Merge pull request #1519 from router-for-me/thinking

feat(translator): support Claude thinking type adaptive

internal/registry/model_definitions_static_data.go

@@ -866,7 +866,7 @@ func GetAntigravityModelConfig() map[string]*AntigravityModelConfig {
 		"gemini-3-flash":             {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}}},
 		"claude-sonnet-4-5-thinking": {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
 		"claude-opus-4-5-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
-		"claude-opus-4-6-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 128000},
+		"claude-opus-4-6-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
 		"claude-sonnet-4-5":          {MaxCompletionTokens: 64000},
 		"gpt-oss-120b-medium":        {},
 		"tab_flash_lite_preview":     {},

internal/translator/antigravity/claude/antigravity_claude_request.go

@@ -344,7 +344,8 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 	// Inject interleaved thinking hint when both tools and thinking are active
 	hasTools := toolDeclCount > 0
 	thinkingResult := gjson.GetBytes(rawJSON, "thinking")
-	hasThinking := thinkingResult.Exists() && thinkingResult.IsObject() && thinkingResult.Get("type").String() == "enabled"
+	thinkingType := thinkingResult.Get("type").String()
+	hasThinking := thinkingResult.Exists() && thinkingResult.IsObject() && (thinkingType == "enabled" || thinkingType == "adaptive")
 	isClaudeThinking := util.IsClaudeThinkingModel(modelName)
 
 	if hasTools && hasThinking && isClaudeThinking {
@@ -377,12 +378,18 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when type==enabled
 	if t := gjson.GetBytes(rawJSON, "thinking"); enableThoughtTranslate && t.Exists() && t.IsObject() {
-		if t.Get("type").String() == "enabled" {
+		switch t.Get("type").String() {
+		case "enabled":
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.includeThoughts", true)
 			}
+		case "adaptive":
+			// Keep adaptive as a high level sentinel; ApplyThinking resolves it
+			// to model-specific max capability.
+			out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingLevel", "high")
+			out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.includeThoughts", true)
 		}
 	}
 	if v := gjson.GetBytes(rawJSON, "temperature"); v.Exists() && v.Type == gjson.Number {

internal/translator/codex/claude/codex_claude_request.go

@@ -222,6 +222,10 @@ func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 					reasoningEffort = effort
 				}
 			}
+		case "adaptive":
+			// Claude adaptive means "enable with max capacity"; keep it as highest level
+			// and let ApplyThinking normalize per target model capability.
+			reasoningEffort = string(thinking.LevelXHigh)
 		case "disabled":
 			if effort, ok := thinking.ConvertBudgetToLevel(0); ok && effort != "" {
 				reasoningEffort = effort

internal/translator/gemini-cli/claude/gemini-cli_claude_request.go

@@ -173,12 +173,18 @@ func ConvertClaudeRequestToCLI(modelName string, inputRawJSON []byte, _ bool) []
 
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when type==enabled
 	if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() {
-		if t.Get("type").String() == "enabled" {
+		switch t.Get("type").String() {
+		case "enabled":
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.includeThoughts", true)
 			}
+		case "adaptive":
+			// Keep adaptive as a high level sentinel; ApplyThinking resolves it
+			// to model-specific max capability.
+			out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingLevel", "high")
+			out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.includeThoughts", true)
 		}
 	}
 	if v := gjson.GetBytes(rawJSON, "temperature"); v.Exists() && v.Type == gjson.Number {

internal/translator/gemini/claude/gemini_claude_request.go

@@ -154,12 +154,18 @@ func ConvertClaudeRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when enabled
 	// Translator only does format conversion, ApplyThinking handles model capability validation.
 	if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() {
-		if t.Get("type").String() == "enabled" {
+		switch t.Get("type").String() {
+		case "enabled":
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())
 				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", budget)
 				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.includeThoughts", true)
 			}
+		case "adaptive":
+			// Keep adaptive as a high level sentinel; ApplyThinking resolves it
+			// to model-specific max capability.
+			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingLevel", "high")
+			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.includeThoughts", true)
 		}
 	}
 	if v := gjson.GetBytes(rawJSON, "temperature"); v.Exists() && v.Type == gjson.Number {

internal/translator/openai/claude/openai_claude_request.go

@@ -75,6 +75,10 @@ func ConvertClaudeRequestToOpenAI(modelName string, inputRawJSON []byte, stream
 						out, _ = sjson.Set(out, "reasoning_effort", effort)
 					}
 				}
+			case "adaptive":
+				// Claude adaptive means "enable with max capacity"; keep it as highest level
+				// and let ApplyThinking normalize per target model capability.
+				out, _ = sjson.Set(out, "reasoning_effort", string(thinking.LevelXHigh))
 			case "disabled":
 				if effort, ok := thinking.ConvertBudgetToLevel(0); ok && effort != "" {
 					out, _ = sjson.Set(out, "reasoning_effort", effort)

test/thinking_conversion_test.go

@@ -2590,6 +2590,135 @@ func TestThinkingE2EMatrix_Body(t *testing.T) {
 	runThinkingTests(t, cases)
 }
 
