Merge pull request #1435 from tianyicui/fix/haiku-4-5-thinking-support

fix: Enable extended thinking support for Claude Haiku 4.5

internal/config/oauth_model_alias_migration.go

@@ -17,6 +17,7 @@ var antigravityModelConversionTable = map[string]string{
 	"gemini-claude-sonnet-4-5":                "claude-sonnet-4-5",
 	"gemini-claude-sonnet-4-5-thinking":       "claude-sonnet-4-5-thinking",
 	"gemini-claude-opus-4-5-thinking":         "claude-opus-4-5-thinking",
+	"gemini-claude-opus-4-6-thinking":         "claude-opus-4-6-thinking",
 }
 
 // defaultAntigravityAliases returns the default oauth-model-alias configuration
@@ -30,6 +31,7 @@ func defaultAntigravityAliases() []OAuthModelAlias {
 		{Name: "claude-sonnet-4-5", Alias: "gemini-claude-sonnet-4-5"},
 		{Name: "claude-sonnet-4-5-thinking", Alias: "gemini-claude-sonnet-4-5-thinking"},
 		{Name: "claude-opus-4-5-thinking", Alias: "gemini-claude-opus-4-5-thinking"},
+		{Name: "claude-opus-4-6-thinking", Alias: "gemini-claude-opus-4-6-thinking"},
 	}
 }
 

internal/config/oauth_model_alias_migration_test.go

@@ -131,6 +131,9 @@ func TestMigrateOAuthModelAlias_ConvertsAntigravityModels(t *testing.T) {
 	if !strings.Contains(content, "claude-opus-4-5-thinking") {
 		t.Fatal("expected missing default alias claude-opus-4-5-thinking to be added")
 	}
+	if !strings.Contains(content, "claude-opus-4-6-thinking") {
+		t.Fatal("expected missing default alias claude-opus-4-6-thinking to be added")
+	}
 }
 
 func TestMigrateOAuthModelAlias_AddsDefaultIfNeitherExists(t *testing.T) {

internal/registry/model_definitions_static_data.go

@@ -15,7 +15,7 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 4.5 Haiku",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
-			// Thinking: not supported for Haiku models
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
 		},
 		{
 			ID:                  "claude-sonnet-4-5-20250929",
@@ -28,6 +28,18 @@ func GetClaudeModels() []*ModelInfo {
 			MaxCompletionTokens: 64000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
 		},
+		{
+			ID:                  "claude-opus-4-6",
+			Object:              "model",
+			Created:             1770318000, // 2026-02-05
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.6 Opus",
+			Description:         "Premium model combining maximum intelligence with practical performance",
+			ContextLength:       1000000,
+			MaxCompletionTokens: 128000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
+		},
 		{
 			ID:                  "claude-opus-4-5-20251101",
 			Object:              "model",
@@ -854,6 +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-sonnet-4-5":          {MaxCompletionTokens: 64000},
 		"gpt-oss-120b-medium":        {},
 		"tab_flash_lite_preview":     {},

