Merge remote-tracking branch 'upstream/main'

internal/translator/antigravity/claude/antigravity_claude_request.go

@@ -22,6 +22,14 @@ import (
 // deriveSessionID generates a stable session ID from the request.
 // Uses the hash of the first user message to identify the conversation.
 func deriveSessionID(rawJSON []byte) string {
+	userIDResult := gjson.GetBytes(rawJSON, "metadata.user_id")
+	if userIDResult.Exists() {
+		userID := userIDResult.String()
+		idx := strings.Index(userID, "session_")
+		if idx != -1 {
+			return userID[idx+8:]
+		}
+	}
 	messages := gjson.GetBytes(rawJSON, "messages")
 	if !messages.IsArray() {
 		return ""
@@ -61,6 +69,7 @@ func deriveSessionID(rawJSON []byte) string {
 // Returns:
 //   - []byte: The transformed request data in Gemini CLI API format
 func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _ bool) []byte {
+	enableThoughtTranslate := true
 	rawJSON := bytes.Clone(inputRawJSON)
 
 	// Derive session ID for signature caching
@@ -124,25 +133,32 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 					if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "thinking" {
 						// Use GetThinkingText to handle wrapped thinking objects
 						thinkingText := thinking.GetThinkingText(contentResult)
-						signatureResult := contentResult.Get("signature")
-						clientSignature := ""
-						if signatureResult.Exists() && signatureResult.String() != "" {
-							clientSignature = signatureResult.String()
-						}
 
 						// Always try cached signature first (more reliable than client-provided)
 						// Client may send stale or invalid signatures from different sessions
 						signature := ""
 						if sessionID != "" && thinkingText != "" {
-							if cachedSig := cache.GetCachedSignature(sessionID, thinkingText); cachedSig != "" {
+							if cachedSig := cache.GetCachedSignature(modelName, sessionID, thinkingText); cachedSig != "" {
 								signature = cachedSig
 								// log.Debugf("Using cached signature for thinking block")
 							}
 						}
 
 						// Fallback to client signature only if cache miss and client signature is valid
-						if signature == "" && cache.HasValidSignature(clientSignature) {
-							signature = clientSignature
+						if signature == "" {
+							signatureResult := contentResult.Get("signature")
+							clientSignature := ""
+							if signatureResult.Exists() && signatureResult.String() != "" {
+								arrayClientSignatures := strings.SplitN(signatureResult.String(), "#", 2)
+								if len(arrayClientSignatures) == 2 {
+									if modelName == arrayClientSignatures[0] {
+										clientSignature = arrayClientSignatures[1]
+									}
+								}
+							}
+							if cache.HasValidSignature(clientSignature) {
+								signature = clientSignature
+							}
 							// log.Debugf("Using client-provided signature for thinking block")
 						}
 
@@ -159,6 +175,7 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 						// Converting to text would break this requirement
 						if isUnsigned {
 							// log.Debugf("Dropping unsigned thinking block (no valid signature)")
+							enableThoughtTranslate = false
 							continue
 						}
 
@@ -386,7 +403,7 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 	}
 
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when type==enabled
-	if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() {
+	if t := gjson.GetBytes(rawJSON, "thinking"); enableThoughtTranslate && t.Exists() && t.IsObject() {
 		if t.Get("type").String() == "enabled" {
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())

internal/translator/antigravity/claude/antigravity_claude_request_test.go

@@ -4,6 +4,7 @@ import (
 	"strings"
 	"testing"
 
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/cache"
 	"github.com/tidwall/gjson"
 )
 
@@ -75,28 +76,42 @@ func TestConvertClaudeRequestToAntigravity_RoleMapping(t *testing.T) {
 func TestConvertClaudeRequestToAntigravity_ThinkingBlocks(t *testing.T) {
 	// Valid signature must be at least 50 characters
 	validSignature := "abc123validSignature1234567890123456789012345678901234567890"
+	thinkingText := "Let me think..."
+
+	// Pre-cache the signature (simulating a response from the same session)
+	// The session ID is derived from the first user message hash
+	// Since there's no user message in this test, we need to add one
 	inputJSON := []byte(`{
 		"model": "claude-sonnet-4-5-thinking",
 		"messages": [
+			{
+				"role": "user",
+				"content": [{"type": "text", "text": "Test user message"}]
+			},
 			{
 				"role": "assistant",
 				"content": [
-					{"type": "thinking", "thinking": "Let me think...", "signature": "` + validSignature + `"},
+					{"type": "thinking", "thinking": "` + thinkingText + `", "signature": "` + validSignature + `"},
 					{"type": "text", "text": "Answer"}
 				]
 			}
 		]
 	}`)
 
+	// Derive session ID and cache the signature
+	sessionID := deriveSessionID(inputJSON)
+	cache.CacheSignature(sessionID, thinkingText, validSignature)
+	defer cache.ClearSignatureCache(sessionID)
+
 	output := ConvertClaudeRequestToAntigravity("claude-sonnet-4-5-thinking", inputJSON, false)
 	outputStr := string(output)
 
-	// Check thinking block conversion
-	firstPart := gjson.Get(outputStr, "request.contents.0.parts.0")
+	// Check thinking block conversion (now in contents.1 due to user message)
+	firstPart := gjson.Get(outputStr, "request.contents.1.parts.0")
 	if !firstPart.Get("thought").Bool() {
 		t.Error("thinking block should have thought: true")
 	}
-	if firstPart.Get("text").String() != "Let me think..." {
+	if firstPart.Get("text").String() != thinkingText {
 		t.Error("thinking text mismatch")
 	}
 	if firstPart.Get("thoughtSignature").String() != validSignature {
@@ -227,13 +242,19 @@ func TestConvertClaudeRequestToAntigravity_ToolUse(t *testing.T) {
 
 func TestConvertClaudeRequestToAntigravity_ToolUse_WithSignature(t *testing.T) {
 	validSignature := "abc123validSignature1234567890123456789012345678901234567890"
+	thinkingText := "Let me think..."
+
 	inputJSON := []byte(`{
 		"model": "claude-sonnet-4-5-thinking",
 		"messages": [
+			{
+				"role": "user",
+				"content": [{"type": "text", "text": "Test user message"}]
+			},
 			{
 				"role": "assistant",
 				"content": [
-					{"type": "thinking", "thinking": "Let me think...", "signature": "` + validSignature + `"},
+					{"type": "thinking", "thinking": "` + thinkingText + `", "signature": "` + validSignature + `"},
 					{
 						"type": "tool_use",
 						"id": "call_123",
@@ -245,11 +266,16 @@ func TestConvertClaudeRequestToAntigravity_ToolUse_WithSignature(t *testing.T) {
 		]
 	}`)
 