+// TestThinkingE2EClaudeAdaptive_Body tests Claude thinking.type=adaptive extended body-only cases.
+// These cases validate that adaptive means "thinking enabled without explicit budget", and
+// cross-protocol conversion should resolve to target-model maximum thinking capability.
+func TestThinkingE2EClaudeAdaptive_Body(t *testing.T) {
+	reg := registry.GetGlobalRegistry()
+	uid := fmt.Sprintf("thinking-e2e-claude-adaptive-%d", time.Now().UnixNano())
+
+	reg.RegisterClient(uid, "test", getTestModels())
+	defer reg.UnregisterClient(uid)
+
+	cases := []thinkingTestCase{
+		// A1: Claude adaptive to OpenAI level model -> highest supported level
+		{
+			name:        "A1",
+			from:        "claude",
+			to:          "openai",
+			model:       "level-model",
+			inputJSON:   `{"model":"level-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField: "reasoning_effort",
+			expectValue: "high",
+			expectErr:   false,
+		},
+		// A2: Claude adaptive to Gemini level subset model -> highest supported level
+		{
+			name:            "A2",
+			from:            "claude",
+			to:              "gemini",
+			model:           "level-subset-model",
+			inputJSON:       `{"model":"level-subset-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField:     "generationConfig.thinkingConfig.thinkingLevel",
+			expectValue:     "high",
+			includeThoughts: "true",
+			expectErr:       false,
+		},
+		// A3: Claude adaptive to Gemini budget model -> max budget
+		{
+			name:            "A3",
+			from:            "claude",
+			to:              "gemini",
+			model:           "gemini-budget-model",
+			inputJSON:       `{"model":"gemini-budget-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField:     "generationConfig.thinkingConfig.thinkingBudget",
+			expectValue:     "20000",
+			includeThoughts: "true",
+			expectErr:       false,
+		},
+		// A4: Claude adaptive to Gemini mixed model -> highest supported level
+		{
+			name:            "A4",
+			from:            "claude",
+			to:              "gemini",
+			model:           "gemini-mixed-model",
+			inputJSON:       `{"model":"gemini-mixed-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField:     "generationConfig.thinkingConfig.thinkingLevel",
+			expectValue:     "high",
+			includeThoughts: "true",
+			expectErr:       false,
+		},
+		// A5: Claude adaptive passthrough for same protocol
+		{
+			name:        "A5",
+			from:        "claude",
+			to:          "claude",
+			model:       "claude-budget-model",
+			inputJSON:   `{"model":"claude-budget-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField: "thinking.type",
+			expectValue: "adaptive",
+			expectErr:   false,
+		},
+		// A6: Claude adaptive to Antigravity budget model -> max budget
+		{
+			name:            "A6",
+			from:            "claude",
+			to:              "antigravity",
+			model:           "antigravity-budget-model",
+			inputJSON:       `{"model":"antigravity-budget-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField:     "request.generationConfig.thinkingConfig.thinkingBudget",
+			expectValue:     "20000",
+			includeThoughts: "true",
+			expectErr:       false,
+		},
+		// A7: Claude adaptive to iFlow GLM -> enabled boolean
+		{
+			name:        "A7",
+			from:        "claude",
+			to:          "iflow",
+			model:       "glm-test",
+			inputJSON:   `{"model":"glm-test","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField: "chat_template_kwargs.enable_thinking",
+			expectValue: "true",
+			expectErr:   false,
+		},
+		// A8: Claude adaptive to iFlow MiniMax -> enabled boolean
+		{
+			name:        "A8",
+			from:        "claude",
+			to:          "iflow",
+			model:       "minimax-test",
+			inputJSON:   `{"model":"minimax-test","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField: "reasoning_split",
+			expectValue: "true",
+			expectErr:   false,
+		},
+		// A9: Claude adaptive to Codex level model -> highest supported level
+		{
+			name:        "A9",
+			from:        "claude",
+			to:          "codex",
+			model:       "level-model",
+			inputJSON:   `{"model":"level-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField: "reasoning.effort",
+			expectValue: "high",
+			expectErr:   false,
+		},
+		// A10: Claude adaptive on non-thinking model should still be stripped
+		{
+			name:        "A10",
+			from:        "claude",
+			to:          "openai",
+			model:       "no-thinking-model",
+			inputJSON:   `{"model":"no-thinking-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
+			expectField: "",
+			expectErr:   false,
+		},
+	}
+
+	runThinkingTests(t, cases)
+}
+
 // getTestModels returns the shared model definitions for E2E tests.
 func getTestModels() []*registry.ModelInfo {
 	return []*registry.ModelInfo{