Merge pull request #488 from router-for-me/gemini

Unify the Gemini executor style

cmd/server/main.go

@@ -139,7 +139,8 @@ func main() {
 
 	wd, err := os.Getwd()
 	if err != nil {
-		log.Fatalf("failed to get working directory: %v", err)
+		log.Errorf("failed to get working directory: %v", err)
+		return
 	}
 
 	// Load environment variables from .env if present.
@@ -233,13 +234,15 @@ func main() {
 		})
 		cancel()
 		if err != nil {
-			log.Fatalf("failed to initialize postgres token store: %v", err)
+			log.Errorf("failed to initialize postgres token store: %v", err)
+			return
 		}
 		examplePath := filepath.Join(wd, "config.example.yaml")
 		ctx, cancel = context.WithTimeout(context.Background(), 30*time.Second)
 		if errBootstrap := pgStoreInst.Bootstrap(ctx, examplePath); errBootstrap != nil {
 			cancel()
-			log.Fatalf("failed to bootstrap postgres-backed config: %v", errBootstrap)
+			log.Errorf("failed to bootstrap postgres-backed config: %v", errBootstrap)
+			return
 		}
 		cancel()
 		configFilePath = pgStoreInst.ConfigPath()
@@ -262,7 +265,8 @@ func main() {
 		if strings.Contains(resolvedEndpoint, "://") {
 			parsed, errParse := url.Parse(resolvedEndpoint)
 			if errParse != nil {
-				log.Fatalf("failed to parse object store endpoint %q: %v", objectStoreEndpoint, errParse)
+				log.Errorf("failed to parse object store endpoint %q: %v", objectStoreEndpoint, errParse)
+				return
 			}
 			switch strings.ToLower(parsed.Scheme) {
 			case "http":
@@ -270,10 +274,12 @@ func main() {
 			case "https":
 				useSSL = true
 			default:
-				log.Fatalf("unsupported object store scheme %q (only http and https are allowed)", parsed.Scheme)
+				log.Errorf("unsupported object store scheme %q (only http and https are allowed)", parsed.Scheme)
+				return
 			}
 			if parsed.Host == "" {
-				log.Fatalf("object store endpoint %q is missing host information", objectStoreEndpoint)
+				log.Errorf("object store endpoint %q is missing host information", objectStoreEndpoint)
+				return
 			}
 			resolvedEndpoint = parsed.Host
 			if parsed.Path != "" && parsed.Path != "/" {
@@ -292,13 +298,15 @@ func main() {
 		}
 		objectStoreInst, err = store.NewObjectTokenStore(objCfg)
 		if err != nil {
-			log.Fatalf("failed to initialize object token store: %v", err)
+			log.Errorf("failed to initialize object token store: %v", err)
+			return
 		}
 		examplePath := filepath.Join(wd, "config.example.yaml")
 		ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
 		if errBootstrap := objectStoreInst.Bootstrap(ctx, examplePath); errBootstrap != nil {
 			cancel()
-			log.Fatalf("failed to bootstrap object-backed config: %v", errBootstrap)
+			log.Errorf("failed to bootstrap object-backed config: %v", errBootstrap)
+			return
 		}
 		cancel()
 		configFilePath = objectStoreInst.ConfigPath()
@@ -323,7 +331,8 @@ func main() {
 		gitStoreInst = store.NewGitTokenStore(gitStoreRemoteURL, gitStoreUser, gitStorePassword)
 		gitStoreInst.SetBaseDir(authDir)
 		if errRepo := gitStoreInst.EnsureRepository(); errRepo != nil {
-			log.Fatalf("failed to prepare git token store: %v", errRepo)
+			log.Errorf("failed to prepare git token store: %v", errRepo)
+			return
 		}
 		configFilePath = gitStoreInst.ConfigPath()
 		if configFilePath == "" {
@@ -332,17 +341,21 @@ func main() {
 		if _, statErr := os.Stat(configFilePath); errors.Is(statErr, fs.ErrNotExist) {
 			examplePath := filepath.Join(wd, "config.example.yaml")
 			if _, errExample := os.Stat(examplePath); errExample != nil {
-				log.Fatalf("failed to find template config file: %v", errExample)
+				log.Errorf("failed to find template config file: %v", errExample)
+				return
 			}
 			if errCopy := misc.CopyConfigTemplate(examplePath, configFilePath); errCopy != nil {
-				log.Fatalf("failed to bootstrap git-backed config: %v", errCopy)
+				log.Errorf("failed to bootstrap git-backed config: %v", errCopy)
+				return
 			}
 			if errCommit := gitStoreInst.PersistConfig(context.Background()); errCommit != nil {
-				log.Fatalf("failed to commit initial git-backed config: %v", errCommit)
+				log.Errorf("failed to commit initial git-backed config: %v", errCommit)
+				return
 			}
 			log.Infof("git-backed config initialized from template: %s", configFilePath)
 		} else if statErr != nil {
-			log.Fatalf("failed to inspect git-backed config: %v", statErr)
+			log.Errorf("failed to inspect git-backed config: %v", statErr)
+			return
 		}
 		cfg, err = config.LoadConfigOptional(configFilePath, isCloudDeploy)
 		if err == nil {
@@ -355,13 +368,15 @@ func main() {
 	} else {
 		wd, err = os.Getwd()
 		if err != nil {
-			log.Fatalf("failed to get working directory: %v", err)
+			log.Errorf("failed to get working directory: %v", err)
+			return
 		}
 		configFilePath = filepath.Join(wd, "config.yaml")
 		cfg, err = config.LoadConfigOptional(configFilePath, isCloudDeploy)
 	}
 	if err != nil {
-		log.Fatalf("failed to load config: %v", err)
+		log.Errorf("failed to load config: %v", err)
+		return
 	}
 	if cfg == nil {
 		cfg = &config.Config{}
@@ -391,7 +406,8 @@ func main() {
 	coreauth.SetQuotaCooldownDisabled(cfg.DisableCooling)
 