internal/runtime/executor/aistudio_executor.go

@@ -141,7 +141,7 @@ func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth,
 		URL:       endpoint,
 		Method:    http.MethodPost,
 		Headers:   wsReq.Headers.Clone(),
-		Body:      bytes.Clone(body.payload),
+		Body:      body.payload,
 		Provider:  e.Identifier(),
 		AuthID:    authID,
 		AuthLabel: authLabel,
@@ -156,14 +156,14 @@ func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth,
 	}
 	recordAPIResponseMetadata(ctx, e.cfg, wsResp.Status, wsResp.Headers.Clone())
 	if len(wsResp.Body) > 0 {
-		appendAPIResponseChunk(ctx, e.cfg, bytes.Clone(wsResp.Body))
+		appendAPIResponseChunk(ctx, e.cfg, wsResp.Body)
 	}
 	if wsResp.Status < 200 || wsResp.Status >= 300 {
 		return resp, statusErr{code: wsResp.Status, msg: string(wsResp.Body)}
 	}
 	reporter.publish(ctx, parseGeminiUsage(wsResp.Body))
 	var param any
-	out := sdktranslator.TranslateNonStream(ctx, body.toFormat, opts.SourceFormat, req.Model, opts.OriginalRequest, bytes.Clone(translatedReq), bytes.Clone(wsResp.Body), &param)
+	out := sdktranslator.TranslateNonStream(ctx, body.toFormat, opts.SourceFormat, req.Model, opts.OriginalRequest, translatedReq, wsResp.Body, &param)
 	resp = cliproxyexecutor.Response{Payload: ensureColonSpacedJSON([]byte(out))}
 	return resp, nil
 }
@@ -199,7 +199,7 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 		URL:       endpoint,
 		Method:    http.MethodPost,
 		Headers:   wsReq.Headers.Clone(),
-		Body:      bytes.Clone(body.payload),
+		Body:      body.payload,
 		Provider:  e.Identifier(),
 		AuthID:    authID,
 		AuthLabel: authLabel,
@@ -225,7 +225,7 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 		}
 		var body bytes.Buffer
 		if len(firstEvent.Payload) > 0 {
-			appendAPIResponseChunk(ctx, e.cfg, bytes.Clone(firstEvent.Payload))
+			appendAPIResponseChunk(ctx, e.cfg, firstEvent.Payload)
 			body.Write(firstEvent.Payload)
 		}
 		if firstEvent.Type == wsrelay.MessageTypeStreamEnd {
@@ -244,7 +244,7 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 				metadataLogged = true
 			}
 			if len(event.Payload) > 0 {
-				appendAPIResponseChunk(ctx, e.cfg, bytes.Clone(event.Payload))
+				appendAPIResponseChunk(ctx, e.cfg, event.Payload)
 				body.Write(event.Payload)
 			}
 			if event.Type == wsrelay.MessageTypeStreamEnd {
@@ -274,12 +274,12 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 				}
 			case wsrelay.MessageTypeStreamChunk:
 				if len(event.Payload) > 0 {
-					appendAPIResponseChunk(ctx, e.cfg, bytes.Clone(event.Payload))
+					appendAPIResponseChunk(ctx, e.cfg, event.Payload)
 					filtered := FilterSSEUsageMetadata(event.Payload)
 					if detail, ok := parseGeminiStreamUsage(filtered); ok {
 						reporter.publish(ctx, detail)
 					}
-					lines := sdktranslator.TranslateStream(ctx, body.toFormat, opts.SourceFormat, req.Model, opts.OriginalRequest, translatedReq, bytes.Clone(filtered), &param)
+					lines := sdktranslator.TranslateStream(ctx, body.toFormat, opts.SourceFormat, req.Model, opts.OriginalRequest, translatedReq, filtered, &param)
 					for i := range lines {
 						out <- cliproxyexecutor.StreamChunk{Payload: ensureColonSpacedJSON([]byte(lines[i]))}
 					}
@@ -293,9 +293,9 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 					metadataLogged = true
 				}
 				if len(event.Payload) > 0 {
-					appendAPIResponseChunk(ctx, e.cfg, bytes.Clone(event.Payload))
+					appendAPIResponseChunk(ctx, e.cfg, event.Payload)
 				}
-				lines := sdktranslator.TranslateStream(ctx, body.toFormat, opts.SourceFormat, req.Model, opts.OriginalRequest, translatedReq, bytes.Clone(event.Payload), &param)
+				lines := sdktranslator.TranslateStream(ctx, body.toFormat, opts.SourceFormat, req.Model, opts.OriginalRequest, translatedReq, event.Payload, &param)
 				for i := range lines {
 					out <- cliproxyexecutor.StreamChunk{Payload: ensureColonSpacedJSON([]byte(lines[i]))}
 				}
@@ -350,7 +350,7 @@ func (e *AIStudioExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.A
 		URL:       endpoint,
 		Method:    http.MethodPost,
 		Headers:   wsReq.Headers.Clone(),
-		Body:      bytes.Clone(body.payload),
+		Body:      body.payload,
 		Provider:  e.Identifier(),
 		AuthID:    authID,
 		AuthLabel: authLabel,
@@ -364,7 +364,7 @@ func (e *AIStudioExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.A
 	}
 	recordAPIResponseMetadata(ctx, e.cfg, resp.Status, resp.Headers.Clone())
 	if len(resp.Body) > 0 {
-		appendAPIResponseChunk(ctx, e.cfg, bytes.Clone(resp.Body))
+		appendAPIResponseChunk(ctx, e.cfg, resp.Body)
 	}
 	if resp.Status < 200 || resp.Status >= 300 {
 		return cliproxyexecutor.Response{}, statusErr{code: resp.Status, msg: string(resp.Body)}
@@ -373,7 +373,7 @@ func (e *AIStudioExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.A
 	if totalTokens <= 0 {
 		return cliproxyexecutor.Response{}, fmt.Errorf("wsrelay: totalTokens missing in response")
 	}
-	translated := sdktranslator.TranslateTokenCount(ctx, body.toFormat, opts.SourceFormat, totalTokens, bytes.Clone(resp.Body))
+	translated := sdktranslator.TranslateTokenCount(ctx, body.toFormat, opts.SourceFormat, totalTokens, resp.Body)
 	return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
 }
 
