**fix(translator): handle non-JSON output parsing for OpenAI function responses**

- Updated `antigravity_openai_request.go` to process non-JSON outputs gracefully by verifying and distinguishing between JSON and plain string formats. - Ensured proper assignment of parsed or raw response to `functionResponse`.
**feat(registry): add context length and update max tokens for Claude model configurations**
2026-02-14 02:10:52 +08:00 · 2025-11-27 16:18:49 +08:00 · 2025-11-27 16:13:25 +08:00 · 2025-11-27 16:03:24 +08:00 · 2025-11-27 15:59:15 +08:00 · 2025-11-26 22:31:05 -08:00
5 changed files with 244 additions and 89 deletions
--- a/internal/registry/model_definitions.go
+++ b/internal/registry/model_definitions.go
@@ -8,65 +8,79 @@ func GetClaudeModels() []*ModelInfo {
 	return []*ModelInfo{

 		{
-			ID:          "claude-haiku-4-5-20251001",
-			Object:      "model",
-			Created:     1759276800, // 2025-10-01
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4.5 Haiku",
+			ID:                  "claude-haiku-4-5-20251001",
+			Object:              "model",
+			Created:             1759276800, // 2025-10-01
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Haiku",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
 		},
 		{
-			ID:          "claude-sonnet-4-5-20250929",
-			Object:      "model",
-			Created:     1759104000, // 2025-09-29
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4.5 Sonnet",
+			ID:                  "claude-sonnet-4-5-20250929",
+			Object:              "model",
+			Created:             1759104000, // 2025-09-29
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Sonnet",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
 		},
 		{
-			ID:          "claude-sonnet-4-5-thinking",
-			Object:      "model",
-			Created:     1759104000, // 2025-09-29
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4.5 Sonnet Thinking",
-			Thinking:    &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
+			ID:                  "claude-sonnet-4-5-thinking",
+			Object:              "model",
+			Created:             1759104000, // 2025-09-29
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Sonnet Thinking",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
-			ID:          "claude-opus-4-5-thinking",
-			Object:      "model",
-			Created:     1761955200, // 2025-11-01
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4.5 Opus Thinking",
-			Thinking:    &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
+			ID:                  "claude-opus-4-5-thinking",
+			Object:              "model",
+			Created:             1761955200, // 2025-11-01
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Opus Thinking",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
-			ID:          "claude-opus-4-5-thinking-low",
-			Object:      "model",
-			Created:     1761955200, // 2025-11-01
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4.5 Opus Thinking Low",
-			Thinking:    &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
+			ID:                  "claude-opus-4-5-thinking-low",
+			Object:              "model",
+			Created:             1761955200, // 2025-11-01
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Opus Thinking Low",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
-			ID:          "claude-opus-4-5-thinking-medium",
-			Object:      "model",
-			Created:     1761955200, // 2025-11-01
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4.5 Opus Thinking Medium",
-			Thinking:    &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
+			ID:                  "claude-opus-4-5-thinking-medium",
+			Object:              "model",
+			Created:             1761955200, // 2025-11-01
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Opus Thinking Medium",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
-			ID:          "claude-opus-4-5-thinking-high",
-			Object:      "model",
-			Created:     1761955200, // 2025-11-01
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4.5 Opus Thinking High",
-			Thinking:    &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
+			ID:                  "claude-opus-4-5-thinking-high",
+			Object:              "model",
+			Created:             1761955200, // 2025-11-01
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Opus Thinking High",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
 			ID:                  "claude-opus-4-5-20251101",
@@ -80,44 +94,54 @@ func GetClaudeModels() []*ModelInfo {
 			MaxCompletionTokens: 64000,
 		},
 		{
-			ID:          "claude-opus-4-1-20250805",
-			Object:      "model",
-			Created:     1722945600, // 2025-08-05
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4.1 Opus",
+			ID:                  "claude-opus-4-1-20250805",
+			Object:              "model",
+			Created:             1722945600, // 2025-08-05
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.1 Opus",
+			ContextLength:       200000,
+			MaxCompletionTokens: 32000,
 		},
 		{
-			ID:          "claude-opus-4-20250514",
-			Object:      "model",
-			Created:     1715644800, // 2025-05-14
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4 Opus",
+			ID:                  "claude-opus-4-20250514",
+			Object:              "model",
+			Created:             1715644800, // 2025-05-14
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4 Opus",
+			ContextLength:       200000,
+			MaxCompletionTokens: 32000,
 		},
 		{
-			ID:          "claude-sonnet-4-20250514",
-			Object:      "model",
-			Created:     1715644800, // 2025-05-14
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 4 Sonnet",
+			ID:                  "claude-sonnet-4-20250514",
+			Object:              "model",
+			Created:             1715644800, // 2025-05-14
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4 Sonnet",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
 		},
 		{
-			ID:          "claude-3-7-sonnet-20250219",
-			Object:      "model",
-			Created:     1708300800, // 2025-02-19
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 3.7 Sonnet",
+			ID:                  "claude-3-7-sonnet-20250219",
+			Object:              "model",
+			Created:             1708300800, // 2025-02-19
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 3.7 Sonnet",
+			ContextLength:       128000,
+			MaxCompletionTokens: 8192,
 		},
 		{
-			ID:          "claude-3-5-haiku-20241022",
-			Object:      "model",
-			Created:     1729555200, // 2024-10-22
-			OwnedBy:     "anthropic",
-			Type:        "claude",
-			DisplayName: "Claude 3.5 Haiku",
+			ID:                  "claude-3-5-haiku-20241022",
+			Object:              "model",
+			Created:             1729555200, // 2024-10-22
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 3.5 Haiku",
+			ContextLength:       128000,
+			MaxCompletionTokens: 8192,
 		},
 	}
 }
--- a/internal/runtime/executor/claude_executor.go
+++ b/internal/runtime/executor/claude_executor.go
@@ -17,6 +17,7 @@ import (
 	claudeauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/claude"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
 	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
@@ -66,6 +67,9 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)