+	// Derive session ID and cache the signature
+	sessionID := deriveSessionID(inputJSON)
+	cache.CacheSignature(sessionID, thinkingText, validSignature)
+	defer cache.ClearSignatureCache(sessionID)
+
 	output := ConvertClaudeRequestToAntigravity("claude-sonnet-4-5-thinking", inputJSON, false)
 	outputStr := string(output)
 
-	// Check function call has the signature from the preceding thinking block
-	part := gjson.Get(outputStr, "request.contents.0.parts.1")
+	// Check function call has the signature from the preceding thinking block (now in contents.1)
+	part := gjson.Get(outputStr, "request.contents.1.parts.1")
 	if part.Get("functionCall.name").String() != "get_weather" {
 		t.Errorf("Expected functionCall, got %s", part.Raw)
 	}
@@ -261,24 +287,35 @@ func TestConvertClaudeRequestToAntigravity_ToolUse_WithSignature(t *testing.T) {
 func TestConvertClaudeRequestToAntigravity_ReorderThinking(t *testing.T) {
 	// Case: text block followed by thinking block -> should be reordered to thinking first
 	validSignature := "abc123validSignature1234567890123456789012345678901234567890"
+	thinkingText := "Planning..."
+
 	inputJSON := []byte(`{
 		"model": "claude-sonnet-4-5-thinking",
 		"messages": [
+			{
+				"role": "user",
+				"content": [{"type": "text", "text": "Test user message"}]
+			},
 			{
 				"role": "assistant",
 				"content": [
 					{"type": "text", "text": "Here is the plan."},
-					{"type": "thinking", "thinking": "Planning...", "signature": "` + validSignature + `"}
+					{"type": "thinking", "thinking": "` + thinkingText + `", "signature": "` + validSignature + `"}
 				]
 			}
 		]
 	}`)
 
+	// Derive session ID and cache the signature
+	sessionID := deriveSessionID(inputJSON)
+	cache.CacheSignature(sessionID, thinkingText, validSignature)
+	defer cache.ClearSignatureCache(sessionID)
+
 	output := ConvertClaudeRequestToAntigravity("claude-sonnet-4-5-thinking", inputJSON, false)
 	outputStr := string(output)
 
-	// Verify order: Thinking block MUST be first
-	parts := gjson.Get(outputStr, "request.contents.0.parts").Array()
+	// Verify order: Thinking block MUST be first (now in contents.1 due to user message)
+	parts := gjson.Get(outputStr, "request.contents.1.parts").Array()
 	if len(parts) != 2 {
 		t.Fatalf("Expected 2 parts, got %d", len(parts))
 	}
@@ -460,6 +497,9 @@ func TestConvertClaudeRequestToAntigravity_TrailingUnsignedThinking_Removed(t *t
 
 func TestConvertClaudeRequestToAntigravity_TrailingSignedThinking_Kept(t *testing.T) {
 	// Last assistant message ends with signed thinking block - should be kept
+	validSignature := "abc123validSignature1234567890123456789012345678901234567890"
+	thinkingText := "Valid thinking..."
+
 	inputJSON := []byte(`{
 		"model": "claude-sonnet-4-5-thinking",
 		"messages": [
@@ -471,12 +511,17 @@ func TestConvertClaudeRequestToAntigravity_TrailingSignedThinking_Kept(t *testin
 				"role": "assistant",
 				"content": [
 					{"type": "text", "text": "Here is my answer"},
-					{"type": "thinking", "thinking": "Valid thinking...", "signature": "abc123validSignature1234567890123456789012345678901234567890"}
+					{"type": "thinking", "thinking": "` + thinkingText + `", "signature": "` + validSignature + `"}
 				]
 			}
 		]
 	}`)
 
+	// Derive session ID and cache the signature
+	sessionID := deriveSessionID(inputJSON)
+	cache.CacheSignature(sessionID, thinkingText, validSignature)
+	defer cache.ClearSignatureCache(sessionID)
+
 	output := ConvertClaudeRequestToAntigravity("claude-sonnet-4-5-thinking", inputJSON, false)
 	outputStr := string(output)
 

internal/translator/antigravity/claude/antigravity_claude_response.go

@@ -73,6 +73,7 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 			SessionID:        deriveSessionID(originalRequestRawJSON),
 		}
 	}
+	modelName := gjson.GetBytes(requestRawJSON, "model").String()
 
 	params := (*param).(*Params)
 
@@ -139,13 +140,13 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 						// log.Debug("Branch: signature_delta")
 
 						if params.SessionID != "" && params.CurrentThinkingText.Len() > 0 {
-							cache.CacheSignature(params.SessionID, params.CurrentThinkingText.String(), thoughtSignature.String())
+							cache.CacheSignature(modelName, params.SessionID, params.CurrentThinkingText.String(), thoughtSignature.String())
 							// log.Debugf("Cached signature for thinking block (sessionID=%s, textLen=%d)", params.SessionID, params.CurrentThinkingText.Len())
 							params.CurrentThinkingText.Reset()
 						}
 
 						output = output + "event: content_block_delta\n"
-						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"signature_delta","signature":""}}`, params.ResponseIndex), "delta.signature", thoughtSignature.String())
+						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"signature_delta","signature":""}}`, params.ResponseIndex), "delta.signature", fmt.Sprintf("%s#%s", cache.GetModelGroup(modelName), thoughtSignature.String()))
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						params.HasContent = true
 					} else if params.ResponseType == 2 { // Continue existing thinking block if already in thinking state
@@ -372,7 +373,7 @@ func resolveStopReason(params *Params) string {
 //   - string: A Claude-compatible JSON response.
 func ConvertAntigravityResponseToClaudeNonStream(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) string {
 	_ = originalRequestRawJSON
-	_ = requestRawJSON
+	modelName := gjson.GetBytes(requestRawJSON, "model").String()
 
 	root := gjson.ParseBytes(rawJSON)
 	promptTokens := root.Get("response.usageMetadata.promptTokenCount").Int()
@@ -437,7 +438,7 @@ func ConvertAntigravityResponseToClaudeNonStream(_ context.Context, _ string, or
 		block := `{"type":"thinking","thinking":""}`
 		block, _ = sjson.Set(block, "thinking", thinkingBuilder.String())
 		if thinkingSignature != "" {
-			block, _ = sjson.Set(block, "signature", thinkingSignature)
+			block, _ = sjson.Set(block, "signature", fmt.Sprintf("%s#%s", cache.GetModelGroup(modelName), thinkingSignature))
 		}
 		responseJSON, _ = sjson.SetRaw(responseJSON, "content.-1", block)
 		thinkingBuilder.Reset()

internal/translator/antigravity/openai/chat-completions/antigravity_openai_request.go