@@ -397,9 +397,9 @@ func (e *AIStudioExecutor) translateRequest(req cliproxyexecutor.Request, opts c
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, stream)
-	payload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), stream)
+	payload := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, stream)
 	payload, err := thinking.ApplyThinking(payload, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
 		return nil, translatedPayload{}, err

internal/runtime/executor/antigravity_executor.go

@@ -137,9 +137,9 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
-	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	translated := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -279,9 +279,9 @@ func (e *AntigravityExecutor) executeClaudeNonStream(ctx context.Context, auth *
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	translated := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 
 	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -671,9 +671,9 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	translated := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 
 	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -875,7 +875,7 @@ func (e *AntigravityExecutor) CountTokens(ctx context.Context, auth *cliproxyaut
 	respCtx := context.WithValue(ctx, "alt", opts.Alt)
 
 	// Prepare payload once (doesn't depend on baseURL)
-	payload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	payload := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	payload, err := thinking.ApplyThinking(payload, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {

internal/runtime/executor/claude_executor.go

@@ -104,9 +104,9 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, stream)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), stream)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, stream)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
@@ -245,9 +245,9 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
@@ -413,7 +413,7 @@ func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	to := sdktranslator.FromString("claude")
 	// Use streaming translation to preserve function calling, except for claude.
 	stream := from != to
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), stream)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, stream)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	if !strings.HasPrefix(baseModel, "claude-3-5-haiku") {

internal/runtime/executor/codex_executor.go

@@ -27,6 +27,11 @@ import (
 	"github.com/google/uuid"
 )
 
+const (
+	codexClientVersion = "0.98.0"
+	codexUserAgent     = "codex_cli_rs/0.98.0 (Mac OS 26.0.1; arm64) Apple_Terminal/464"
+)
+
 var dataTag = []byte("data:")
 
 // CodexExecutor is a stateless executor for Codex (OpenAI Responses API entrypoint).
@@ -92,9 +97,9 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -202,9 +207,9 @@ func (e *CodexExecutor) executeCompact(ctx context.Context, auth *cliproxyauth.A
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -292,9 +297,9 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -400,7 +405,7 @@ func (e *CodexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	body, err := thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -637,10 +642,10 @@ func applyCodexHeaders(r *http.Request, auth *cliproxyauth.Auth, token string, s
 		ginHeaders = ginCtx.Request.Header
 	}
 
-	misc.EnsureHeader(r.Header, ginHeaders, "Version", "0.21.0")
+	misc.EnsureHeader(r.Header, ginHeaders, "Version", codexClientVersion)
 	misc.EnsureHeader(r.Header, ginHeaders, "Openai-Beta", "responses=experimental")
 	misc.EnsureHeader(r.Header, ginHeaders, "Session_id", uuid.NewString())
-	misc.EnsureHeader(r.Header, ginHeaders, "User-Agent", "codex_cli_rs/0.50.0 (Mac OS 26.0.1; arm64) Apple_Terminal/464")
+	misc.EnsureHeader(r.Header, ginHeaders, "User-Agent", codexUserAgent)
 
 	if stream {
 		r.Header.Set("Accept", "text/event-stream")

internal/runtime/executor/gemini_cli_executor.go

@@ -123,9 +123,9 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
-	basePayload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	basePayload := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	basePayload, err = thinking.ApplyThinking(basePayload, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -277,9 +277,9 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	basePayload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	basePayload := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 
 	basePayload, err = thinking.ApplyThinking(basePayload, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -408,7 +408,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 					}
 				}
 