+	// Ensure max_tokens > thinking.budget_tokens when thinking is enabled
+	body = ensureMaxTokensForThinking(req.Model, body)
+
 	// Extract betas from body and convert to header
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
@@ -165,6 +169,9 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	body = checkSystemInstructions(body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)

+	// Ensure max_tokens > thinking.budget_tokens when thinking is enabled
+	body = ensureMaxTokensForThinking(req.Model, body)
+
 	// Extract betas from body and convert to header
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
@@ -447,6 +454,42 @@ func (e *ClaudeExecutor) injectThinkingConfig(modelName string, body []byte) []b
 	return body
 }

+// ensureMaxTokensForThinking ensures max_tokens > thinking.budget_tokens when thinking is enabled.
+// Anthropic API requires this constraint; violating it returns a 400 error.
+// This function should be called after all thinking configuration is finalized.
+// It looks up the model's MaxCompletionTokens from the registry to use as the cap.
+func ensureMaxTokensForThinking(modelName string, body []byte) []byte {
+	thinkingType := gjson.GetBytes(body, "thinking.type").String()
+	if thinkingType != "enabled" {
+		return body
+	}
+
+	budgetTokens := gjson.GetBytes(body, "thinking.budget_tokens").Int()
+	if budgetTokens <= 0 {
+		return body
+	}
+
+	maxTokens := gjson.GetBytes(body, "max_tokens").Int()
+
+	// Look up the model's max completion tokens from the registry
+	maxCompletionTokens := 0
+	if modelInfo := registry.GetGlobalRegistry().GetModelInfo(modelName); modelInfo != nil {
+		maxCompletionTokens = modelInfo.MaxCompletionTokens
+	}
+
+	// Fall back to budget + buffer if registry lookup fails or returns 0
+	const fallbackBuffer = 4000
+	requiredMaxTokens := budgetTokens + fallbackBuffer
+	if maxCompletionTokens > 0 {
+		requiredMaxTokens = int64(maxCompletionTokens)
+	}
+
+	if maxTokens < requiredMaxTokens {
+		body, _ = sjson.SetBytes(body, "max_tokens", requiredMaxTokens)
+	}
+	return body
+}
+
 func (e *ClaudeExecutor) resolveUpstreamModel(alias string, auth *cliproxyauth.Auth) string {
 	if alias == "" {
 		return ""
--- a/internal/translator/antigravity/openai/chat-completions/antigravity_openai_request.go
+++ b/internal/translator/antigravity/openai/chat-completions/antigravity_openai_request.go
@@ -271,7 +271,15 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 								if resp == "" {
 									resp = "{}"
 								}
-								toolNode, _ = sjson.SetBytes(toolNode, "parts."+itoa(pp)+".functionResponse.response.result", []byte(resp))
+								// Handle non-JSON output gracefully (matches dev branch approach)
+								if resp != "null" {
+									parsed := gjson.Parse(resp)
+									if parsed.Type == gjson.JSON {
+										toolNode, _ = sjson.SetRawBytes(toolNode, "parts."+itoa(pp)+".functionResponse.response.result", []byte(parsed.Raw))
+									} else {
+										toolNode, _ = sjson.SetBytes(toolNode, "parts."+itoa(pp)+".functionResponse.response.result", resp)
+									}
+								}
 								pp++
 							}
 						}
--- a/internal/translator/gemini/openai/responses/gemini_openai-responses_request.go
+++ b/internal/translator/gemini/openai/responses/gemini_openai-responses_request.go
@@ -33,7 +33,83 @@ func ConvertOpenAIResponsesRequestToGemini(modelName string, inputRawJSON []byte