@@ -66,6 +66,13 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 		out, _ = sjson.SetBytes(out, "request.generationConfig.maxOutputTokens", maxTok.Num)
 	}
 
+	// Candidate count (OpenAI 'n' parameter)
+	if n := gjson.GetBytes(rawJSON, "n"); n.Exists() && n.Type == gjson.Number {
+		if val := n.Int(); val > 1 {
+			out, _ = sjson.SetBytes(out, "request.generationConfig.candidateCount", val)
+		}
+	}
+
 	// Map OpenAI modalities -> Gemini CLI request.generationConfig.responseModalities
 	// e.g. "modalities": ["image", "text"] -> ["IMAGE", "TEXT"]
 	if mods := gjson.GetBytes(rawJSON, "modalities"); mods.Exists() && mods.IsArray() {

internal/translator/codex/claude/codex_claude_response.go

@@ -117,8 +117,12 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 		} else {
 			template, _ = sjson.Set(template, "delta.stop_reason", "end_turn")
 		}
-		template, _ = sjson.Set(template, "usage.input_tokens", rootResult.Get("response.usage.input_tokens").Int())
-		template, _ = sjson.Set(template, "usage.output_tokens", rootResult.Get("response.usage.output_tokens").Int())
+		inputTokens, outputTokens, cachedTokens := extractResponsesUsage(rootResult.Get("response.usage"))
+		template, _ = sjson.Set(template, "usage.input_tokens", inputTokens)
+		template, _ = sjson.Set(template, "usage.output_tokens", outputTokens)
+		if cachedTokens > 0 {
+			template, _ = sjson.Set(template, "usage.cache_read_input_tokens", cachedTokens)
+		}
 
 		output = "event: message_delta\n"
 		output += fmt.Sprintf("data: %s\n\n", template)
@@ -204,8 +208,12 @@ func ConvertCodexResponseToClaudeNonStream(_ context.Context, _ string, original
 	out := `{"id":"","type":"message","role":"assistant","model":"","content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":0,"output_tokens":0}}`
 	out, _ = sjson.Set(out, "id", responseData.Get("id").String())
 	out, _ = sjson.Set(out, "model", responseData.Get("model").String())
-	out, _ = sjson.Set(out, "usage.input_tokens", responseData.Get("usage.input_tokens").Int())
-	out, _ = sjson.Set(out, "usage.output_tokens", responseData.Get("usage.output_tokens").Int())
+	inputTokens, outputTokens, cachedTokens := extractResponsesUsage(responseData.Get("usage"))
+	out, _ = sjson.Set(out, "usage.input_tokens", inputTokens)
+	out, _ = sjson.Set(out, "usage.output_tokens", outputTokens)
+	if cachedTokens > 0 {
+		out, _ = sjson.Set(out, "usage.cache_read_input_tokens", cachedTokens)
+	}
 
 	hasToolCall := false
 
@@ -308,12 +316,27 @@ func ConvertCodexResponseToClaudeNonStream(_ context.Context, _ string, original
 		out, _ = sjson.SetRaw(out, "stop_sequence", stopSequence.Raw)
 	}
 
-	if responseData.Get("usage.input_tokens").Exists() || responseData.Get("usage.output_tokens").Exists() {
-		out, _ = sjson.Set(out, "usage.input_tokens", responseData.Get("usage.input_tokens").Int())
-		out, _ = sjson.Set(out, "usage.output_tokens", responseData.Get("usage.output_tokens").Int())
+	return out
+}
+
+func extractResponsesUsage(usage gjson.Result) (int64, int64, int64) {
+	if !usage.Exists() || usage.Type == gjson.Null {
+		return 0, 0, 0
 	}
 
-	return out
+	inputTokens := usage.Get("input_tokens").Int()
+	outputTokens := usage.Get("output_tokens").Int()
+	cachedTokens := usage.Get("input_tokens_details.cached_tokens").Int()
+
+	if cachedTokens > 0 {
+		if inputTokens >= cachedTokens {
+			inputTokens -= cachedTokens
+		} else {
+			inputTokens = 0
+		}
+	}
+
+	return inputTokens, outputTokens, cachedTokens
 }
 
 // buildReverseMapFromClaudeOriginalShortToOriginal builds a map[short]original from original Claude request tools.

internal/translator/gemini-cli/openai/chat-completions/gemini-cli_openai_request.go

@@ -63,6 +63,13 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 		out, _ = sjson.SetBytes(out, "request.generationConfig.topK", tkr.Num)
 	}
 
+	// Candidate count (OpenAI 'n' parameter)
+	if n := gjson.GetBytes(rawJSON, "n"); n.Exists() && n.Type == gjson.Number {
+		if val := n.Int(); val > 1 {
+			out, _ = sjson.SetBytes(out, "request.generationConfig.candidateCount", val)
+		}
+	}
+
 	// Map OpenAI modalities -> Gemini CLI request.generationConfig.responseModalities
 	// e.g. "modalities": ["image", "text"] -> ["IMAGE", "TEXT"]
 	if mods := gjson.GetBytes(rawJSON, "modalities"); mods.Exists() && mods.IsArray() {

internal/translator/gemini/openai/chat-completions/gemini_openai_request.go

@@ -63,6 +63,13 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 		out, _ = sjson.SetBytes(out, "generationConfig.topK", tkr.Num)
 	}
 
+	// Candidate count (OpenAI 'n' parameter)
+	if n := gjson.GetBytes(rawJSON, "n"); n.Exists() && n.Type == gjson.Number {
+		if val := n.Int(); val > 1 {
+			out, _ = sjson.SetBytes(out, "generationConfig.candidateCount", val)
+		}
+	}
+
 	// Map OpenAI modalities -> Gemini generationConfig.responseModalities
 	// e.g. "modalities": ["image", "text"] -> ["IMAGE", "TEXT"]
 	if mods := gjson.GetBytes(rawJSON, "modalities"); mods.Exists() && mods.IsArray() {

internal/translator/gemini/openai/chat-completions/gemini_openai_response.go

@@ -21,7 +21,8 @@ import (
 // convertGeminiResponseToOpenAIChatParams holds parameters for response conversion.
 type convertGeminiResponseToOpenAIChatParams struct {
 	UnixTimestamp int64
-	FunctionIndex int
+	// FunctionIndex tracks tool call indices per candidate index to support multiple candidates.
+	FunctionIndex map[int]int
 }
 