 	if err = logging.ConfigureLogOutput(cfg.LoggingToFile); err != nil {
-		log.Fatalf("failed to configure log output: %v", err)
+		log.Errorf("failed to configure log output: %v", err)
+		return
 	}
 
 	log.Infof("CLIProxyAPI Version: %s, Commit: %s, BuiltAt: %s", buildinfo.Version, buildinfo.Commit, buildinfo.BuildDate)
@@ -400,7 +416,8 @@ func main() {
 	util.SetLogLevel(cfg)
 
 	if resolvedAuthDir, errResolveAuthDir := util.ResolveAuthDir(cfg.AuthDir); errResolveAuthDir != nil {
-		log.Fatalf("failed to resolve auth directory: %v", errResolveAuthDir)
+		log.Errorf("failed to resolve auth directory: %v", errResolveAuthDir)
+		return
 	} else {
 		cfg.AuthDir = resolvedAuthDir
 	}

config.example.yaml

@@ -1,3 +1,7 @@
+# Server host/interface to bind to. Default is empty ("") to bind all interfaces (IPv4 + IPv6).
+# Use "127.0.0.1" or "localhost" to restrict access to local machine only.
+host: ""
+
 # Server port
 port: 8317
 
@@ -96,7 +100,7 @@ ws-auth: false
 #     excluded-models:
 #       - "claude-opus-4-5-20251101" # exclude specific models (exact match)
 #       - "claude-3-*"               # wildcard matching prefix (e.g. claude-3-7-sonnet-20250219)
-#       - "*-think"                  # wildcard matching suffix (e.g. claude-opus-4-5-thinking)
+#       - "*-thinking"               # wildcard matching suffix (e.g. claude-opus-4-5-thinking)
 #       - "*haiku*"                  # wildcard matching substring (e.g. claude-3-5-haiku-20241022)
 
 # OpenAI compatibility providers
@@ -134,6 +138,8 @@ ws-auth: false
 #   upstream-api-key: ""
 #   # Restrict Amp management routes (/api/auth, /api/user, etc.) to localhost only (recommended)
 #   restrict-management-to-localhost: true
+#   # Force model mappings to run before checking local API keys (default: false)
+#   force-model-mappings: false
 #   # Amp Model Mappings
 #   # Route unavailable Amp models to alternative models available in your local proxy.
 #   # Useful when Amp CLI requests models you don't have access to (e.g., Claude Opus 4.5)

internal/api/handlers/management/auth_files.go

@@ -713,14 +713,16 @@ func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 	// Generate PKCE codes
 	pkceCodes, err := claude.GeneratePKCECodes()
 	if err != nil {
-		log.Fatalf("Failed to generate PKCE codes: %v", err)
+		log.Errorf("Failed to generate PKCE codes: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate PKCE codes"})
 		return
 	}
 
 	// Generate random state parameter
 	state, err := misc.GenerateRandomState()
 	if err != nil {
-		log.Fatalf("Failed to generate state parameter: %v", err)
+		log.Errorf("Failed to generate state parameter: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate state parameter"})
 		return
 	}
 
@@ -730,7 +732,8 @@ func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 	// Generate authorization URL (then override redirect_uri to reuse server port)
 	authURL, state, err := anthropicAuth.GenerateAuthURL(state, pkceCodes)
 	if err != nil {
-		log.Fatalf("Failed to generate authorization URL: %v", err)
+		log.Errorf("Failed to generate authorization URL: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate authorization url"})
 		return
 	}
 
