5")
             #expect(provider.capabilities.supportsVision == true)
             #expect(provider.capabilities.supportsTools == true)
             #expect(provider.capabilities.supportsStreaming == true)
@@ -89,7 +89,7 @@ struct ProviderSystemTests {
             }
 
             #expect(throws: TachikomaError.self) {
-                try OpenAIProvider(model: .gpt4o, configuration: config)
+                try OpenAIProvider(model: .gpt55, configuration: config)
             }
 
             #expect(throws: TachikomaError.self) {
@@ -102,9 +102,9 @@ struct ProviderSystemTests {
 
     @Test
     func `Model Capabilities - Vision Support`() {
-        #expect(Model.openai(.gpt4o).supportsVision == true)
-        #expect(Model.openai(.gpt4oMini).supportsVision == true)
-        #expect(Model.openai(.gpt41).supportsVision == false)
+        #expect(Model.openai(.gpt55).supportsVision == true)
+        #expect(Model.openai(.gpt5Mini).supportsVision == true)
+        #expect(Model.openai(.custom("text-only-openai")).supportsVision == false)
 
         #expect(Model.anthropic(.opus4).supportsVision == true)
         #expect(Model.anthropic(.sonnet4).supportsVision == true)
@@ -121,8 +121,8 @@ struct ProviderSystemTests {
 
     @Test
     func `Model Capabilities - Tool Support`() {
-        #expect(Model.openai(.gpt4o).supportsTools == true)
-        #expect(Model.openai(.gpt41).supportsTools == true)
+        #expect(Model.openai(.gpt55).supportsTools == true)
+        #expect(Model.openai(.gpt55).supportsTools == true)
 
         #expect(Model.anthropic(.opus4).supportsTools == true)
         #expect(Model.anthropic(.sonnet4).supportsTools == true)
@@ -136,7 +136,7 @@ struct ProviderSystemTests {
 
     @Test
     func `Model Capabilities - Streaming Support`() {
-        #expect(Model.openai(.gpt4o).supportsStreaming == true)
+        #expect(Model.openai(.gpt55).supportsStreaming == true)
         #expect(Model.anthropic(.opus4).supportsStreaming == true)
         #expect(Model.grok(.grok4).supportsStreaming == true)
         #expect(Model.ollama(.llama33).supportsStreaming == true)

Tests/TachikomaTests/Providers/OpenAIResponsesProviderTests.swift

@@ -23,6 +23,7 @@ struct OpenAIResponsesProviderTests {
         let config = self.openAIConfig()
 
         let gpt5Models: [LanguageModel.OpenAI] = [
+            .gpt55,
             .gpt52,
             .gpt51,
             .gpt5,
@@ -44,8 +45,8 @@ struct OpenAIResponsesProviderTests {
     }
 
     @Test
-    func `GPT-5.1 text.verbosity parameter is set correctly`() throws {
-        // Test that the text.verbosity parameter is properly configured for GPT-5.1
+    func `GPT-5.5 text.verbosity parameter is set correctly`() throws {
+        // Test that the text.verbosity parameter is properly configured for GPT-5.5
         let config = self.openAIConfig()
 
         // Skip if no API key
@@ -54,7 +55,7 @@ struct OpenAIResponsesProviderTests {
         }
 
         let provider = try OpenAIResponsesProvider(
-            model: .gpt51,
+            model: .gpt55,
             configuration: config,
         )
 
@@ -69,18 +70,18 @@ struct OpenAIResponsesProviderTests {
 
         // We can't directly test the internal request building without making it public
         // But we can verify the provider is configured correctly
-        #expect(provider.modelId == "gpt-5.1")
+        #expect(provider.modelId == "gpt-5.5")
         #expect(provider.capabilities.supportsTools == true)
         #expect(provider.capabilities.supportsVision == true)
     }
 
     @Test
-    func `Reasoning models use Responses API`() throws {
-        // Test that reasoning-oriented models also use the OpenAIResponsesProvider
+    func `GPT-5 models use Responses API`() throws {
+        // Test that GPT-5 models use the OpenAIResponsesProvider
         let config = self.openAIConfig()
 