 // functionCallIDCounter provides a process-wide unique counter for function call identifiers.
@@ -42,13 +43,20 @@ var functionCallIDCounter uint64
 // Returns:
 //   - []string: A slice of strings, each containing an OpenAI-compatible JSON response
 func ConvertGeminiResponseToOpenAI(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) []string {
+	// Initialize parameters if nil.
 	if *param == nil {
 		*param = &convertGeminiResponseToOpenAIChatParams{
 			UnixTimestamp: 0,
-			FunctionIndex: 0,
+			FunctionIndex: make(map[int]int),
 		}
 	}
 
+	// Ensure the Map is initialized (handling cases where param might be reused from older context).
+	p := (*param).(*convertGeminiResponseToOpenAIChatParams)
+	if p.FunctionIndex == nil {
+		p.FunctionIndex = make(map[int]int)
+	}
+
 	if bytes.HasPrefix(rawJSON, []byte("data:")) {
 		rawJSON = bytes.TrimSpace(rawJSON[5:])
 	}
@@ -57,151 +65,179 @@ func ConvertGeminiResponseToOpenAI(_ context.Context, _ string, originalRequestR
 		return []string{}
 	}
 
-	// Initialize the OpenAI SSE template.
-	template := `{"id":"","object":"chat.completion.chunk","created":12345,"model":"model","choices":[{"index":0,"delta":{"role":null,"content":null,"reasoning_content":null,"tool_calls":null},"finish_reason":null,"native_finish_reason":null}]}`
+	// Initialize the OpenAI SSE base template.
+	// We use a base template and clone it for each candidate to support multiple candidates.
+	baseTemplate := `{"id":"","object":"chat.completion.chunk","created":12345,"model":"model","choices":[{"index":0,"delta":{"role":null,"content":null,"reasoning_content":null,"tool_calls":null},"finish_reason":null,"native_finish_reason":null}]}`
 
 	// Extract and set the model version.
 	if modelVersionResult := gjson.GetBytes(rawJSON, "modelVersion"); modelVersionResult.Exists() {
-		template, _ = sjson.Set(template, "model", modelVersionResult.String())
+		baseTemplate, _ = sjson.Set(baseTemplate, "model", modelVersionResult.String())
 	}
 
 	// Extract and set the creation timestamp.
 	if createTimeResult := gjson.GetBytes(rawJSON, "createTime"); createTimeResult.Exists() {
 		t, err := time.Parse(time.RFC3339Nano, createTimeResult.String())
 		if err == nil {
-			(*param).(*convertGeminiResponseToOpenAIChatParams).UnixTimestamp = t.Unix()
+			p.UnixTimestamp = t.Unix()
 		}
-		template, _ = sjson.Set(template, "created", (*param).(*convertGeminiResponseToOpenAIChatParams).UnixTimestamp)
+		baseTemplate, _ = sjson.Set(baseTemplate, "created", p.UnixTimestamp)
 	} else {
-		template, _ = sjson.Set(template, "created", (*param).(*convertGeminiResponseToOpenAIChatParams).UnixTimestamp)
+		baseTemplate, _ = sjson.Set(baseTemplate, "created", p.UnixTimestamp)
 	}
 
 	// Extract and set the response ID.
 	if responseIDResult := gjson.GetBytes(rawJSON, "responseId"); responseIDResult.Exists() {
-		template, _ = sjson.Set(template, "id", responseIDResult.String())
-	}
-
-	// Extract and set the finish reason.
-	if finishReasonResult := gjson.GetBytes(rawJSON, "candidates.0.finishReason"); finishReasonResult.Exists() {
-		template, _ = sjson.Set(template, "choices.0.finish_reason", strings.ToLower(finishReasonResult.String()))
-		template, _ = sjson.Set(template, "choices.0.native_finish_reason", strings.ToLower(finishReasonResult.String()))
+		baseTemplate, _ = sjson.Set(baseTemplate, "id", responseIDResult.String())
 	}
 
 	// Extract and set usage metadata (token counts).
+	// Usage is applied to the base template so it appears in the chunks.
 	if usageResult := gjson.GetBytes(rawJSON, "usageMetadata"); usageResult.Exists() {
 		cachedTokenCount := usageResult.Get("cachedContentTokenCount").Int()
 		if candidatesTokenCountResult := usageResult.Get("candidatesTokenCount"); candidatesTokenCountResult.Exists() {
-			template, _ = sjson.Set(template, "usage.completion_tokens", candidatesTokenCountResult.Int())
+			baseTemplate, _ = sjson.Set(baseTemplate, "usage.completion_tokens", candidatesTokenCountResult.Int())
 		}
 		if totalTokenCountResult := usageResult.Get("totalTokenCount"); totalTokenCountResult.Exists() {
-			template, _ = sjson.Set(template, "usage.total_tokens", totalTokenCountResult.Int())
+			baseTemplate, _ = sjson.Set(baseTemplate, "usage.total_tokens", totalTokenCountResult.Int())
 		}
 		promptTokenCount := usageResult.Get("promptTokenCount").Int() - cachedTokenCount
 		thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
-		template, _ = sjson.Set(template, "usage.prompt_tokens", promptTokenCount+thoughtsTokenCount)
+		baseTemplate, _ = sjson.Set(baseTemplate, "usage.prompt_tokens", promptTokenCount+thoughtsTokenCount)
 		if thoughtsTokenCount > 0 {
-			template, _ = sjson.Set(template, "usage.completion_tokens_details.reasoning_tokens", thoughtsTokenCount)
+			baseTemplate, _ = sjson.Set(baseTemplate, "usage.completion_tokens_details.reasoning_tokens", thoughtsTokenCount)
 		}
 		// Include cached token count if present (indicates prompt caching is working)
 		if cachedTokenCount > 0 {
 			var err error
-			template, err = sjson.Set(template, "usage.prompt_tokens_details.cached_tokens", cachedTokenCount)
+			baseTemplate, err = sjson.Set(baseTemplate, "usage.prompt_tokens_details.cached_tokens", cachedTokenCount)
 			if err != nil {
 				log.Warnf("gemini openai response: failed to set cached_tokens in streaming: %v", err)
 			}
 		}
 	}
 