@@ -872,7 +875,7 @@ func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 		}
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
-			log.Fatalf("Failed to save authentication tokens: %v", errSave)
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
 			oauthStatus[state] = "Failed to save authentication tokens"
 			return
 		}
@@ -1045,7 +1048,7 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 		gemAuth := geminiAuth.NewGeminiAuth()
 		gemClient, errGetClient := gemAuth.GetAuthenticatedClient(ctx, &ts, h.cfg, true)
 		if errGetClient != nil {
-			log.Fatalf("failed to get authenticated client: %v", errGetClient)
+			log.Errorf("failed to get authenticated client: %v", errGetClient)
 			oauthStatus[state] = "Failed to get authenticated client"
 			return
 		}
@@ -1110,7 +1113,7 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 		}
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
-			log.Fatalf("Failed to save token to file: %v", errSave)
+			log.Errorf("Failed to save token to file: %v", errSave)
 			oauthStatus[state] = "Failed to save token to file"
 			return
 		}
@@ -1131,14 +1134,16 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 	// Generate PKCE codes
 	pkceCodes, err := codex.GeneratePKCECodes()
 	if err != nil {
-		log.Fatalf("Failed to generate PKCE codes: %v", err)
+		log.Errorf("Failed to generate PKCE codes: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate PKCE codes"})
 		return
 	}
 
 	// Generate random state parameter
 	state, err := misc.GenerateRandomState()
 	if err != nil {
-		log.Fatalf("Failed to generate state parameter: %v", err)
+		log.Errorf("Failed to generate state parameter: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate state parameter"})
 		return
 	}
 
@@ -1148,7 +1153,8 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 	// Generate authorization URL
 	authURL, err := openaiAuth.GenerateAuthURL(state, pkceCodes)
 	if err != nil {
-		log.Fatalf("Failed to generate authorization URL: %v", err)
+		log.Errorf("Failed to generate authorization URL: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate authorization url"})
 		return
 	}
 
@@ -1283,7 +1289,7 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
 			oauthStatus[state] = "Failed to save authentication tokens"
-			log.Fatalf("Failed to save authentication tokens: %v", errSave)
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
 			return
 		}
 		fmt.Printf("Authentication successful! Token saved to %s\n", savedPath)
@@ -1318,7 +1324,8 @@ func (h *Handler) RequestAntigravityToken(c *gin.Context) {
 
 	state, errState := misc.GenerateRandomState()
 	if errState != nil {
-		log.Fatalf("Failed to generate state parameter: %v", errState)
+		log.Errorf("Failed to generate state parameter: %v", errState)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate state parameter"})
 		return
 	}
 
@@ -1514,7 +1521,7 @@ func (h *Handler) RequestAntigravityToken(c *gin.Context) {
 		}
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
-			log.Fatalf("Failed to save token to file: %v", errSave)
+			log.Errorf("Failed to save token to file: %v", errSave)
 			oauthStatus[state] = "Failed to save token to file"
 			return
 		}
@@ -1543,7 +1550,8 @@ func (h *Handler) RequestQwenToken(c *gin.Context) {
 	// Generate authorization URL
 	deviceFlow, err := qwenAuth.InitiateDeviceFlow(ctx)
 	if err != nil {
-		log.Fatalf("Failed to generate authorization URL: %v", err)
+		log.Errorf("Failed to generate authorization URL: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate authorization url"})
 		return
 	}
 	authURL := deviceFlow.VerificationURIComplete
@@ -1570,7 +1578,7 @@ func (h *Handler) RequestQwenToken(c *gin.Context) {
 		}
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
-			log.Fatalf("Failed to save authentication tokens: %v", errSave)
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
 			oauthStatus[state] = "Failed to save authentication tokens"
 			return
 		}
@@ -1674,7 +1682,7 @@ func (h *Handler) RequestIFlowToken(c *gin.Context) {
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
 			oauthStatus[state] = "Failed to save authentication tokens"
-			log.Fatalf("Failed to save authentication tokens: %v", errSave)
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
 			return
 		}
 
@@ -2103,6 +2111,7 @@ func checkCloudAPIIsEnabled(ctx context.Context, httpClient *http.Client, projec
 				continue
 			}
 		}
+		_ = resp.Body.Close()
 		return false, fmt.Errorf("project activation required: %s", errMessage)
 	}
 	return true, nil

