feat(providers): default ChatGPT OAuth to GPT-5.5

Default ChatGPT Subscription (OAuth) provider model selection to GPT-5.5 and update model metadata, tests, and docs.

CHANGELOG.md

@@ -23,6 +23,12 @@ All notable changes to GoClaw are documented here. For full documentation, see [
   with downgrade protection. WebSocket events `package.update.*` for owner
   clients. See `docs/packages-github.md` § "Updating Installed Packages".
 
+### Changed
+
+- **ChatGPT Subscription (OAuth)** — default model and backend-owned model catalog
+  now prefer `gpt-5.5`, with reasoning metadata and context-window defaults updated
+  for provider-first model selection.
+
 ### Breaking Changes
 
 - **Context pruning now opt-in.** Previously tool-result trimming ran by default

cmd/auth.go

@@ -45,7 +45,7 @@ func authStatusCmd() *cobra.Command {
 					name = provider
 				}
 				fmt.Printf("ChatGPT OAuth account: active (alias: %s)\n", name)
-				fmt.Printf("Use model prefix '%s/' in agent config (e.g. %s/gpt-5.4).\n", name, name)
+				fmt.Printf("Use model prefix '%s/' in agent config (e.g. %s/gpt-5.5).\n", name, name)
 			} else {
 				fmt.Printf("No ChatGPT OAuth tokens found for alias '%s'.\n", provider)
 				fmt.Println("Use the web UI to authenticate this ChatGPT OAuth account.")

docs/02-providers.md

@@ -62,7 +62,7 @@ Streaming fallback is conservative: backup models are tried only if the stream f
 |----------|------|----------|---------------|
 | **anthropic** | Native HTTP + SSE | API key required | `claude-sonnet-4-5-20250929` |
 | **claude_cli** | stdio subprocess + MCP | Binary path (default: `claude`) | `sonnet` |
-| **codex** | OAuth Responses API | OAuth token source | `gpt-5.3-codex` |
+| **codex** | OAuth Responses API | OAuth token source | `gpt-5.5` |
 | **acp** | JSON-RPC 2.0 subagents | Binary + workspace dir | `claude` |
 | **dashscope** | OpenAI-compat wrapper | API key + custom models | `qwen3-max` |
 | **openai** (+ 10+ variants) | OpenAI-compatible | API key + endpoint URL | Model-specific |
@@ -571,7 +571,7 @@ Claude CLI inherits thinking support from the underlying Claude model. Thinking
 
 ## 12. Codex Provider
 
-The Codex provider integrates with OpenAI's ChatGPT Responses API (OAuth-based), enabling access to gpt-5.3-codex model through the chatgpt.com backend. Unlike standard OpenAI endpoints, Codex uses OAuth token refresh and a custom response format with "phase" markers.
+The Codex provider integrates with OpenAI's ChatGPT Responses API (OAuth-based), defaulting to `gpt-5.5` through the chatgpt.com backend. Unlike standard OpenAI endpoints, Codex uses OAuth token refresh and a custom response format with "phase" markers.
 
 ### Configuration
 
@@ -582,7 +582,7 @@ tokenSource := &MyTokenSource{} // implements TokenSource interface
 provider := NewCodexProvider("codex", tokenSource, "", "")
 // or specify custom API base and model:
 provider := NewCodexProvider("codex", tokenSource,
-  "https://chatgpt.com/backend-api", "gpt-5.3-codex")
+  "https://chatgpt.com/backend-api", "gpt-5.5")
 ```
 
 ### API Endpoint
@@ -601,7 +601,7 @@ Codex returns structured responses with phase markers:
 ```json
 {
   "id": "...",
-  "model": "gpt-5.3-codex",
+  "model": "gpt-5.5",
   "choices": [{
     "message": {
       "role": "assistant",