-				segments := sdktranslator.TranslateStream(respCtx, to, from, attemptModel, opts.OriginalRequest, reqBody, bytes.Clone([]byte("[DONE]")), &param)
+				segments := sdktranslator.TranslateStream(respCtx, to, from, attemptModel, opts.OriginalRequest, reqBody, []byte("[DONE]"), &param)
 				for i := range segments {
 					out <- cliproxyexecutor.StreamChunk{Payload: []byte(segments[i])}
 				}
@@ -435,7 +435,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 				out <- cliproxyexecutor.StreamChunk{Payload: []byte(segments[i])}
 			}
 
-			segments = sdktranslator.TranslateStream(respCtx, to, from, attemptModel, opts.OriginalRequest, reqBody, bytes.Clone([]byte("[DONE]")), &param)
+			segments = sdktranslator.TranslateStream(respCtx, to, from, attemptModel, opts.OriginalRequest, reqBody, []byte("[DONE]"), &param)
 			for i := range segments {
 				out <- cliproxyexecutor.StreamChunk{Payload: []byte(segments[i])}
 			}
@@ -487,7 +487,7 @@ func (e *GeminiCLIExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.
 	// The loop variable attemptModel is only used as the concrete model id sent to the upstream
 	// Gemini CLI endpoint when iterating fallback variants.
 	for range models {
-		payload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+		payload := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 		payload, err = thinking.ApplyThinking(payload, req.Model, from.String(), to.String(), e.Identifier())
 		if err != nil {

internal/runtime/executor/gemini_executor.go

@@ -120,9 +120,9 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -227,9 +227,9 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -325,7 +325,7 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 				out <- cliproxyexecutor.StreamChunk{Payload: []byte(lines[i])}
 			}
 		}
-		lines := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, bytes.Clone([]byte("[DONE]")), &param)
+		lines := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, []byte("[DONE]"), &param)
 		for i := range lines {
 			out <- cliproxyexecutor.StreamChunk{Payload: []byte(lines[i])}
 		}
@@ -346,7 +346,7 @@ func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
-	translatedReq := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	translatedReq := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	translatedReq, err := thinking.ApplyThinking(translatedReq, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {

internal/runtime/executor/gemini_vertex_executor.go

@@ -322,9 +322,9 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 		if len(opts.OriginalRequest) > 0 {
 			originalPayloadSource = opts.OriginalRequest
 		}
-		originalPayload := bytes.Clone(originalPayloadSource)
+		originalPayload := originalPayloadSource
 		originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
-		body = sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+		body = sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 		body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 		if err != nil {
@@ -437,9 +437,9 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -542,9 +542,9 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -667,9 +667,9 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -785,7 +785,7 @@ func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 
-	translatedReq := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	translatedReq := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	translatedReq, err := thinking.ApplyThinking(translatedReq, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -869,7 +869,7 @@ func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 
-	translatedReq := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	translatedReq := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	translatedReq, err := thinking.ApplyThinking(translatedReq, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {

internal/runtime/executor/iflow_executor.go

@@ -91,9 +91,9 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), "iflow", e.Identifier())
@@ -194,9 +194,9 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), "iflow", e.Identifier())
@@ -298,7 +298,7 @@ func (e *IFlowExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	enc, err := tokenizerForModel(baseModel)
 	if err != nil {

internal/runtime/executor/logging_helpers.go

@@ -80,7 +80,7 @@ func recordAPIRequest(ctx context.Context, cfg *config.Config, info upstreamRequ
 	writeHeaders(builder, info.Headers)
 	builder.WriteString("\nBody:\n")
 	if len(info.Body) > 0 {
-		builder.WriteString(string(bytes.Clone(info.Body)))
+		builder.WriteString(string(info.Body))
 	} else {
 		builder.WriteString("<empty>")
 	}
@@ -152,7 +152,7 @@ func appendAPIResponseChunk(ctx context.Context, cfg *config.Config, chunk []byt
 	if cfg == nil || !cfg.RequestLog {
 		return
 	}
-	data := bytes.TrimSpace(bytes.Clone(chunk))
+	data := bytes.TrimSpace(chunk)
 	if len(data) == 0 {
 		return
 	}

internal/runtime/executor/openai_compat_executor.go

@@ -92,9 +92,9 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, opts.Stream)
-	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), opts.Stream)
+	translated := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, opts.Stream)
 	requestedModel := payloadRequestedModel(opts, req.Model)
 	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", translated, originalTranslated, requestedModel)
 	if opts.Alt == "responses/compact" {
@@ -194,9 +194,9 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	translated := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 	requestedModel := payloadRequestedModel(opts, req.Model)
 	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", translated, originalTranslated, requestedModel)
 