internal/api/handlers/management/config_lists.go

@@ -706,3 +706,155 @@ func normalizeClaudeKey(entry *config.ClaudeKey) {
 	}
 	entry.Models = normalized
 }
+
+// GetAmpCode returns the complete ampcode configuration.
+func (h *Handler) GetAmpCode(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"ampcode": config.AmpCode{}})
+		return
+	}
+	c.JSON(200, gin.H{"ampcode": h.cfg.AmpCode})
+}
+
+// GetAmpUpstreamURL returns the ampcode upstream URL.
+func (h *Handler) GetAmpUpstreamURL(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"upstream-url": ""})
+		return
+	}
+	c.JSON(200, gin.H{"upstream-url": h.cfg.AmpCode.UpstreamURL})
+}
+
+// PutAmpUpstreamURL updates the ampcode upstream URL.
+func (h *Handler) PutAmpUpstreamURL(c *gin.Context) {
+	h.updateStringField(c, func(v string) { h.cfg.AmpCode.UpstreamURL = strings.TrimSpace(v) })
+}
+
+// DeleteAmpUpstreamURL clears the ampcode upstream URL.
+func (h *Handler) DeleteAmpUpstreamURL(c *gin.Context) {
+	h.cfg.AmpCode.UpstreamURL = ""
+	h.persist(c)
+}
+
+// GetAmpUpstreamAPIKey returns the ampcode upstream API key.
+func (h *Handler) GetAmpUpstreamAPIKey(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"upstream-api-key": ""})
+		return
+	}
+	c.JSON(200, gin.H{"upstream-api-key": h.cfg.AmpCode.UpstreamAPIKey})
+}
+
+// PutAmpUpstreamAPIKey updates the ampcode upstream API key.
+func (h *Handler) PutAmpUpstreamAPIKey(c *gin.Context) {
+	h.updateStringField(c, func(v string) { h.cfg.AmpCode.UpstreamAPIKey = strings.TrimSpace(v) })
+}
+
+// DeleteAmpUpstreamAPIKey clears the ampcode upstream API key.
+func (h *Handler) DeleteAmpUpstreamAPIKey(c *gin.Context) {
+	h.cfg.AmpCode.UpstreamAPIKey = ""
+	h.persist(c)
+}
+
+// GetAmpRestrictManagementToLocalhost returns the localhost restriction setting.
+func (h *Handler) GetAmpRestrictManagementToLocalhost(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"restrict-management-to-localhost": true})
+		return
+	}
+	c.JSON(200, gin.H{"restrict-management-to-localhost": h.cfg.AmpCode.RestrictManagementToLocalhost})
+}
+
+// PutAmpRestrictManagementToLocalhost updates the localhost restriction setting.
+func (h *Handler) PutAmpRestrictManagementToLocalhost(c *gin.Context) {
+	h.updateBoolField(c, func(v bool) { h.cfg.AmpCode.RestrictManagementToLocalhost = v })
+}
+
+// GetAmpModelMappings returns the ampcode model mappings.
+func (h *Handler) GetAmpModelMappings(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"model-mappings": []config.AmpModelMapping{}})
+		return
+	}
+	c.JSON(200, gin.H{"model-mappings": h.cfg.AmpCode.ModelMappings})
+}
+
+// PutAmpModelMappings replaces all ampcode model mappings.
+func (h *Handler) PutAmpModelMappings(c *gin.Context) {
+	var body struct {
+		Value []config.AmpModelMapping `json:"value"`
+	}
+	if err := c.ShouldBindJSON(&body); err != nil {
+		c.JSON(400, gin.H{"error": "invalid body"})
+		return
+	}
+	h.cfg.AmpCode.ModelMappings = body.Value
+	h.persist(c)
+}
+
+// PatchAmpModelMappings adds or updates model mappings.
+func (h *Handler) PatchAmpModelMappings(c *gin.Context) {
+	var body struct {
+		Value []config.AmpModelMapping `json:"value"`
+	}
+	if err := c.ShouldBindJSON(&body); err != nil {
+		c.JSON(400, gin.H{"error": "invalid body"})
+		return
+	}
+
+	existing := make(map[string]int)
+	for i, m := range h.cfg.AmpCode.ModelMappings {
+		existing[strings.TrimSpace(m.From)] = i
+	}
+
+	for _, newMapping := range body.Value {
+		from := strings.TrimSpace(newMapping.From)
+		if idx, ok := existing[from]; ok {
+			h.cfg.AmpCode.ModelMappings[idx] = newMapping
+		} else {
+			h.cfg.AmpCode.ModelMappings = append(h.cfg.AmpCode.ModelMappings, newMapping)
+			existing[from] = len(h.cfg.AmpCode.ModelMappings) - 1
+		}
+	}
+	h.persist(c)
+}
+
+// DeleteAmpModelMappings removes specified model mappings by "from" field.
+func (h *Handler) DeleteAmpModelMappings(c *gin.Context) {
+	var body struct {
+		Value []string `json:"value"`
+	}
+	if err := c.ShouldBindJSON(&body); err != nil || len(body.Value) == 0 {
+		h.cfg.AmpCode.ModelMappings = nil
+		h.persist(c)
+		return
+	}
+
+	toRemove := make(map[string]bool)
+	for _, from := range body.Value {
+		toRemove[strings.TrimSpace(from)] = true
+	}
+
+	newMappings := make([]config.AmpModelMapping, 0, len(h.cfg.AmpCode.ModelMappings))
+	for _, m := range h.cfg.AmpCode.ModelMappings {
+		if !toRemove[strings.TrimSpace(m.From)] {
+			newMappings = append(newMappings, m)
+		}
+	}
+	h.cfg.AmpCode.ModelMappings = newMappings
+	h.persist(c)
+}
+
+// GetAmpForceModelMappings returns whether model mappings are forced.
+func (h *Handler) GetAmpForceModelMappings(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"force-model-mappings": false})
+		return
+	}
+	c.JSON(200, gin.H{"force-model-mappings": h.cfg.AmpCode.ForceModelMappings})
+}
+
+// PutAmpForceModelMappings updates the force model mappings setting.
+func (h *Handler) PutAmpForceModelMappings(c *gin.Context) {
+	h.updateBoolField(c, func(v bool) { h.cfg.AmpCode.ForceModelMappings = v })
+}