-        let reasoningModels: [LanguageModel.OpenAI] = [
-            .o4Mini,
+        let responsesModels: [LanguageModel.OpenAI] = [
+            .gpt55,
             .gpt52,
             .gpt51,
             .gpt5,
@@ -88,7 +89,7 @@ struct OpenAIResponsesProviderTests {
             .gpt5Thinking,
         ]
 
-        for model in reasoningModels {
+        for model in responsesModels {
             let provider = try ProviderFactory.createProvider(
                 for: .openai(model),
                 configuration: config,
@@ -102,11 +103,10 @@ struct OpenAIResponsesProviderTests {
     }
 
     @Test
-    func `Legacy models use standard OpenAI provider`() throws {
-        // Test that non-GPT-5/reasoning models use the standard OpenAIProvider
+    func `Custom OpenAI models use standard OpenAI provider`() throws {
         let config = self.openAIConfig()
 
-        let legacyModels: [LanguageModel.OpenAI] = [.gpt4o, .gpt4oMini, .gpt41]
+        let legacyModels: [LanguageModel.OpenAI] = [.custom("custom-openai")]
 
         for model in legacyModels {
             let provider = try ProviderFactory.createProvider(
@@ -116,7 +116,7 @@ struct OpenAIResponsesProviderTests {
 
             #expect(
                 provider is OpenAIProvider,
-                "Legacy model \(model) should use OpenAIProvider",
+                "Custom model \(model) should use OpenAIProvider",
             )
         }
     }
@@ -472,12 +472,12 @@ struct OpenAIResponsesProviderTests {
             #expect(request.url?.path == "/v1/responses")
             let payload = Self.responsesStreamPayload(chunks: [
                 Self.streamChunkJSON(content: "Hello", finishReason: nil),
-                Self.streamChunkJSON(content: "Hello world", finishReason: nil),
+                Self.streamChunkJSON(content: " world", finishReason: nil),
                 Self.streamChunkJSON(content: nil, finishReason: "stop"),
             ])
             return NetworkMocking.streamResponse(for: request, data: payload)
         } operation: { session in
-            let provider = try OpenAIResponsesProvider(model: .o4Mini, configuration: config, session: session)
+            let provider = try OpenAIResponsesProvider(model: .gpt55, configuration: config, session: session)
             let stream = try await provider.streamText(request: self.sampleRequest)
 
             var collected = ""
@@ -566,29 +566,18 @@ struct OpenAIResponsesProviderTests {
     }
 
     private static func streamChunkJSON(content: String?, finishReason: String?) -> String {
-        var delta: [String: Any] = [
-            "role": "assistant",
-        ]
         if let content {
-            delta["content"] = content
-        }
-
-        var choice: [String: Any] = [
-            "index": 0,
-            "delta": delta,
-        ]
-        if let finishReason {
-            choice["finish_reason"] = finishReason
+            let chunk: [String: Any] = [
+                "type": "response.output_text.delta",
+                "delta": content,
+            ]
+            let data = try! JSONSerialization.data(withJSONObject: chunk)
+            return String(data: data, encoding: .utf8)!
         }
 
         let chunk: [String: Any] = [
-            "id": "resp_stream",
-            "object": "response",
-            "created": 1_700_000_000,
-            "model": "o4-mini",
-            "choices": [choice],
+            "type": finishReason == nil ? "response.output_text.done" : "response.completed",
         ]
-
         let data = try! JSONSerialization.data(withJSONObject: chunk)
         return String(data: data, encoding: .utf8)!
     }

Tests/TachikomaTests/Providers/ProviderEndToEndTests.swift

@@ -57,7 +57,7 @@ struct ProviderEndToEndTests {
             let config = Self.makeConfiguration { config in
                 config.setAPIKey("sk-live-openai", for: .openai)
             }
-            let provider = try OpenAIProvider(model: .gpt4o, configuration: config)
+            let provider = try OpenAIProvider(model: .gpt55, configuration: config)
             let response = try await provider.generateText(request: Self.basicRequest)
             #expect(response.text == "OpenAI chat success")
         }
@@ -410,7 +410,7 @@ struct ProviderEndToEndTests {
             "id": "chatcmpl-123",
             "object": "chat.completion",
             "created": 1_723_000_000,
-            "model": "gpt-4o",
+            "model": "gpt-5.5",
             "choices": [
                 [
                     "index": 0,