@@ -306,7 +306,7 @@ func (e *OpenAICompatExecutor) CountTokens(ctx context.Context, auth *cliproxyau
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
-	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	translated := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	modelForCounting := baseModel
 

internal/runtime/executor/qwen_executor.go

@@ -85,9 +85,9 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
@@ -176,9 +176,9 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 	if len(opts.OriginalRequest) > 0 {
 		originalPayloadSource = opts.OriginalRequest
 	}
-	originalPayload := bytes.Clone(originalPayloadSource)
+	originalPayload := originalPayloadSource
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, true)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
@@ -260,7 +260,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 				out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
 			}
 		}
-		doneChunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, bytes.Clone([]byte("[DONE]")), &param)
+		doneChunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, []byte("[DONE]"), &param)
 		for i := range doneChunks {
 			out <- cliproxyexecutor.StreamChunk{Payload: []byte(doneChunks[i])}
 		}
@@ -278,7 +278,7 @@ func (e *QwenExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth,
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
-	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, req.Payload, false)
 
 	modelName := gjson.GetBytes(body, "model").String()
 	if strings.TrimSpace(modelName) == "" {

internal/translator/antigravity/claude/antigravity_claude_request.go

@@ -6,7 +6,6 @@
 package claude
 
 import (
-	"bytes"
 	"strings"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/cache"
@@ -37,7 +36,7 @@ import (
 //   - []byte: The transformed request data in Gemini CLI API format
 func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _ bool) []byte {
 	enableThoughtTranslate := true
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 
 	// system instruction
 	systemInstructionJSON := ""

internal/translator/antigravity/gemini/antigravity_gemini_request.go

@@ -6,7 +6,6 @@
 package gemini
 
 import (
-	"bytes"
 	"fmt"
 	"strings"
 
@@ -34,7 +33,7 @@ import (
 // Returns:
 //   - []byte: The transformed request data in Gemini API format
 func ConvertGeminiRequestToAntigravity(modelName string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	template := ""
 	template = `{"project":"","request":{},"model":""}`
 	template, _ = sjson.SetRaw(template, "request", string(rawJSON))

internal/translator/antigravity/openai/responses/antigravity_openai-responses_request.go

@@ -1,14 +1,12 @@
 package responses
 
 import (
-	"bytes"
-
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/antigravity/gemini"
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/openai/responses"
 )
 
 func ConvertOpenAIResponsesRequestToAntigravity(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	rawJSON = ConvertOpenAIResponsesRequestToGemini(modelName, rawJSON, stream)
 	return ConvertGeminiRequestToAntigravity(modelName, rawJSON, stream)
 }

internal/translator/claude/gemini-cli/claude_gemini-cli_request.go

@@ -6,8 +6,6 @@
 package geminiCLI
 
 import (
-	"bytes"
-
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/claude/gemini"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -30,7 +28,7 @@ import (
 // Returns:
 //   - []byte: The transformed request data in Claude Code API format
 func ConvertGeminiCLIRequestToClaude(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 
 	modelResult := gjson.GetBytes(rawJSON, "model")
 	// Extract the inner request object and promote it to the top level

internal/translator/claude/gemini/claude_gemini_request.go

@@ -6,7 +6,6 @@
 package gemini
 
 import (
-	"bytes"
 	"crypto/rand"
 	"crypto/sha256"
 	"encoding/hex"
@@ -46,7 +45,7 @@ var (
 // Returns:
 //   - []byte: The transformed request data in Claude Code API format
 func ConvertGeminiRequestToClaude(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 
 	if account == "" {
 		u, _ := uuid.NewRandom()

internal/translator/claude/openai/chat-completions/claude_openai_request.go

@@ -6,7 +6,6 @@
 package chat_completions
 
 import (
-	"bytes"
 	"crypto/rand"
 	"crypto/sha256"
 	"encoding/hex"
@@ -44,7 +43,7 @@ var (
 // Returns:
 //   - []byte: The transformed request data in Claude Code API format
 func ConvertOpenAIRequestToClaude(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 
 	if account == "" {
 		u, _ := uuid.NewRandom()