internal/api/handlers/management/handler.go

@@ -240,16 +240,6 @@ func (h *Handler) updateBoolField(c *gin.Context, set func(bool)) {
 		Value *bool `json:"value"`
 	}
 	if err := c.ShouldBindJSON(&body); err != nil || body.Value == nil {
-		var m map[string]any
-		if err2 := c.ShouldBindJSON(&m); err2 == nil {
-			for _, v := range m {
-				if b, ok := v.(bool); ok {
-					set(b)
-					h.persist(c)
-					return
-				}
-			}
-		}
 		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid body"})
 		return
 	}

internal/api/middleware/response_writer.go

@@ -232,7 +232,16 @@ func (w *ResponseWriterWrapper) Finalize(c *gin.Context) error {
 			w.streamDone = nil
 		}
 
+		// Write API Request and Response to the streaming log before closing
 		if w.streamWriter != nil {
+			apiRequest := w.extractAPIRequest(c)
+			if len(apiRequest) > 0 {
+				_ = w.streamWriter.WriteAPIRequest(apiRequest)
+			}
+			apiResponse := w.extractAPIResponse(c)
+			if len(apiResponse) > 0 {
+				_ = w.streamWriter.WriteAPIResponse(apiResponse)
+			}
 			if err := w.streamWriter.Close(); err != nil {
 				w.streamWriter = nil
 				return err

internal/api/modules/amp/amp.go

@@ -100,6 +100,16 @@ func (m *AmpModule) Name() string {
 	return "amp-routing"
 }
 
+// forceModelMappings returns whether model mappings should take precedence over local API keys
+func (m *AmpModule) forceModelMappings() bool {
+	m.configMu.RLock()
+	defer m.configMu.RUnlock()
+	if m.lastConfig == nil {
+		return false
+	}
+	return m.lastConfig.ForceModelMappings
+}
+
 // Register sets up Amp routes if configured.
 // This implements the RouteModuleV2 interface with Context.
 // Routes are registered only once via sync.Once for idempotent behavior.

internal/api/modules/amp/fallback_handlers.go

@@ -28,6 +28,9 @@ const (
 	RouteTypeNoProvider AmpRouteType = "NO_PROVIDER"
 )
 
+// MappedModelContextKey is the Gin context key for passing mapped model names.
+const MappedModelContextKey = "mapped_model"
+
 // logAmpRouting logs the routing decision for an Amp request with structured fields
 func logAmpRouting(routeType AmpRouteType, requestedModel, resolvedModel, provider, path string) {
 	fields := log.Fields{
@@ -74,23 +77,29 @@ func logAmpRouting(routeType AmpRouteType, requestedModel, resolvedModel, provid
 // FallbackHandler wraps a standard handler with fallback logic to ampcode.com
 // when the model's provider is not available in CLIProxyAPI
 type FallbackHandler struct {
-	getProxy    func() *httputil.ReverseProxy
-	modelMapper ModelMapper
+	getProxy           func() *httputil.ReverseProxy
+	modelMapper        ModelMapper
+	forceModelMappings func() bool
 }
 
 // NewFallbackHandler creates a new fallback handler wrapper
 // The getProxy function allows lazy evaluation of the proxy (useful when proxy is created after routes)
 func NewFallbackHandler(getProxy func() *httputil.ReverseProxy) *FallbackHandler {
 	return &FallbackHandler{
-		getProxy: getProxy,
+		getProxy:           getProxy,
+		forceModelMappings: func() bool { return false },
 	}
 }
 
 // NewFallbackHandlerWithMapper creates a new fallback handler with model mapping support
-func NewFallbackHandlerWithMapper(getProxy func() *httputil.ReverseProxy, mapper ModelMapper) *FallbackHandler {
+func NewFallbackHandlerWithMapper(getProxy func() *httputil.ReverseProxy, mapper ModelMapper, forceModelMappings func() bool) *FallbackHandler {
+	if forceModelMappings == nil {
+		forceModelMappings = func() bool { return false }
+	}
 	return &FallbackHandler{
-		getProxy:    getProxy,
-		modelMapper: mapper,
+		getProxy:           getProxy,
+		modelMapper:        mapper,
+		forceModelMappings: forceModelMappings,
 	}
 }
 
@@ -124,35 +133,68 @@ func (fh *FallbackHandler) WrapHandler(handler gin.HandlerFunc) gin.HandlerFunc
 			return
 		}
 
-		// Normalize model (handles Gemini thinking suffixes)
-		normalizedModel, _ := util.NormalizeGeminiThinkingModel(modelName)
-
-		// Check if we have providers for this model
-		providers := util.GetProviderName(normalizedModel)
+		// Normalize model (handles dynamic thinking suffixes)
+		normalizedModel, _ := util.NormalizeThinkingModel(modelName)
 
 		// Track resolved model for logging (may change if mapping is applied)
 		resolvedModel := normalizedModel
 		usedMapping := false
+		var providers []string
 