-	// Process the main content part of the response.
-	partsResult := gjson.GetBytes(rawJSON, "candidates.0.content.parts")
-	hasFunctionCall := false
-	if partsResult.IsArray() {
-		partResults := partsResult.Array()
-		for i := 0; i < len(partResults); i++ {
-			partResult := partResults[i]
-			partTextResult := partResult.Get("text")
-			functionCallResult := partResult.Get("functionCall")
-			inlineDataResult := partResult.Get("inlineData")
-			if !inlineDataResult.Exists() {
-				inlineDataResult = partResult.Get("inline_data")
-			}
-			thoughtSignatureResult := partResult.Get("thoughtSignature")
-			if !thoughtSignatureResult.Exists() {
-				thoughtSignatureResult = partResult.Get("thought_signature")
+	var responseStrings []string
+	candidates := gjson.GetBytes(rawJSON, "candidates")
+
+	// Iterate over all candidates to support candidate_count > 1.
+	if candidates.IsArray() {
+		candidates.ForEach(func(_, candidate gjson.Result) bool {
+			// Clone the template for the current candidate.
+			template := baseTemplate
+
+			// Set the specific index for this candidate.
+			candidateIndex := int(candidate.Get("index").Int())
+			template, _ = sjson.Set(template, "choices.0.index", candidateIndex)
+
+			// Extract and set the finish reason.
+			if finishReasonResult := candidate.Get("finishReason"); finishReasonResult.Exists() {
+				template, _ = sjson.Set(template, "choices.0.finish_reason", strings.ToLower(finishReasonResult.String()))
+				template, _ = sjson.Set(template, "choices.0.native_finish_reason", strings.ToLower(finishReasonResult.String()))
 			}
 
-			hasThoughtSignature := thoughtSignatureResult.Exists() && thoughtSignatureResult.String() != ""
-			hasContentPayload := partTextResult.Exists() || functionCallResult.Exists() || inlineDataResult.Exists()
+			partsResult := candidate.Get("content.parts")
+			hasFunctionCall := false
 
-			// Skip pure thoughtSignature parts but keep any actual payload in the same part.
-			if hasThoughtSignature && !hasContentPayload {
-				continue
+			if partsResult.IsArray() {
+				partResults := partsResult.Array()
+				for i := 0; i < len(partResults); i++ {
+					partResult := partResults[i]
+					partTextResult := partResult.Get("text")
+					functionCallResult := partResult.Get("functionCall")
+					inlineDataResult := partResult.Get("inlineData")
+					if !inlineDataResult.Exists() {
+						inlineDataResult = partResult.Get("inline_data")
+					}
+					thoughtSignatureResult := partResult.Get("thoughtSignature")
+					if !thoughtSignatureResult.Exists() {
+						thoughtSignatureResult = partResult.Get("thought_signature")
+					}
+
+					hasThoughtSignature := thoughtSignatureResult.Exists() && thoughtSignatureResult.String() != ""
+					hasContentPayload := partTextResult.Exists() || functionCallResult.Exists() || inlineDataResult.Exists()
+
+					// Skip pure thoughtSignature parts but keep any actual payload in the same part.
+					if hasThoughtSignature && !hasContentPayload {
+						continue
+					}
+
+					if partTextResult.Exists() {
+						text := partTextResult.String()
+						// Handle text content, distinguishing between regular content and reasoning/thoughts.
+						if partResult.Get("thought").Bool() {
+							template, _ = sjson.Set(template, "choices.0.delta.reasoning_content", text)
+						} else {
+							template, _ = sjson.Set(template, "choices.0.delta.content", text)
+						}
+						template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
+					} else if functionCallResult.Exists() {
+						// Handle function call content.
+						hasFunctionCall = true
+						toolCallsResult := gjson.Get(template, "choices.0.delta.tool_calls")
+
+						// Retrieve the function index for this specific candidate.
+						functionCallIndex := p.FunctionIndex[candidateIndex]
+						p.FunctionIndex[candidateIndex]++
+
+						if toolCallsResult.Exists() && toolCallsResult.IsArray() {
+							functionCallIndex = len(toolCallsResult.Array())
+						} else {
+							template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
+						}
+
+						functionCallTemplate := `{"id": "","index": 0,"type": "function","function": {"name": "","arguments": ""}}`
+						fcName := functionCallResult.Get("name").String()
+						functionCallTemplate, _ = sjson.Set(functionCallTemplate, "id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&functionCallIDCounter, 1)))
+						functionCallTemplate, _ = sjson.Set(functionCallTemplate, "index", functionCallIndex)
+						functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.name", fcName)
+						if fcArgsResult := functionCallResult.Get("args"); fcArgsResult.Exists() {
+							functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.arguments", fcArgsResult.Raw)
+						}
+						template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
+						template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallTemplate)
+					} else if inlineDataResult.Exists() {
+						data := inlineDataResult.Get("data").String()
+						if data == "" {
+							continue
+						}
+						mimeType := inlineDataResult.Get("mimeType").String()
+						if mimeType == "" {
+							mimeType = inlineDataResult.Get("mime_type").String()
+						}
+						if mimeType == "" {
+							mimeType = "image/png"
+						}
+						imageURL := fmt.Sprintf("data:%s;base64,%s", mimeType, data)
+						imagesResult := gjson.Get(template, "choices.0.delta.images")
+						if !imagesResult.Exists() || !imagesResult.IsArray() {
+							template, _ = sjson.SetRaw(template, "choices.0.delta.images", `[]`)
+						}
+						imageIndex := len(gjson.Get(template, "choices.0.delta.images").Array())
+						imagePayload := `{"type":"image_url","image_url":{"url":""}}`
+						imagePayload, _ = sjson.Set(imagePayload, "index", imageIndex)
+						imagePayload, _ = sjson.Set(imagePayload, "image_url.url", imageURL)
+						template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
+						template, _ = sjson.SetRaw(template, "choices.0.delta.images.-1", imagePayload)
+					}
+				}
 			}
 