internal/translator/claude/openai/responses/claude_openai-responses_request.go

@@ -1,7 +1,6 @@
 package responses
 
 import (
-	"bytes"
 	"crypto/rand"
 	"crypto/sha256"
 	"encoding/hex"
@@ -32,7 +31,7 @@ var (
 // - max_output_tokens -> max_tokens
 // - stream passthrough via parameter
 func ConvertOpenAIResponsesRequestToClaude(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 
 	if account == "" {
 		u, _ := uuid.NewRandom()

internal/translator/codex/claude/codex_claude_request.go

@@ -6,7 +6,6 @@
 package claude
 
 import (
-	"bytes"
 	"fmt"
 	"strconv"
 	"strings"
@@ -35,7 +34,7 @@ import (
 // Returns:
 //   - []byte: The transformed request data in internal client format
 func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 
 	template := `{"model":"","instructions":"","input":[]}`
 

internal/translator/codex/gemini-cli/codex_gemini-cli_request.go

@@ -6,8 +6,6 @@
 package geminiCLI
 
 import (
-	"bytes"
-
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/codex/gemini"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -30,7 +28,7 @@ import (
 // Returns:
 //   - []byte: The transformed request data in Codex API format
 func ConvertGeminiCLIRequestToCodex(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 
 	rawJSON = []byte(gjson.GetBytes(rawJSON, "request").Raw)
 	rawJSON, _ = sjson.SetBytes(rawJSON, "model", modelName)

internal/translator/codex/gemini/codex_gemini_request.go

@@ -6,7 +6,6 @@
 package gemini
 
 import (
-	"bytes"
 	"crypto/rand"
 	"fmt"
 	"math/big"
@@ -37,7 +36,7 @@ import (
 // Returns:
 //   - []byte: The transformed request data in Codex API format
 func ConvertGeminiRequestToCodex(modelName string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	// Base template
 	out := `{"model":"","instructions":"","input":[]}`
 

internal/translator/codex/openai/chat-completions/codex_openai_request.go

@@ -7,8 +7,6 @@
 package chat_completions
 
 import (
-	"bytes"
-
 	"strconv"
 	"strings"
 
@@ -29,7 +27,7 @@ import (
 // Returns:
 //   - []byte: The transformed request data in OpenAI Responses API format
 func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	// Start with empty JSON object
 	out := `{"instructions":""}`
 

internal/translator/codex/openai/responses/codex_openai-responses_request.go

@@ -1,7 +1,6 @@
 package responses
 
 import (
-	"bytes"
 	"fmt"
 
 	"github.com/tidwall/gjson"
@@ -9,7 +8,7 @@ import (
 )
 
 func ConvertOpenAIResponsesRequestToCodex(modelName string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 
 	inputResult := gjson.GetBytes(rawJSON, "input")
 	if inputResult.Type == gjson.String {

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

@@ -35,7 +35,7 @@ const geminiCLIClaudeThoughtSignature = "skip_thought_signature_validator"
 // Returns:
 //   - []byte: The transformed request data in Gemini CLI API format
 func ConvertClaudeRequestToCLI(modelName string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	rawJSON = bytes.Replace(rawJSON, []byte(`"url":{"type":"string","format":"uri",`), []byte(`"url":{"type":"string",`), -1)
 
 	// Build output Gemini CLI request JSON

internal/translator/gemini-cli/gemini/gemini-cli_gemini_request.go

@@ -6,7 +6,6 @@
 package gemini
 
 import (
-	"bytes"
 	"fmt"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
@@ -33,7 +32,7 @@ import (
 // Returns:
 //   - []byte: The transformed request data in Gemini API format
 func ConvertGeminiRequestToGeminiCLI(_ string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	template := ""
 	template = `{"project":"","request":{},"model":""}`
 	template, _ = sjson.SetRaw(template, "request", string(rawJSON))

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

@@ -3,7 +3,6 @@
 package chat_completions
 
 import (
-	"bytes"
 	"fmt"
 	"strings"
 