docs/18-http-api.md

@@ -148,7 +148,7 @@ can override or inherit routing behavior while keeping the main `provider` field
 ```json
 {
   "provider": "openai-codex",
-  "model": "gpt-5.4",
+  "model": "gpt-5.5",
   "chatgpt_oauth_routing": {
     "override_mode": "custom",
     "strategy": "round_robin"
@@ -214,7 +214,7 @@ Agents can now store capability-aware GPT-5/Codex reasoning intent under top-lev
 ```json
 {
   "provider": "openai-codex",
-  "model": "gpt-5.4",
+  "model": "gpt-5.5",
   "reasoning_config": {
     "override_mode": "inherit"
   }
@@ -364,11 +364,11 @@ Example response:
 {
   "models": [
     {
-      "id": "gpt-5.4",
-      "name": "GPT-5.4",
+      "id": "gpt-5.5",
+      "name": "GPT-5.5",
       "reasoning": {
         "levels": ["none", "low", "medium", "high", "xhigh"],
-        "default_effort": "none"
+        "default_effort": "medium"
       }
     },
     {

docs/project-changelog.md

@@ -6,6 +6,18 @@ Significant changes, features, and fixes in reverse chronological order.
 
 ## 2026-05-18
 
+### Providers: ChatGPT OAuth GPT-5.5 default
+
+**Changed**
+
+- Updated ChatGPT Subscription (OAuth) default model, provider model catalog, reasoning metadata, and docs examples to prefer `gpt-5.5`.
+
+**Tests**
+
+- Updated focused provider/model catalog, reasoning, registry, and token-window coverage for `gpt-5.5`.
+
+---
+
 ### Deployment: Codex CLI service-user auth
 
 **Fixes**

internal/http/provider_models_catalog.go

@@ -1,5 +1,7 @@
 package http
 
+import "github.com/nextlevelbuilder/goclaw/internal/providers"
+
 // bailianModels returns a hardcoded list of models available on the
 // Bailian Coding platform (coding-intl.dashscope.aliyuncs.com).
 // The platform does not expose a /v1/models endpoint.
@@ -84,6 +86,7 @@ func acpModels() []ModelInfo {
 // chatGPTOAuthModels returns models available via ChatGPT OAuth integration.
 func chatGPTOAuthModels() []ModelInfo {
 	return withReasoningCapabilities([]ModelInfo{
+		{ID: providers.DefaultCodexModel, Name: "GPT-5.5"},
 		{ID: "gpt-5.4", Name: "GPT-5.4"},
 		{ID: "gpt-5.4-mini", Name: "GPT-5.4 Mini"},
 		{ID: "gpt-5.3-codex", Name: "GPT-5.3 Codex"},

internal/http/provider_models_test.go

@@ -97,22 +97,22 @@ func TestProvidersHandlerListProviderModelsChatGPTOAuthIncludesReasoningMetadata
 
 	var found bool
 	for _, model := range result.Models {
-		if model.ID != "gpt-5.4" {
+		if model.ID != "gpt-5.5" {
 			continue
 		}
 		found = true
 		if model.Reasoning == nil {
-			t.Fatal("gpt-5.4 reasoning = nil, want capability metadata")
+			t.Fatal("gpt-5.5 reasoning = nil, want capability metadata")
 		}
-		if model.Reasoning.DefaultEffort != "none" {
-			t.Fatalf("gpt-5.4 default_effort = %q, want none", model.Reasoning.DefaultEffort)
+		if model.Reasoning.DefaultEffort != "medium" {
+			t.Fatalf("gpt-5.5 default_effort = %q, want medium", model.Reasoning.DefaultEffort)
 		}
 		if got := model.Reasoning.Levels; len(got) != 5 || got[4] != "xhigh" {
-			t.Fatalf("gpt-5.4 levels = %#v, want none..xhigh", got)
+			t.Fatalf("gpt-5.5 levels = %#v, want none..xhigh", got)
 		}
 	}
 	if !found {
-		t.Fatal("gpt-5.4 not found in ChatGPT OAuth model list")
+		t.Fatal("gpt-5.5 not found in ChatGPT OAuth model list")
 	}
 }
 