-			if partTextResult.Exists() {
-				text := partTextResult.String()
-				// Handle text content, distinguishing between regular content and reasoning/thoughts.
-				if partResult.Get("thought").Bool() {
-					template, _ = sjson.Set(template, "choices.0.delta.reasoning_content", text)
-				} else {
-					template, _ = sjson.Set(template, "choices.0.delta.content", text)
-				}
-				template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
-			} else if functionCallResult.Exists() {
-				// Handle function call content.
-				hasFunctionCall = true
-				toolCallsResult := gjson.Get(template, "choices.0.delta.tool_calls")
-				functionCallIndex := (*param).(*convertGeminiResponseToOpenAIChatParams).FunctionIndex
-				(*param).(*convertGeminiResponseToOpenAIChatParams).FunctionIndex++
-				if toolCallsResult.Exists() && toolCallsResult.IsArray() {
-					functionCallIndex = len(toolCallsResult.Array())
-				} else {
-					template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
-				}
-
-				functionCallTemplate := `{"id": "","index": 0,"type": "function","function": {"name": "","arguments": ""}}`
-				fcName := functionCallResult.Get("name").String()
-				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&functionCallIDCounter, 1)))
-				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "index", functionCallIndex)
-				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.name", fcName)
-				if fcArgsResult := functionCallResult.Get("args"); fcArgsResult.Exists() {
-					functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.arguments", fcArgsResult.Raw)
-				}
-				template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
-				template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallTemplate)
-			} else if inlineDataResult.Exists() {
-				data := inlineDataResult.Get("data").String()
-				if data == "" {
-					continue
-				}
-				mimeType := inlineDataResult.Get("mimeType").String()
-				if mimeType == "" {
-					mimeType = inlineDataResult.Get("mime_type").String()
-				}
-				if mimeType == "" {
-					mimeType = "image/png"
-				}
-				imageURL := fmt.Sprintf("data:%s;base64,%s", mimeType, data)
-				imagesResult := gjson.Get(template, "choices.0.delta.images")
-				if !imagesResult.Exists() || !imagesResult.IsArray() {
-					template, _ = sjson.SetRaw(template, "choices.0.delta.images", `[]`)
-				}
-				imageIndex := len(gjson.Get(template, "choices.0.delta.images").Array())
-				imagePayload := `{"type":"image_url","image_url":{"url":""}}`
-				imagePayload, _ = sjson.Set(imagePayload, "index", imageIndex)
-				imagePayload, _ = sjson.Set(imagePayload, "image_url.url", imageURL)
-				template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
-				template, _ = sjson.SetRaw(template, "choices.0.delta.images.-1", imagePayload)
+			if hasFunctionCall {
+				template, _ = sjson.Set(template, "choices.0.finish_reason", "tool_calls")
+				template, _ = sjson.Set(template, "choices.0.native_finish_reason", "tool_calls")
 			}
+
+			responseStrings = append(responseStrings, template)
+			return true // continue loop
+		})
+	} else {
+		// If there are no candidates (e.g., a pure usageMetadata chunk), return the usage chunk if present.
+		if gjson.GetBytes(rawJSON, "usageMetadata").Exists() && len(responseStrings) == 0 {
+			responseStrings = append(responseStrings, baseTemplate)
 		}
 	}
 
-	if hasFunctionCall {
-		template, _ = sjson.Set(template, "choices.0.finish_reason", "tool_calls")
-		template, _ = sjson.Set(template, "choices.0.native_finish_reason", "tool_calls")
-	}
-
-	return []string{template}
+	return responseStrings
 }
 
 // ConvertGeminiResponseToOpenAINonStream converts a non-streaming Gemini response to a non-streaming OpenAI response.
@@ -219,7 +255,9 @@ func ConvertGeminiResponseToOpenAI(_ context.Context, _ string, originalRequestR
 //   - string: An OpenAI-compatible JSON response containing all message content and metadata
 func ConvertGeminiResponseToOpenAINonStream(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) string {
 	var unixTimestamp int64
-	template := `{"id":"","object":"chat.completion","created":123456,"model":"model","choices":[{"index":0,"message":{"role":"assistant","content":null,"reasoning_content":null,"tool_calls":null},"finish_reason":null,"native_finish_reason":null}]}`
+	// Initialize template with an empty choices array to support multiple candidates.
+	template := `{"id":"","object":"chat.completion","created":123456,"model":"model","choices":[]}`
+
 	if modelVersionResult := gjson.GetBytes(rawJSON, "modelVersion"); modelVersionResult.Exists() {
 		template, _ = sjson.Set(template, "model", modelVersionResult.String())
 	}
@@ -238,11 +276,6 @@ func ConvertGeminiResponseToOpenAINonStream(_ context.Context, _ string, origina
 		template, _ = sjson.Set(template, "id", responseIDResult.String())
 	}
 
-	if finishReasonResult := gjson.GetBytes(rawJSON, "candidates.0.finishReason"); finishReasonResult.Exists() {
-		template, _ = sjson.Set(template, "choices.0.finish_reason", strings.ToLower(finishReasonResult.String()))
-		template, _ = sjson.Set(template, "choices.0.native_finish_reason", strings.ToLower(finishReasonResult.String()))
-	}
-
 	if usageResult := gjson.GetBytes(rawJSON, "usageMetadata"); usageResult.Exists() {
 		if candidatesTokenCountResult := usageResult.Get("candidatesTokenCount"); candidatesTokenCountResult.Exists() {
 			template, _ = sjson.Set(template, "usage.completion_tokens", candidatesTokenCountResult.Int())
@@ -267,74 +300,96 @@ func ConvertGeminiResponseToOpenAINonStream(_ context.Context, _ string, origina
 		}
 	}
 
-	// Process the main content part of the response.
-	partsResult := gjson.GetBytes(rawJSON, "candidates.0.content.parts")
-	hasFunctionCall := false
-	if partsResult.IsArray() {
-		partsResults := partsResult.Array()
-		for i := 0; i < len(partsResults); i++ {
-			partResult := partsResults[i]
-			partTextResult := partResult.Get("text")
-			functionCallResult := partResult.Get("functionCall")
-			inlineDataResult := partResult.Get("inlineData")
-			if !inlineDataResult.Exists() {
-				inlineDataResult = partResult.Get("inline_data")
+	// Process the main content part of the response for all candidates.
+	candidates := gjson.GetBytes(rawJSON, "candidates")
+	if candidates.IsArray() {
+		candidates.ForEach(func(_, candidate gjson.Result) bool {
+			// Construct a single Choice object.
+			choiceTemplate := `{"index":0,"message":{"role":"assistant","content":null,"reasoning_content":null,"tool_calls":null},"finish_reason":null,"native_finish_reason":null}`
+
+			// Set the index for this choice.
+			choiceTemplate, _ = sjson.Set(choiceTemplate, "index", candidate.Get("index").Int())
+
+			// Set finish reason.
+			if finishReasonResult := candidate.Get("finishReason"); finishReasonResult.Exists() {
+				choiceTemplate, _ = sjson.Set(choiceTemplate, "finish_reason", strings.ToLower(finishReasonResult.String()))
+				choiceTemplate, _ = sjson.Set(choiceTemplate, "native_finish_reason", strings.ToLower(finishReasonResult.String()))
 			}
 