-		if len(providers) == 0 {
-			// No providers configured - check if we have a model mapping
+		// Check if model mappings should be forced ahead of local API keys
+		forceMappings := fh.forceModelMappings != nil && fh.forceModelMappings()
+
+		if forceMappings {
+			// FORCE MODE: Check model mappings FIRST (takes precedence over local API keys)
+			// This allows users to route Amp requests to their preferred OAuth providers
 			if fh.modelMapper != nil {
 				if mappedModel := fh.modelMapper.MapModel(normalizedModel); mappedModel != "" {
-					// Mapping found - rewrite the model in request body
-					bodyBytes = rewriteModelInRequest(bodyBytes, mappedModel)
-					c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
-					resolvedModel = mappedModel
-					usedMapping = true
-
-					// Get providers for the mapped model
-					providers = util.GetProviderName(mappedModel)
-
-					// Continue to handler with remapped model
-					goto handleRequest
+					// Mapping found - check if we have a provider for the mapped model
+					mappedProviders := util.GetProviderName(mappedModel)
+					if len(mappedProviders) > 0 {
+						// Mapping found and provider available - rewrite the model in request body
+						bodyBytes = rewriteModelInRequest(bodyBytes, mappedModel)
+						c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+						// Store mapped model in context for handlers that check it (like gemini bridge)
+						c.Set(MappedModelContextKey, mappedModel)
+						resolvedModel = mappedModel
+						usedMapping = true
+						providers = mappedProviders
+					}
 				}
 			}
 
-			// No mapping found - check if we have a proxy for fallback
+			// If no mapping applied, check for local providers
+			if !usedMapping {
+				providers = util.GetProviderName(normalizedModel)
+			}
+		} else {
+			// DEFAULT MODE: Check local providers first, then mappings as fallback
+			providers = util.GetProviderName(normalizedModel)
+
+			if len(providers) == 0 {
+				// No providers configured - check if we have a model mapping
+				if fh.modelMapper != nil {
+					if mappedModel := fh.modelMapper.MapModel(normalizedModel); mappedModel != "" {
+						// Mapping found - check if we have a provider for the mapped model
+						mappedProviders := util.GetProviderName(mappedModel)
+						if len(mappedProviders) > 0 {
+							// Mapping found and provider available - rewrite the model in request body
+							bodyBytes = rewriteModelInRequest(bodyBytes, mappedModel)
+							c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+							// Store mapped model in context for handlers that check it (like gemini bridge)
+							c.Set(MappedModelContextKey, mappedModel)
+							resolvedModel = mappedModel
+							usedMapping = true
+							providers = mappedProviders
+						}
+					}
+				}
+			}
+		}
+
+		// If no providers available, fallback to ampcode.com
+		if len(providers) == 0 {
 			proxy := fh.getProxy()
 			if proxy != nil {
 				// Log: Forwarding to ampcode.com (uses Amp credits)
@@ -170,8 +212,6 @@ func (fh *FallbackHandler) WrapHandler(handler gin.HandlerFunc) gin.HandlerFunc
 			logAmpRouting(RouteTypeNoProvider, modelName, "", "", requestPath)
 		}
 
-	handleRequest:
-
 		// Log the routing decision
 		providerName := ""
 		if len(providers) > 0 {

internal/api/modules/amp/gemini_bridge.go

@@ -4,7 +4,6 @@ import (
 	"strings"
 
 	"github.com/gin-gonic/gin"
-	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers/gemini"
 )
 
 // createGeminiBridgeHandler creates a handler that bridges AMP CLI's non-standard Gemini paths
@@ -15,16 +14,31 @@ import (
 //
 // This extracts the model+method from the AMP path and sets it as the :action parameter
 // so the standard Gemini handler can process it.
-func createGeminiBridgeHandler(geminiHandler *gemini.GeminiAPIHandler) gin.HandlerFunc {
+//
+// The handler parameter should be a Gemini-compatible handler that expects the :action param.
+func createGeminiBridgeHandler(handler gin.HandlerFunc) gin.HandlerFunc {
 	return func(c *gin.Context) {
 		// Get the full path from the catch-all parameter
 		path := c.Param("path")
 
 		// Extract model:method from AMP CLI path format
 		// Example: /publishers/google/models/gemini-3-pro-preview:streamGenerateContent
-		if idx := strings.Index(path, "/models/"); idx >= 0 {
-			// Extract everything after "/models/"
-			actionPart := path[idx+8:] // Skip "/models/"
+		const modelsPrefix = "/models/"
+		if idx := strings.Index(path, modelsPrefix); idx >= 0 {
+			// Extract everything after modelsPrefix
+			actionPart := path[idx+len(modelsPrefix):]
+
+			// Check if model was mapped by FallbackHandler
+			if mappedModel, exists := c.Get(MappedModelContextKey); exists {
+				if strModel, ok := mappedModel.(string); ok && strModel != "" {
+					// Replace the model part in the action
+					// actionPart is like "model-name:method"
+					if colonIdx := strings.Index(actionPart, ":"); colonIdx > 0 {
+						method := actionPart[colonIdx:] // ":method"
+						actionPart = strModel + method
+					}
+				}
+			}
 