 	// Convert input messages to Gemini contents format
 	if input := root.Get("input"); input.Exists() && input.IsArray() {
-		input.ForEach(func(_, item gjson.Result) bool {
+		items := input.Array()
+
+		// Normalize consecutive function calls and outputs so each call is immediately followed by its response
+		normalized := make([]gjson.Result, 0, len(items))
+		for i := 0; i < len(items); {
+			item := items[i]
+			itemType := item.Get("type").String()
+			itemRole := item.Get("role").String()
+			if itemType == "" && itemRole != "" {
+				itemType = "message"
+			}
+
+			if itemType == "function_call" {
+				var calls []gjson.Result
+				var outputs []gjson.Result
+
+				for i < len(items) {
+					next := items[i]
+					nextType := next.Get("type").String()
+					nextRole := next.Get("role").String()
+					if nextType == "" && nextRole != "" {
+						nextType = "message"
+					}
+					if nextType != "function_call" {
+						break
+					}
+					calls = append(calls, next)
+					i++
+				}
+
+				for i < len(items) {
+					next := items[i]
+					nextType := next.Get("type").String()
+					nextRole := next.Get("role").String()
+					if nextType == "" && nextRole != "" {
+						nextType = "message"
+					}
+					if nextType != "function_call_output" {
+						break
+					}
+					outputs = append(outputs, next)
+					i++
+				}
+
+				if len(calls) > 0 {
+					outputMap := make(map[string]gjson.Result, len(outputs))
+					for _, out := range outputs {
+						outputMap[out.Get("call_id").String()] = out
+					}
+					for _, call := range calls {
+						normalized = append(normalized, call)
+						callID := call.Get("call_id").String()
+						if resp, ok := outputMap[callID]; ok {
+							normalized = append(normalized, resp)
+							delete(outputMap, callID)
+						}
+					}
+					for _, out := range outputs {
+						if _, ok := outputMap[out.Get("call_id").String()]; ok {
+							normalized = append(normalized, out)
+						}
+					}
+					continue
+				}
+			}
+
+			if itemType == "function_call_output" {
+				normalized = append(normalized, item)
+				i++
+				continue
+			}
+
+			normalized = append(normalized, item)
+			i++
+		}
+
+		for _, item := range normalized {
 			itemType := item.Get("type").String()
 			itemRole := item.Get("role").String()
 			if itemType == "" && itemRole != "" {
@@ -59,7 +135,7 @@ func ConvertOpenAIResponsesRequestToGemini(modelName string, inputRawJSON []byte
 							out, _ = sjson.SetRaw(out, "system_instruction", systemInstr)
 						}
 					}
-					return true
+					continue
 				}

 				// Handle regular messages
@@ -186,7 +262,8 @@ func ConvertOpenAIResponsesRequestToGemini(modelName string, inputRawJSON []byte
 			case "function_call_output":
 				// Handle function call outputs - convert to function message with functionResponse
 				callID := item.Get("call_id").String()
-				output := item.Get("output").String()
+				// Use .Raw to preserve the JSON encoding (includes quotes for strings)
+				outputRaw := item.Get("output").Str

 				functionContent := `{"role":"function","parts":[]}`
 				functionResponse := `{"functionResponse":{"name":"","response":{}}}`
@@ -209,18 +286,19 @@ func ConvertOpenAIResponsesRequestToGemini(modelName string, inputRawJSON []byte

 				functionResponse, _ = sjson.Set(functionResponse, "functionResponse.name", functionName)