internal/providers/adapter_codex.go

@@ -26,7 +26,7 @@ func NewCodexAdapter(cfg ProviderConfig) (ProviderAdapter, error) {
 
 	model := cfg.Model
 	if model == "" {
-		model = "gpt-5.4"
+		model = DefaultCodexModel
 	}
 
 	var ts TokenSource
@@ -53,7 +53,7 @@ func (a *CodexAdapter) Capabilities() ProviderCapabilities {
 		Vision:           true,
 		CacheControl:     false,
 		ImageGeneration:  true, // Codex (OpenAI Responses API) supports native image_generation tool
-		MaxContextWindow: 1_000_000,
+		MaxContextWindow: 1_050_000,
 		TokenizerID:      "o200k_base",
 	}
 }

internal/providers/adapter_codex_test.go

@@ -35,12 +35,23 @@ func TestCodexAdapter_Defaults(t *testing.T) {
 	if !caps.Thinking || !caps.Vision {
 		t.Errorf("expected Thinking+Vision, got %+v", caps)
 	}
-	if caps.MaxContextWindow != 1_000_000 {
-		t.Errorf("MaxContextWindow = %d, want 1_000_000", caps.MaxContextWindow)
+	if caps.MaxContextWindow != 1_050_000 {
+		t.Errorf("MaxContextWindow = %d, want 1_050_000", caps.MaxContextWindow)
 	}
 	if caps.TokenizerID != "o200k_base" {
 		t.Errorf("TokenizerID = %q, want o200k_base", caps.TokenizerID)
 	}
+	body, _, err := a.ToRequest(ChatRequest{Messages: []Message{{Role: "user", Content: "hi"}}})
+	if err != nil {
+		t.Fatalf("ToRequest error: %v", err)
+	}
+	var payload map[string]any
+	if err := json.Unmarshal(body, &payload); err != nil {
+		t.Fatalf("decode body: %v", err)
+	}
+	if payload["model"] != DefaultCodexModel {
+		t.Errorf("default model = %v, want %s", payload["model"], DefaultCodexModel)
+	}
 }
 
 // TestCodexAdapter_BaseURLTrimmed verifies trailing slashes are stripped from

internal/providers/codex.go

@@ -15,6 +15,9 @@ type CodexRoutingDefaults struct {
 	ExtraProviderNames []string
 }
 
+// DefaultCodexModel is the default model for ChatGPT Subscription (OAuth).
+const DefaultCodexModel = "gpt-5.5"
+
 // CodexProvider implements Provider for the OpenAI Responses API,
 // used with ChatGPT subscription via OAuth (Codex flow).
 // Wire format: POST /codex/responses on chatgpt.com backend.
@@ -37,7 +40,7 @@ func NewCodexProvider(name string, tokenSource TokenSource, apiBase, defaultMode
 	apiBase = strings.TrimRight(apiBase, "/")
 
 	if defaultModel == "" {
-		defaultModel = "gpt-5.4"
+		defaultModel = DefaultCodexModel
 	}
 
 	return &CodexProvider{
@@ -78,7 +81,7 @@ func (p *CodexProvider) Capabilities() ProviderCapabilities {
 		Vision:           true,
 		CacheControl:     false,
 		ImageGeneration:  true, // Codex (OpenAI Responses API) supports native image_generation tool
-		MaxContextWindow: 1_000_000,
+		MaxContextWindow: 1_050_000,
 		TokenizerID:      "o200k_base",
 	}
 }

internal/providers/codex_test.go

@@ -47,8 +47,8 @@ func TestCodexProviderName(t *testing.T) {
 
 func TestCodexProviderDefaultModel(t *testing.T) {
 	p := NewCodexProvider("test", &staticTokenSource{token: "test"}, "", "")
-	if p.DefaultModel() != "gpt-5.4" {
-		t.Errorf("DefaultModel() = %q, want %q", p.DefaultModel(), "gpt-5.4")
+	if p.DefaultModel() != DefaultCodexModel {
+		t.Errorf("DefaultModel() = %q, want %q", p.DefaultModel(), DefaultCodexModel)
 	}
 