 			// Set this as the :action parameter that the Gemini handler expects
 			c.Params = append(c.Params, gin.Param{
@@ -32,8 +46,8 @@ func createGeminiBridgeHandler(geminiHandler *gemini.GeminiAPIHandler) gin.Handl
 				Value: actionPart,
 			})
 
-			// Call the standard Gemini handler
-			geminiHandler.GeminiHandler(c)
+			// Call the handler
+			handler(c)
 			return
 		}
 

internal/api/modules/amp/gemini_bridge_test.go

@@ -0,0 +1,93 @@
+package amp
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+)
+
+func TestCreateGeminiBridgeHandler_ActionParameterExtraction(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	tests := []struct {
+		name           string
+		path           string
+		mappedModel    string // empty string means no mapping
+		expectedAction string
+	}{
+		{
+			name:           "no_mapping_uses_url_model",
+			path:           "/publishers/google/models/gemini-pro:generateContent",
+			mappedModel:    "",
+			expectedAction: "gemini-pro:generateContent",
+		},
+		{
+			name:           "mapped_model_replaces_url_model",
+			path:           "/publishers/google/models/gemini-exp:generateContent",
+			mappedModel:    "gemini-2.0-flash",
+			expectedAction: "gemini-2.0-flash:generateContent",
+		},
+		{
+			name:           "mapping_preserves_method",
+			path:           "/publishers/google/models/gemini-2.5-preview:streamGenerateContent",
+			mappedModel:    "gemini-flash",
+			expectedAction: "gemini-flash:streamGenerateContent",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var capturedAction string
+
+			mockGeminiHandler := func(c *gin.Context) {
+				capturedAction = c.Param("action")
+				c.JSON(http.StatusOK, gin.H{"captured": capturedAction})
+			}
+
+			// Use the actual createGeminiBridgeHandler function
+			bridgeHandler := createGeminiBridgeHandler(mockGeminiHandler)
+
+			r := gin.New()
+			if tt.mappedModel != "" {
+				r.Use(func(c *gin.Context) {
+					c.Set(MappedModelContextKey, tt.mappedModel)
+					c.Next()
+				})
+			}
+			r.POST("/api/provider/google/v1beta1/*path", bridgeHandler)
+
+			req := httptest.NewRequest(http.MethodPost, "/api/provider/google/v1beta1"+tt.path, nil)
+			w := httptest.NewRecorder()
+			r.ServeHTTP(w, req)
+
+			if w.Code != http.StatusOK {
+				t.Fatalf("Expected status 200, got %d", w.Code)
+			}
+			if capturedAction != tt.expectedAction {
+				t.Errorf("Expected action '%s', got '%s'", tt.expectedAction, capturedAction)
+			}
+		})
+	}
+}
+
+func TestCreateGeminiBridgeHandler_InvalidPath(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	mockHandler := func(c *gin.Context) {
+		c.JSON(http.StatusOK, gin.H{"ok": true})
+	}
+	bridgeHandler := createGeminiBridgeHandler(mockHandler)
+
+	r := gin.New()
+	r.POST("/api/provider/google/v1beta1/*path", bridgeHandler)
+
+	req := httptest.NewRequest(http.MethodPost, "/api/provider/google/v1beta1/invalid/path", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status 400 for invalid path, got %d", w.Code)
+	}
+}

internal/api/modules/amp/routes.go

@@ -9,7 +9,6 @@ import (
 
 	"github.com/gin-gonic/gin"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/logging"
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers"
 	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers/claude"
 	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers/gemini"
@@ -157,6 +156,7 @@ func (m *AmpModule) registerManagementRoutes(engine *gin.Engine, baseHandler *ha
 	rootMiddleware := []gin.HandlerFunc{m.managementAvailabilityMiddleware(), noCORSMiddleware(), m.localhostOnlyMiddleware()}
 	engine.GET("/threads/*path", append(rootMiddleware, proxyHandler)...)
 	engine.GET("/threads.rss", append(rootMiddleware, proxyHandler)...)
+	engine.GET("/news.rss", append(rootMiddleware, proxyHandler)...)
 
 	// Root-level auth routes for CLI login flow
 	// Amp uses multiple auth routes: /auth/cli-login, /auth/callback, /auth/sign-in, /auth/logout
@@ -169,30 +169,22 @@ func (m *AmpModule) registerManagementRoutes(engine *gin.Engine, baseHandler *ha
 	// We bridge these to our standard Gemini handler to enable local OAuth.
 	// If no local OAuth is available, falls back to ampcode.com proxy.
 	geminiHandlers := gemini.NewGeminiAPIHandler(baseHandler)
-	geminiBridge := createGeminiBridgeHandler(geminiHandlers)
-	geminiV1Beta1Fallback := NewFallbackHandler(func() *httputil.ReverseProxy {
+	geminiBridge := createGeminiBridgeHandler(geminiHandlers.GeminiHandler)
+	geminiV1Beta1Fallback := NewFallbackHandlerWithMapper(func() *httputil.ReverseProxy {
 		return m.getProxy()
-	})
+	}, m.modelMapper, m.forceModelMappings)
 	geminiV1Beta1Handler := geminiV1Beta1Fallback.WrapHandler(geminiBridge)
 