-			if partTextResult.Exists() {
-				// Append text content, distinguishing between regular content and reasoning.
-				if partResult.Get("thought").Bool() {
-					template, _ = sjson.Set(template, "choices.0.message.reasoning_content", partTextResult.String())
-				} else {
-					template, _ = sjson.Set(template, "choices.0.message.content", partTextResult.String())
-				}
-				template, _ = sjson.Set(template, "choices.0.message.role", "assistant")
-			} else if functionCallResult.Exists() {
-				// Append function call content to the tool_calls array.
-				hasFunctionCall = true
-				toolCallsResult := gjson.Get(template, "choices.0.message.tool_calls")
-				if !toolCallsResult.Exists() || !toolCallsResult.IsArray() {
-					template, _ = sjson.SetRaw(template, "choices.0.message.tool_calls", `[]`)
-				}
-				functionCallItemTemplate := `{"id": "","type": "function","function": {"name": "","arguments": ""}}`
-				fcName := functionCallResult.Get("name").String()
-				functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&functionCallIDCounter, 1)))
-				functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.name", fcName)
-				if fcArgsResult := functionCallResult.Get("args"); fcArgsResult.Exists() {
-					functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.arguments", fcArgsResult.Raw)
-				}
-				template, _ = sjson.Set(template, "choices.0.message.role", "assistant")
-				template, _ = sjson.SetRaw(template, "choices.0.message.tool_calls.-1", functionCallItemTemplate)
-			} else if inlineDataResult.Exists() {
-				data := inlineDataResult.Get("data").String()
-				if data == "" {
-					continue
-				}
-				mimeType := inlineDataResult.Get("mimeType").String()
-				if mimeType == "" {
-					mimeType = inlineDataResult.Get("mime_type").String()
-				}
-				if mimeType == "" {
-					mimeType = "image/png"
-				}
-				imageURL := fmt.Sprintf("data:%s;base64,%s", mimeType, data)
-				imagesResult := gjson.Get(template, "choices.0.message.images")
-				if !imagesResult.Exists() || !imagesResult.IsArray() {
-					template, _ = sjson.SetRaw(template, "choices.0.message.images", `[]`)
-				}
-				imageIndex := len(gjson.Get(template, "choices.0.message.images").Array())
-				imagePayload := `{"type":"image_url","image_url":{"url":""}}`
-				imagePayload, _ = sjson.Set(imagePayload, "index", imageIndex)
-				imagePayload, _ = sjson.Set(imagePayload, "image_url.url", imageURL)
-				template, _ = sjson.Set(template, "choices.0.message.role", "assistant")
-				template, _ = sjson.SetRaw(template, "choices.0.message.images.-1", imagePayload)
-			}
-		}
-	}
+			partsResult := candidate.Get("content.parts")
+			hasFunctionCall := false
+			if partsResult.IsArray() {
+				partsResults := partsResult.Array()
+				for i := 0; i < len(partsResults); i++ {
+					partResult := partsResults[i]
+					partTextResult := partResult.Get("text")
+					functionCallResult := partResult.Get("functionCall")
+					inlineDataResult := partResult.Get("inlineData")
+					if !inlineDataResult.Exists() {
+						inlineDataResult = partResult.Get("inline_data")
+					}
 
-	if hasFunctionCall {
-		template, _ = sjson.Set(template, "choices.0.finish_reason", "tool_calls")
-		template, _ = sjson.Set(template, "choices.0.native_finish_reason", "tool_calls")
+					if partTextResult.Exists() {
+						// Append text content, distinguishing between regular content and reasoning.
+						if partResult.Get("thought").Bool() {
+							oldVal := gjson.Get(choiceTemplate, "message.reasoning_content").String()
+							choiceTemplate, _ = sjson.Set(choiceTemplate, "message.reasoning_content", oldVal+partTextResult.String())
+						} else {
+							oldVal := gjson.Get(choiceTemplate, "message.content").String()
+							choiceTemplate, _ = sjson.Set(choiceTemplate, "message.content", oldVal+partTextResult.String())
+						}
+						choiceTemplate, _ = sjson.Set(choiceTemplate, "message.role", "assistant")
+					} else if functionCallResult.Exists() {
+						// Append function call content to the tool_calls array.
+						hasFunctionCall = true
+						toolCallsResult := gjson.Get(choiceTemplate, "message.tool_calls")
+						if !toolCallsResult.Exists() || !toolCallsResult.IsArray() {
+							choiceTemplate, _ = sjson.SetRaw(choiceTemplate, "message.tool_calls", `[]`)
+						}
+						functionCallItemTemplate := `{"id": "","type": "function","function": {"name": "","arguments": ""}}`
+						fcName := functionCallResult.Get("name").String()
+						functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&functionCallIDCounter, 1)))
+						functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.name", fcName)
+						if fcArgsResult := functionCallResult.Get("args"); fcArgsResult.Exists() {
+							functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.arguments", fcArgsResult.Raw)
+						}
+						choiceTemplate, _ = sjson.Set(choiceTemplate, "message.role", "assistant")
+						choiceTemplate, _ = sjson.SetRaw(choiceTemplate, "message.tool_calls.-1", functionCallItemTemplate)
+					} else if inlineDataResult.Exists() {
+						data := inlineDataResult.Get("data").String()
+						if data != "" {
+							mimeType := inlineDataResult.Get("mimeType").String()
+							if mimeType == "" {
+								mimeType = inlineDataResult.Get("mime_type").String()
+							}
+							if mimeType == "" {
+								mimeType = "image/png"
+							}
+							imageURL := fmt.Sprintf("data:%s;base64,%s", mimeType, data)
+							imagesResult := gjson.Get(choiceTemplate, "message.images")
+							if !imagesResult.Exists() || !imagesResult.IsArray() {
+								choiceTemplate, _ = sjson.SetRaw(choiceTemplate, "message.images", `[]`)
+							}
+							imageIndex := len(gjson.Get(choiceTemplate, "message.images").Array())
+							imagePayload := `{"type":"image_url","image_url":{"url":""}}`
+							imagePayload, _ = sjson.Set(imagePayload, "index", imageIndex)
+							imagePayload, _ = sjson.Set(imagePayload, "image_url.url", imageURL)
+							choiceTemplate, _ = sjson.Set(choiceTemplate, "message.role", "assistant")
+							choiceTemplate, _ = sjson.SetRaw(choiceTemplate, "message.images.-1", imagePayload)
+						}
+					}
+				}
+			}
+
+			if hasFunctionCall {
+				choiceTemplate, _ = sjson.Set(choiceTemplate, "finish_reason", "tool_calls")
+				choiceTemplate, _ = sjson.Set(choiceTemplate, "native_finish_reason", "tool_calls")
+			}
+
+			// Append the constructed choice to the main choices array.
+			template, _ = sjson.SetRaw(template, "choices.-1", choiceTemplate)
+			return true
+		})
 	}
 
 	return template

internal/translator/openai/claude/openai_claude_request.go

@@ -88,7 +88,7 @@ func ConvertClaudeRequestToOpenAI(modelName string, inputRawJSON []byte, stream
 	var messagesJSON = "[]"
 