 	p2 := NewCodexProvider("test", &staticTokenSource{token: "test"}, "", "o3")

internal/providers/forward_compat_openai.go

@@ -11,11 +11,10 @@ var openAIForwardCompatMap = map[string]struct {
 	Patch    *ModelSpec
 }{
 	"gpt-5.5": {
-		Template: "gpt-5.4",
-		Patch:    &ModelSpec{ContextWindow: 1_000_000, MaxTokens: 200_000},
+		Template: "gpt-5.5",
 	},
 	"gpt-5.6": {
-		Template: "gpt-5.4",
+		Template: "gpt-5.5",
 		Patch:    &ModelSpec{ContextWindow: 2_000_000, MaxTokens: 200_000},
 	},
 	"o5-mini": {
@@ -24,7 +23,7 @@ var openAIForwardCompatMap = map[string]struct {
 	},
 }
 
-// ResolveForwardCompat handles models like "gpt-5.5" by cloning from "gpt-5.4".
+// ResolveForwardCompat handles future model aliases by cloning from the latest known template.
 func (r *OpenAIForwardCompat) ResolveForwardCompat(modelID string, registry ModelRegistry) *ModelSpec {
 	// Direct map lookup (exact match)
 	if entry, ok := openAIForwardCompatMap[modelID]; ok {

internal/providers/model_registry.go

@@ -154,6 +154,7 @@ func SeedDefaultModels(r *InMemoryRegistry) {
 
 	// OpenAI models
 	for _, s := range []ModelSpec{
+		{ID: "gpt-5.5", Provider: "openai", ContextWindow: 1_050_000, MaxTokens: 128_000, Reasoning: true, Vision: true, TokenizerID: "o200k_base"},
 		{ID: "gpt-5.4", Provider: "openai", ContextWindow: 1_000_000, MaxTokens: 100_000, Reasoning: true, Vision: true, TokenizerID: "o200k_base"},
 		{ID: "gpt-5.2", Provider: "openai", ContextWindow: 256_000, MaxTokens: 64_000, Reasoning: true, Vision: true, TokenizerID: "o200k_base"},
 		{ID: "gpt-4o", Provider: "openai", ContextWindow: 128_000, MaxTokens: 16_384, Reasoning: false, Vision: true, TokenizerID: "o200k_base"},

internal/providers/model_registry_test.go

@@ -263,10 +263,10 @@ func TestCloneFromTemplatePatchBooleanFields(t *testing.T) {
 	registry := &InMemoryRegistry{}
 
 	template := ModelSpec{
-		ID:       "base",
-		Provider: "openai",
+		ID:        "base",
+		Provider:  "openai",
 		Reasoning: false,
-		Vision:   false,
+		Vision:    false,
 	}
 	registry.Register(template)
 
@@ -371,7 +371,7 @@ func TestOpenAIForwardCompatResolveExactMatch(t *testing.T) {
 	resolver := &OpenAIForwardCompat{}
 	registry.RegisterResolver("openai", resolver)
 
-	// Try to resolve gpt-5.5 which should use gpt-5.4 as template with patch
+	// gpt-5.5 is a seeded model, so direct resolution should win.
 	resolved := registry.Resolve("openai", "gpt-5.5")
 