Tests/TachikomaTests/UIIntegrationTests.swift

@@ -102,7 +102,7 @@ struct UIIntegrationTests {
 
         let streamResult = StreamTextResult(
             stream: textStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 
@@ -159,7 +159,7 @@ struct UIIntegrationTests {
 
         let streamResult = StreamTextResult(
             stream: textStream,
-            model: .openai(.gpt4o),
+            model: .openai(.gpt55),
             settings: .default,
         )
 

Tests/TachikomaTests/UnifiedErrorsTests.swift

@@ -60,7 +60,7 @@ struct UnifiedErrorsTests {
             statusCode: 429,
             responseBody: "{\"error\": \"rate_limit\"}",
             provider: "openai",
-            modelId: "gpt-4",
+            modelId: "gpt-5.5",
             requestId: "req-123",
             retryAfter: 30,
             metadata: ["tokens_used": "5000"],
@@ -120,7 +120,7 @@ struct UnifiedErrorsTests {
             statusCode: 500,
             responseBody: "Internal server error",
             provider: "anthropic",
-            modelId: "claude-3",
+            modelId: "claude-opus-4-7",
             requestId: "req-456",
             errorType: .serverError,
             message: "Server encountered an error",

Tests/TachikomaTests/Utilities/UsageTrackingTests.swift

@@ -48,7 +48,7 @@ struct UsageTrackingTests {
         let tracker = UsageTracker(forTesting: true)
 
         let sessionId = tracker.startSession()
-        let model = LanguageModel.openai(.gpt4oMini)
+        let model = LanguageModel.openai(.gpt5Mini)
         let usage = Usage(inputTokens: 100, outputTokens: 50)
 
         // Record usage
@@ -65,7 +65,7 @@ struct UsageTrackingTests {
         #expect(session?.totalTokens == 150)
 
         let operation = session?.operations.first
-        #expect(operation?.modelId == "gpt-4o-mini")
+        #expect(operation?.modelId == "gpt-5-mini")
         #expect(operation?.providerName == "OpenAI")
         #expect(operation?.usage.inputTokens == 100)
         #expect(operation?.usage.outputTokens == 50)
@@ -83,7 +83,7 @@ struct UsageTrackingTests {
         let tracker = UsageTracker(forTesting: true)
 
         let sessionId = tracker.startSession()
-        let model = LanguageModel.openai(.gpt4oMini)
+        let model = LanguageModel.openai(.gpt5Mini)
 
         // Record multiple operations
         tracker.recordUsage(
@@ -117,10 +117,10 @@ struct UsageTrackingTests {
         let usage = Usage(inputTokens: 1_000_000, outputTokens: 1_000_000) // 1M tokens each for easy calculation
 
         // Test OpenAI pricing
-        let gpt4oMiniCost = calculator.calculateCost(for: .openai(.gpt4oMini), usage: usage)
-        #expect(gpt4oMiniCost.input == 0.15) // $0.15 per million input tokens
-        #expect(gpt4oMiniCost.output == 0.60) // $0.60 per million output tokens
-        #expect(gpt4oMiniCost.total == 0.75)
+        let gpt5MiniCost = calculator.calculateCost(for: .openai(.gpt5Mini), usage: usage)
+        #expect(gpt5MiniCost.input == 1.00)
+        #expect(gpt5MiniCost.output == 4.00)
+        #expect(gpt5MiniCost.total == 5.00)
 
         // Test Anthropic pricing
         let claudeHaikuCost = calculator.calculateCost(for: .anthropic(.haiku45), usage: usage)
@@ -145,7 +145,7 @@ struct UsageTrackingTests {
         let session1 = tracker.startSession()
         tracker.recordUsage(
             sessionId: session1,
-            model: .openai(.gpt4oMini),
+            model: .openai(.gpt5Mini),
             usage: Usage(inputTokens: 100, outputTokens: 50),
             operation: .textGeneration,
         )
@@ -173,7 +173,7 @@ struct UsageTrackingTests {
 
         // Check model breakdown
         #expect(totalUsage.modelBreakdown.count == 2)
-        #expect(totalUsage.modelBreakdown[LanguageModel.openai(.gpt4oMini).modelId] != nil)
+        #expect(totalUsage.modelBreakdown[LanguageModel.openai(.gpt5Mini).modelId] != nil)
         #expect(totalUsage.modelBreakdown[LanguageModel.anthropic(.haiku45).modelId] != nil)
 