-				// Parse output JSON string and set as response content
-				if output != "" {
-					outputResult := gjson.Parse(output)
-					functionResponse, _ = sjson.Set(functionResponse, "functionResponse.response.result", outputResult.Raw)
+				// Set the raw JSON output directly (preserves string encoding)
+				if outputRaw != "" && outputRaw != "null" {
+					output := gjson.Parse(outputRaw)
+					if output.Type == gjson.JSON {
+						functionResponse, _ = sjson.SetRaw(functionResponse, "functionResponse.response.result", output.Raw)
+					} else {
+						functionResponse, _ = sjson.Set(functionResponse, "functionResponse.response.result", outputRaw)
+					}
 				}
-
 				functionContent, _ = sjson.SetRaw(functionContent, "parts.-1", functionResponse)
 				out, _ = sjson.SetRaw(out, "contents.-1", functionContent)
 			}
-
-			return true
-		})
+		}
 	} else if input.Exists() && input.Type == gjson.String {
 		// Simple string input conversion to user message
 		userContent := `{"role":"user","parts":[{"text":""}]}`
--- a/internal/translator/openai/openai/responses/openai_openai-responses_request.go
+++ b/internal/translator/openai/openai/responses/openai_openai-responses_request.go
@@ -202,6 +202,8 @@ func ConvertOpenAIResponsesRequestToOpenAIChatCompletions(modelName string, inpu
 			out, _ = sjson.Set(out, "reasoning_effort", "medium")
 		case "high":
 			out, _ = sjson.Set(out, "reasoning_effort", "high")
+		case "xhigh":
+			out, _ = sjson.Set(out, "reasoning_effort", "xhigh")
 		default:
 			out, _ = sjson.Set(out, "reasoning_effort", "auto")
 		}
Author	SHA1	Message	Date
Luis Pater	72c7ef7647	fix(translator): handle non-JSON output parsing for OpenAI function responses - Updated `antigravity_openai_request.go` to process non-JSON outputs gracefully by verifying and distinguishing between JSON and plain string formats. - Ensured proper assignment of parsed or raw response to `functionResponse`.	2025-11-27 16:18:49 +08:00
Luis Pater	d2e4639b2a	feat(registry): add context length and update max tokens for Claude model configurations - Added `ContextLength` field with a value of 200,000 to all applicable Claude model definitions. - Standardized `MaxCompletionTokens` values across models for consistency and alignment.	2025-11-27 16:13:25 +08:00
Luis Pater	08321223c4	Merge pull request #340 from nestharus/fix/339-thinking-openai-gemini-compat fix(thinking): resolve OpenAI/Gemini compatibility for thinking model…	2025-11-27 16:03:24 +08:00
Luis Pater	7e30157590	Fixed: #354 fix(translator): add support for "xhigh" reasoning effort in OpenAI responses - Updated handling in `openai_openai-responses_request.go` to include the new "xhigh" reasoning effort level.	2025-11-27 15:59:15 +08:00
nestharus	e73cdf5cff	fix(claude): ensure max_tokens exceeds thinking budget for thinking models Fixes an issue where Claude thinking models would return 400 errors when the thinking.budget_tokens was greater than or equal to max_tokens. Changes: - Add MaxCompletionTokens: 128000 to all Claude thinking model definitions - Add ensureMaxTokensForThinking() function in claude_executor.go that: - Checks if thinking is enabled with a budget_tokens value - Looks up the model's MaxCompletionTokens from the registry - Ensures max_tokens is set to at least the model's MaxCompletionTokens - Falls back to budget_tokens + 4000 buffer if registry lookup fails This ensures Anthropic API constraint (max_tokens > thinking.budget_tokens) is always satisfied when using extended thinking features. Fixes: #339 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>	2025-11-26 22:31:05 -08:00
Luis Pater	39621a0340	fix(translator): normalize function calls and outputs for consistent input processing - Implemented logic to pair consecutive function calls and their outputs, ensuring proper sequencing for processing. - Adjusted `gemini_openai-responses_request.go` to normalize message structures and maintain expected flow.	2025-11-27 10:25:45 +08:00
Luis Pater	346b663079	fix(translator): handle non-JSON output gracefully in function call outputs - Updated handling of `output` in `gemini_openai-responses_request.go` to use `.Str` instead of `.Raw` when parsing non-JSON string outputs. - Added checks to distinguish between JSON and non-JSON `output` types for accurate `functionResponse` construction.	2025-11-27 09:40:00 +08:00
Luis Pater	0bcae68c6c	fix(translator): preserve raw JSON encoding in function call outputs - Updated handling of `output` in `gemini_openai-responses_request.go` to use `.Raw` instead of `.String` for preserving original JSON encoding. - Ensured proper setting of raw JSON output when constructing `functionResponse`.	2025-11-27 08:26:53 +08:00