 	if resolved == nil {
@@ -380,12 +380,11 @@ func TestOpenAIForwardCompatResolveExactMatch(t *testing.T) {
 	if resolved.ID != "gpt-5.5" {
 		t.Errorf("expected ID=gpt-5.5, got %s", resolved.ID)
 	}
-	// Should have patched values from the map
-	if resolved.ContextWindow != 1_000_000 {
-		t.Errorf("expected ContextWindow=1000000 from patch, got %d", resolved.ContextWindow)
+	if resolved.ContextWindow != 1_050_000 {
+		t.Errorf("expected ContextWindow=1050000, got %d", resolved.ContextWindow)
 	}
-	if resolved.MaxTokens != 200_000 {
-		t.Errorf("expected MaxTokens=200000 from patch, got %d", resolved.MaxTokens)
+	if resolved.MaxTokens != 128_000 {
+		t.Errorf("expected MaxTokens=128000, got %d", resolved.MaxTokens)
 	}
 }
 
@@ -403,9 +402,12 @@ func TestOpenAIForwardCompatResolvePrefixMatch(t *testing.T) {
 	if resolved.ID != "gpt-5.5-turbo" {
 		t.Errorf("expected ID=gpt-5.5-turbo, got %s", resolved.ID)
 	}
-	// Should use gpt-5.4 as template with gpt-5.5 patch
-	if resolved.MaxTokens != 200_000 {
-		t.Errorf("expected MaxTokens=200000 from patch, got %d", resolved.MaxTokens)
+	// Should use gpt-5.5 as the latest known template.
+	if resolved.ContextWindow != 1_050_000 {
+		t.Errorf("expected ContextWindow=1050000 from template, got %d", resolved.ContextWindow)
+	}
+	if resolved.MaxTokens != 128_000 {
+		t.Errorf("expected MaxTokens=128000 from template, got %d", resolved.MaxTokens)
 	}
 }
 
@@ -462,8 +464,8 @@ func TestCloneFromTemplatePatchZeroValuesIgnored(t *testing.T) {
 
 	// Patch with zero values should be ignored
 	patch := &ModelSpec{
-		ContextWindow: 0, // Should be ignored
-		MaxTokens:     0, // Should be ignored
+		ContextWindow: 0,  // Should be ignored
+		MaxTokens:     0,  // Should be ignored
 		TokenizerID:   "", // Should be ignored
 		Cost: ModelCost{
 			InputPer1M: 0, // Should be ignored

internal/providers/native_image.go

@@ -41,7 +41,7 @@ func ValidateImageModel(model string) (string, error) {
 
 // NativeImageRequest describes a single image generation request.
 type NativeImageRequest struct {
-	// Model is the parent LLM model for the Responses API call (e.g. "gpt-5.4").
+	// Model is the parent LLM model for the Responses API call (e.g. "gpt-5.5").
 	// NOT the image model — see ImageModel below.
 	// If empty, the provider uses its own default LLM model.
 	Model string

internal/providers/openai_request.go

@@ -181,7 +181,7 @@ func (p *OpenAIProvider) buildRequestBody(model string, req ChatRequest, stream
 	if v, ok := req.Options[OptTemperature]; ok {
 		// Certain model families don't support custom temperature (locked to default).
 		// This is a model-level constraint, not provider-specific — applies to both OpenAI and Azure.
-		// Note: gpt-5.X flagship models (gpt-5.1, gpt-5.4) DO support temperature;
+		// Note: gpt-5.X flagship models (gpt-5.1, gpt-5.4, gpt-5.5) DO support temperature;
 		// only the mini/nano reasoning variants reject it.
 		skipTemp := strings.HasPrefix(capabilityModel, "gpt-5-mini") || strings.HasPrefix(capabilityModel, "gpt-5-nano") || strings.HasPrefix(capabilityModel, "o1") || strings.HasPrefix(capabilityModel, "o3") || strings.HasPrefix(capabilityModel, "o4")
 		if !skipTemp {