@@ -28,7 +27,7 @@ const geminiCLIFunctionThoughtSignature = "skip_thought_signature_validator"
 // Returns:
 //   - []byte: The transformed request data in Gemini CLI API format
 func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	// Base envelope (no default thinkingConfig)
 	out := []byte(`{"project":"","request":{"contents":[]},"model":"gemini-2.5-pro"}`)
 

internal/translator/gemini-cli/openai/responses/gemini-cli_openai-responses_request.go

@@ -1,14 +1,12 @@
 package responses
 
 import (
-	"bytes"
-
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini-cli/gemini"
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/openai/responses"
 )
 
 func ConvertOpenAIResponsesRequestToGeminiCLI(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	rawJSON = ConvertOpenAIResponsesRequestToGemini(modelName, rawJSON, stream)
 	return ConvertGeminiRequestToGeminiCLI(modelName, rawJSON, stream)
 }

internal/translator/gemini/claude/gemini_claude_request.go

@@ -28,7 +28,7 @@ const geminiClaudeThoughtSignature = "skip_thought_signature_validator"
 // Returns:
 //   - []byte: The transformed request in Gemini CLI format.
 func ConvertClaudeRequestToGemini(modelName string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	rawJSON = bytes.Replace(rawJSON, []byte(`"url":{"type":"string","format":"uri",`), []byte(`"url":{"type":"string",`), -1)
 
 	// Build output Gemini CLI request JSON

internal/translator/gemini/gemini-cli/gemini_gemini-cli_request.go

@@ -6,7 +6,6 @@
 package geminiCLI
 
 import (
-	"bytes"
 	"fmt"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
@@ -19,7 +18,7 @@ import (
 // It extracts the model name, system instruction, message contents, and tool declarations
 // from the raw JSON request and returns them in the format expected by the internal client.
 func ConvertGeminiCLIRequestToGemini(_ string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	modelResult := gjson.GetBytes(rawJSON, "model")
 	rawJSON = []byte(gjson.GetBytes(rawJSON, "request").Raw)
 	rawJSON, _ = sjson.SetBytes(rawJSON, "model", modelResult.String())

internal/translator/gemini/gemini/gemini_gemini_request.go

@@ -4,7 +4,6 @@
 package gemini
 
 import (
-	"bytes"
 	"fmt"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
@@ -19,7 +18,7 @@ import (
 //
 // It keeps the payload otherwise unchanged.
 func ConvertGeminiRequestToGemini(_ string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	// Fast path: if no contents field, only attach safety settings
 	contents := gjson.GetBytes(rawJSON, "contents")
 	if !contents.Exists() {

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

@@ -3,7 +3,6 @@
 package chat_completions
 
 import (
-	"bytes"
 	"fmt"
 	"strings"
 
@@ -28,7 +27,7 @@ const geminiFunctionThoughtSignature = "skip_thought_signature_validator"
 // Returns:
 //   - []byte: The transformed request data in Gemini API format
 func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	// Base envelope (no default thinkingConfig)
 	out := []byte(`{"contents":[]}`)
 

internal/translator/gemini/openai/responses/gemini_openai-responses_request.go

@@ -1,7 +1,6 @@
 package responses
 
 import (
-	"bytes"
 	"strings"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
@@ -12,7 +11,7 @@ import (
 const geminiResponsesThoughtSignature = "skip_thought_signature_validator"
 
 func ConvertOpenAIResponsesRequestToGemini(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 
 	// Note: modelName and stream parameters are part of the fixed method signature
 	_ = modelName // Unused but required by interface

internal/translator/openai/claude/openai_claude_request.go

@@ -6,7 +6,6 @@
 package claude
 
 import (
-	"bytes"
 	"strings"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
@@ -18,7 +17,7 @@ import (
 // It extracts the model name, system instruction, message contents, and tool declarations
 // from the raw JSON request and returns them in the format expected by the OpenAI API.
 func ConvertClaudeRequestToOpenAI(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	// Base OpenAI Chat Completions API template
 	out := `{"model":"","messages":[]}`
 