         // Check operation breakdown
@@ -191,7 +191,7 @@ struct UsageTrackingTests {
         let sessionId = tracker.startSession()
         tracker.recordUsage(
             sessionId: sessionId,
-            model: .openai(.gpt4oMini),
+            model: .openai(.gpt5Mini),
             usage: Usage(inputTokens: 1000, outputTokens: 500),
             operation: .textGeneration,
         )
@@ -212,7 +212,7 @@ struct UsageTrackingTests {
         #expect(formattedReport.contains("Operations: 1"))
         #expect(formattedReport.contains("Total Tokens: 1500"))
         #expect(formattedReport.contains("OpenAI"))
-        #expect(formattedReport.contains("gpt-4o-mini"))
+        #expect(formattedReport.contains("gpt-5-mini"))
         #expect(formattedReport.contains("Text Generation"))
     }
 
@@ -223,7 +223,7 @@ struct UsageTrackingTests {
         let sessionId = tracker.startSession()
         tracker.recordUsage(
             sessionId: sessionId,
-            model: .openai(.gpt4oMini),
+            model: .openai(.gpt5Mini),
             usage: Usage(inputTokens: 100, outputTokens: 50),
             operation: .textGeneration,
         )

docs/models.md

@@ -8,14 +8,11 @@ Tachikoma ships with a built-in model catalog (`CaseIterable` enums) plus suppor
 
 ## OpenAI (`LanguageModel.OpenAI`)
 
-- `o4-mini`
+- `gpt-5.5`
 - `gpt-5.2`, `gpt-5.1`
 - `gpt-5`, `gpt-5-pro`, `gpt-5-mini`, `gpt-5-nano`
 - `gpt-5-thinking`, `gpt-5-thinking-mini`, `gpt-5-thinking-nano`
 - `gpt-5-chat-latest`
-- `gpt-4.1`, `gpt-4.1-mini`
-- `gpt-4o`, `gpt-4o-mini`, `gpt-4o-realtime-preview`
-- `gpt-4-turbo`, `gpt-3.5-turbo`
 
 Notes:
 - Mini/Nano variants exist only for **GPT‑5** (not for GPT‑5.1 / GPT‑5.2).

docs/modern-api.md

@@ -9,7 +9,7 @@ Based on the Vercel AI SDK patterns from `/Users/steipete/Downloads/ai-sdk.md`,
 The Vercel AI SDK provides these core patterns:
 - **Provider-specific imports**: `import { openai } from '@ai-sdk/openai'`
 - **Core functions**: `generateText()`, `streamText()`, `generateObject()`, `streamObject()`
-- **Model specification**: `model: openai('gpt-4o')`, `model: anthropic('claude-sonnet-4')`
+- **Model specification**: `model: openai('gpt-5.5')`, `model: anthropic('claude-sonnet-4')`
 - **Unified API**: Same function signatures across all providers
 - **Tool integration**: Simple `tools: { toolName: toolDefinition }` object
 - **Streaming first**: Built-in streaming support with AsyncSequence patterns
@@ -55,7 +55,7 @@ The Vercel AI SDK provides these core patterns:
 - [ ] **Replace LanguageModel enum with LanguageModel protocol**
   - [ ] Create `LanguageModel` protocol with provider metadata
   - [ ] Create provider-specific model structs: `OpenAIModel`, `AnthropicModel`, etc.
-  - [ ] Implement model creation: `openai("gpt-4o")`, `anthropic("claude-sonnet-4")`
+  - [ ] Implement model creation: `openai("gpt-5.5")`, `anthropic("claude-sonnet-4")`
   - [ ] Add model capabilities: `supportsTools`, `supportsStreaming`, `supportsImages`
 
 #### 1.3 Core API Functions Redesign
@@ -105,7 +105,7 @@ The Vercel AI SDK provides these core patterns:
   - [ ] Google Gemini API integration
 
 - [ ] **Provider-specific features**
-  - [ ] Reasoning support for GPT-5 and o4-mini models
+  - [ ] Reasoning support for GPT-5 thinking models
   - [ ] Vision support for multimodal models
   - [ ] Function calling for compatible models
   - [ ] Streaming optimization per provider