internal/providers/reasoning_capability.go

@@ -25,6 +25,7 @@ type reasoningCapabilityEntry struct {
 }
 
 var reasoningCapabilityEntries = []reasoningCapabilityEntry{
+	{id: "gpt-5.5", capability: ReasoningCapability{Levels: []string{"none", "low", "medium", "high", "xhigh"}, DefaultEffort: "medium"}},
 	{id: "gpt-5.4-mini", capability: ReasoningCapability{Levels: []string{"none", "low", "medium", "high", "xhigh"}, DefaultEffort: "none"}},
 	{id: "gpt-5-mini", capability: ReasoningCapability{Levels: []string{"none", "low", "medium", "high", "xhigh"}, DefaultEffort: "none"}},
 	{id: "gpt-5.4", capability: ReasoningCapability{Levels: []string{"none", "low", "medium", "high", "xhigh"}, DefaultEffort: "none"}},

internal/providers/reasoning_resolution_test.go

@@ -30,6 +30,14 @@ func TestLookupReasoningCapability(t *testing.T) {
 	if capability.Supports("low") {
 		t.Fatal("expected gpt-5.1-codex-max to reject low")
 	}
+
+	capability = LookupReasoningCapability("gpt-5.5")
+	if capability == nil {
+		t.Fatal("LookupReasoningCapability(gpt-5.5) = nil, want capability")
+	}
+	if capability.DefaultEffort != "medium" {
+		t.Fatalf("gpt-5.5 default_effort = %q, want medium", capability.DefaultEffort)
+	}
 }
 
 func TestResolveReasoningDecisionDowngradesUnsupportedEffort(t *testing.T) {

internal/tokencount/tiktoken_counter_test.go

@@ -82,6 +82,7 @@ func TestModelContextWindow(t *testing.T) {
 	}{
 		{"claude-sonnet-4-5-20250929", 200_000},
 		{"gpt-4o-mini", 128_000},
+		{"gpt-5.5", 1_050_000},
 		{"gpt-5.4", 1_000_000},
 		{"unknown-model", 200_000}, // conservative default
 	}

internal/tokencount/token_counter.go

@@ -31,8 +31,8 @@ type TokenizerID string
 
 const (
 	TokenizerCL100K   TokenizerID = "cl100k_base" // Claude, GPT-3.5/4
-	TokenizerO200K    TokenizerID = "o200k_base"   // GPT-4o, GPT-5
-	TokenizerFallback TokenizerID = "fallback"      // rune-count / 3
+	TokenizerO200K    TokenizerID = "o200k_base"  // GPT-4o, GPT-5
+	TokenizerFallback TokenizerID = "fallback"    // rune-count / 3
 )
 
 // ModelInfo maps a model name prefix to its tokenizer + context window.
@@ -47,6 +47,7 @@ var DefaultRegistry = map[string]ModelInfo{
 	"claude-":   {TokenizerCL100K, 200_000},
 	"gpt-4o":    {TokenizerO200K, 128_000},
 	"gpt-4":     {TokenizerCL100K, 128_000},
+	"gpt-5.5":   {TokenizerO200K, 1_050_000},
 	"gpt-5":     {TokenizerO200K, 1_000_000},
 	"qwen-":     {TokenizerCL100K, 128_000},
 	"deepseek-": {TokenizerCL100K, 128_000},

ui/web/src/pages/providers/provider-oauth-section.tsx

@@ -61,7 +61,7 @@ export function OAuthSection({
       <p className="mt-1">
         {t("oauth.modelPrefixHint")} <code className="rounded bg-muted px-1 font-mono">{provider}/</code>{" "}
         {t("oauth.modelPrefixExample", {
-          example: `${provider}/gpt-5.4`,
+          example: `${provider}/gpt-5.5`,
         })}
       </p>
     </div>
@@ -244,7 +244,7 @@ export function OAuthSection({
         <p className="mt-1">
           {t("oauth.modelPrefixHint")} <code className="rounded bg-muted px-1 font-mono">{resolvedProviderName}/</code>{" "}
           {t("oauth.modelPrefixExample", {
-            example: `${resolvedProviderName}/gpt-5.4`,
+            example: `${resolvedProviderName}/gpt-5.5`,
           })}
         </p>
       </div>