internal/translator/openai/gemini-cli/openai_gemini_request.go

@@ -6,8 +6,6 @@
 package geminiCLI
 
 import (
-	"bytes"
-
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/openai/gemini"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -17,7 +15,7 @@ import (
 // It extracts the model name, generation config, message contents, and tool declarations
 // from the raw JSON request and returns them in the format expected by the OpenAI API.
 func ConvertGeminiCLIRequestToOpenAI(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	rawJSON = []byte(gjson.GetBytes(rawJSON, "request").Raw)
 	rawJSON, _ = sjson.SetBytes(rawJSON, "model", modelName)
 	if gjson.GetBytes(rawJSON, "systemInstruction").Exists() {

internal/translator/openai/gemini/openai_gemini_request.go

@@ -6,7 +6,6 @@
 package gemini
 
 import (
-	"bytes"
 	"crypto/rand"
 	"fmt"
 	"math/big"
@@ -21,7 +20,7 @@ import (
 // It extracts the model name, generation config, message contents, and tool declarations
 // from the raw JSON request and returns them in the format expected by the OpenAI API.
 func ConvertGeminiRequestToOpenAI(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	// Base OpenAI Chat Completions API template
 	out := `{"model":"","messages":[]}`
 

internal/translator/openai/openai/chat-completions/openai_openai_request.go

@@ -3,7 +3,6 @@
 package chat_completions
 
 import (
-	"bytes"
 	"github.com/tidwall/sjson"
 )
 
@@ -25,7 +24,7 @@ func ConvertOpenAIRequestToOpenAI(modelName string, inputRawJSON []byte, _ bool)
 		// If there's an error, return the original JSON or handle the error appropriately.
 		// For now, we'll return the original, but in a real scenario, logging or a more robust error
 		// handling mechanism would be needed.
-		return bytes.Clone(inputRawJSON)
+		return inputRawJSON
 	}
 	return updatedJSON
 }

internal/translator/openai/openai/responses/openai_openai-responses_request.go

@@ -1,7 +1,6 @@
 package responses
 
 import (
-	"bytes"
 	"strings"
 
 	"github.com/tidwall/gjson"
@@ -28,7 +27,7 @@ import (
 // Returns:
 //   - []byte: The transformed request data in OpenAI chat completions format
 func ConvertOpenAIResponsesRequestToOpenAIChatCompletions(modelName string, inputRawJSON []byte, stream bool) []byte {
-	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON := inputRawJSON
 	// Base OpenAI chat completions template with default values
 	out := `{"model":"","messages":[],"stream":false}`
 

internal/util/claude_model_test.go

@@ -11,6 +11,7 @@ func TestIsClaudeThinkingModel(t *testing.T) {
 		// Claude thinking models - should return true
 		{"claude-sonnet-4-5-thinking", "claude-sonnet-4-5-thinking", true},
 		{"claude-opus-4-5-thinking", "claude-opus-4-5-thinking", true},
+		{"claude-opus-4-6-thinking", "claude-opus-4-6-thinking", true},
 		{"Claude-Sonnet-Thinking uppercase", "Claude-Sonnet-4-5-Thinking", true},
 		{"claude thinking mixed case", "Claude-THINKING-Model", true},
 

sdk/api/handlers/handlers.go

@@ -408,7 +408,7 @@ func (h *BaseAPIHandler) ExecuteWithAuthManager(ctx context.Context, handlerType
 		}
 		return nil, &interfaces.ErrorMessage{StatusCode: status, Error: err, Addon: addon}
 	}
-	return cloneBytes(resp.Payload), nil
+	return resp.Payload, nil
 }
 
 // ExecuteCountWithAuthManager executes a non-streaming request via the core auth manager.
@@ -451,7 +451,7 @@ func (h *BaseAPIHandler) ExecuteCountWithAuthManager(ctx context.Context, handle
 		}
 		return nil, &interfaces.ErrorMessage{StatusCode: status, Error: err, Addon: addon}
 	}
-	return cloneBytes(resp.Payload), nil
+	return resp.Payload, nil
 }
 
 // ExecuteStreamWithAuthManager executes a streaming request via the core auth manager.
@@ -696,7 +696,7 @@ func (h *BaseAPIHandler) WriteErrorResponse(c *gin.Context, msg *interfaces.Erro
 	var previous []byte
 	if existing, exists := c.Get("API_RESPONSE"); exists {
 		if existingBytes, ok := existing.([]byte); ok && len(existingBytes) > 0 {
-			previous = bytes.Clone(existingBytes)
+			previous = existingBytes
 		}
 	}
 	appendAPIResponse(c, body)