 	// Handle system message first
-	systemMsgJSON := `{"role":"system","content":[{"type":"text","text":"Use ANY tool, the parameters MUST accord with RFC 8259 (The JavaScript Object Notation (JSON) Data Interchange Format), the keys and value MUST be enclosed in double quotes."}]}`
+	systemMsgJSON := `{"role":"system","content":[]}`
 	if system := root.Get("system"); system.Exists() {
 		if system.Type == gjson.String {
 			if system.String() != "" {

internal/translator/openai/claude/openai_claude_response.go

@@ -289,21 +289,17 @@ func convertOpenAIStreamingChunkToAnthropic(rawJSON []byte, param *ConvertOpenAI
 	// Only process if usage has actual values (not null)
 	if param.FinishReason != "" {
 		usage := root.Get("usage")
-		var inputTokens, outputTokens int64
+		var inputTokens, outputTokens, cachedTokens int64
 		if usage.Exists() && usage.Type != gjson.Null {
-			// Check if usage has actual token counts
-			promptTokens := usage.Get("prompt_tokens")
-			completionTokens := usage.Get("completion_tokens")
-
-			if promptTokens.Exists() && completionTokens.Exists() {
-				inputTokens = promptTokens.Int()
-				outputTokens = completionTokens.Int()
-			}
+			inputTokens, outputTokens, cachedTokens = extractOpenAIUsage(usage)
 			// Send message_delta with usage
 			messageDeltaJSON := `{"type":"message_delta","delta":{"stop_reason":"","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
 			messageDeltaJSON, _ = sjson.Set(messageDeltaJSON, "delta.stop_reason", mapOpenAIFinishReasonToAnthropic(param.FinishReason))
 			messageDeltaJSON, _ = sjson.Set(messageDeltaJSON, "usage.input_tokens", inputTokens)
 			messageDeltaJSON, _ = sjson.Set(messageDeltaJSON, "usage.output_tokens", outputTokens)
+			if cachedTokens > 0 {
+				messageDeltaJSON, _ = sjson.Set(messageDeltaJSON, "usage.cache_read_input_tokens", cachedTokens)
+			}
 			results = append(results, "event: message_delta\ndata: "+messageDeltaJSON+"\n\n")
 			param.MessageDeltaSent = true
 
@@ -423,13 +419,12 @@ func convertOpenAINonStreamingToAnthropic(rawJSON []byte) []string {
 
 	// Set usage information
 	if usage := root.Get("usage"); usage.Exists() {
-		out, _ = sjson.Set(out, "usage.input_tokens", usage.Get("prompt_tokens").Int())
-		out, _ = sjson.Set(out, "usage.output_tokens", usage.Get("completion_tokens").Int())
-		reasoningTokens := int64(0)
-		if v := usage.Get("completion_tokens_details.reasoning_tokens"); v.Exists() {
-			reasoningTokens = v.Int()
+		inputTokens, outputTokens, cachedTokens := extractOpenAIUsage(usage)
+		out, _ = sjson.Set(out, "usage.input_tokens", inputTokens)
+		out, _ = sjson.Set(out, "usage.output_tokens", outputTokens)
+		if cachedTokens > 0 {
+			out, _ = sjson.Set(out, "usage.cache_read_input_tokens", cachedTokens)
 		}
-		out, _ = sjson.Set(out, "usage.reasoning_tokens", reasoningTokens)
 	}
 
 	return []string{out}
@@ -674,8 +669,12 @@ func ConvertOpenAIResponseToClaudeNonStream(_ context.Context, _ string, origina
 	}
 
 	if respUsage := root.Get("usage"); respUsage.Exists() {
-		out, _ = sjson.Set(out, "usage.input_tokens", respUsage.Get("prompt_tokens").Int())
-		out, _ = sjson.Set(out, "usage.output_tokens", respUsage.Get("completion_tokens").Int())
+		inputTokens, outputTokens, cachedTokens := extractOpenAIUsage(respUsage)
+		out, _ = sjson.Set(out, "usage.input_tokens", inputTokens)
+		out, _ = sjson.Set(out, "usage.output_tokens", outputTokens)
+		if cachedTokens > 0 {
+			out, _ = sjson.Set(out, "usage.cache_read_input_tokens", cachedTokens)
+		}
 	}
 
 	if !stopReasonSet {
@@ -692,3 +691,23 @@ func ConvertOpenAIResponseToClaudeNonStream(_ context.Context, _ string, origina
 func ClaudeTokenCount(ctx context.Context, count int64) string {
 	return fmt.Sprintf(`{"input_tokens":%d}`, count)
 }
+
+func extractOpenAIUsage(usage gjson.Result) (int64, int64, int64) {
+	if !usage.Exists() || usage.Type == gjson.Null {
+		return 0, 0, 0
+	}
+
+	inputTokens := usage.Get("prompt_tokens").Int()
+	outputTokens := usage.Get("completion_tokens").Int()
+	cachedTokens := usage.Get("prompt_tokens_details.cached_tokens").Int()
+
+	if cachedTokens > 0 {
+		if inputTokens >= cachedTokens {
+			inputTokens -= cachedTokens
+		} else {
+			inputTokens = 0
+		}
+	}
+
+	return inputTokens, outputTokens, cachedTokens
+}

internal/translator/openai/gemini/openai_gemini_request.go

@@ -77,7 +77,13 @@ func ConvertGeminiRequestToOpenAI(modelName string, inputRawJSON []byte, stream
 			}
 		}
 
+		// Candidate count (OpenAI 'n' parameter)
+		if candidateCount := genConfig.Get("candidateCount"); candidateCount.Exists() {
+			out, _ = sjson.Set(out, "n", candidateCount.Int())
+		}
+
 		// Map Gemini thinkingConfig to OpenAI reasoning_effort.
+		// Always perform conversion to support allowCompat models that may not be in registry
 		if thinkingConfig := genConfig.Get("thinkingConfig"); thinkingConfig.Exists() && thinkingConfig.IsObject() {
 			if thinkingLevel := thinkingConfig.Get("thinkingLevel"); thinkingLevel.Exists() {
 				effort := strings.ToLower(strings.TrimSpace(thinkingLevel.String()))