-	// Route POST model calls through Gemini bridge when a local provider exists, otherwise proxy.
+	// Route POST model calls through Gemini bridge with FallbackHandler.
+	// FallbackHandler checks provider -> mapping -> proxy fallback automatically.
 	// All other methods (e.g., GET model listing) always proxy to upstream to preserve Amp CLI behavior.
 	ampAPI.Any("/provider/google/v1beta1/*path", func(c *gin.Context) {
 		if c.Request.Method == "POST" {
-			// Attempt to extract the model name from the AMP-style path
 			if path := c.Param("path"); strings.Contains(path, "/models/") {
-				modelPart := path[strings.Index(path, "/models/")+len("/models/"):]
-				if colonIdx := strings.Index(modelPart, ":"); colonIdx > 0 {
-					modelPart = modelPart[:colonIdx]
-				}
-				if modelPart != "" {
-					normalized, _ := util.NormalizeGeminiThinkingModel(modelPart)
-					// Only handle locally when we have a provider; otherwise fall back to proxy
-					if providers := util.GetProviderName(normalized); len(providers) > 0 {
-						geminiV1Beta1Handler(c)
-						return
-					}
-				}
+				// POST with /models/ path -> use Gemini bridge with fallback handler
+				// FallbackHandler will check provider/mapping and proxy if needed
+				geminiV1Beta1Handler(c)
+				return
 			}
 		}
 		// Non-POST or no local provider available -> proxy upstream
@@ -218,7 +210,7 @@ func (m *AmpModule) registerProviderAliases(engine *gin.Engine, baseHandler *han
 	// Also includes model mapping support for routing unavailable models to alternatives
 	fallbackHandler := NewFallbackHandlerWithMapper(func() *httputil.ReverseProxy {
 		return m.getProxy()
-	}, m.modelMapper)
+	}, m.modelMapper, m.forceModelMappings)
 
 	// Provider-specific routes under /api/provider/:provider
 	ampProviders := engine.Group("/api/provider")

internal/api/server.go

@@ -300,7 +300,7 @@ func NewServer(cfg *config.Config, authManager *auth.Manager, accessManager *sdk
 
 	// Create HTTP server
 	s.server = &http.Server{
-		Addr:    fmt.Sprintf(":%d", cfg.Port),
+		Addr:    fmt.Sprintf("%s:%d", cfg.Host, cfg.Port),
 		Handler: engine,
 	}
 
@@ -520,6 +520,26 @@ func (s *Server) registerManagementRoutes() {
 		mgmt.PUT("/ws-auth", s.mgmt.PutWebsocketAuth)
 		mgmt.PATCH("/ws-auth", s.mgmt.PutWebsocketAuth)
 
+		mgmt.GET("/ampcode", s.mgmt.GetAmpCode)
+		mgmt.GET("/ampcode/upstream-url", s.mgmt.GetAmpUpstreamURL)
+		mgmt.PUT("/ampcode/upstream-url", s.mgmt.PutAmpUpstreamURL)
+		mgmt.PATCH("/ampcode/upstream-url", s.mgmt.PutAmpUpstreamURL)
+		mgmt.DELETE("/ampcode/upstream-url", s.mgmt.DeleteAmpUpstreamURL)
+		mgmt.GET("/ampcode/upstream-api-key", s.mgmt.GetAmpUpstreamAPIKey)
+		mgmt.PUT("/ampcode/upstream-api-key", s.mgmt.PutAmpUpstreamAPIKey)
+		mgmt.PATCH("/ampcode/upstream-api-key", s.mgmt.PutAmpUpstreamAPIKey)
+		mgmt.DELETE("/ampcode/upstream-api-key", s.mgmt.DeleteAmpUpstreamAPIKey)
+		mgmt.GET("/ampcode/restrict-management-to-localhost", s.mgmt.GetAmpRestrictManagementToLocalhost)
+		mgmt.PUT("/ampcode/restrict-management-to-localhost", s.mgmt.PutAmpRestrictManagementToLocalhost)
+		mgmt.PATCH("/ampcode/restrict-management-to-localhost", s.mgmt.PutAmpRestrictManagementToLocalhost)
+		mgmt.GET("/ampcode/model-mappings", s.mgmt.GetAmpModelMappings)
+		mgmt.PUT("/ampcode/model-mappings", s.mgmt.PutAmpModelMappings)
+		mgmt.PATCH("/ampcode/model-mappings", s.mgmt.PatchAmpModelMappings)
+		mgmt.DELETE("/ampcode/model-mappings", s.mgmt.DeleteAmpModelMappings)
+		mgmt.GET("/ampcode/force-model-mappings", s.mgmt.GetAmpForceModelMappings)
+		mgmt.PUT("/ampcode/force-model-mappings", s.mgmt.PutAmpForceModelMappings)
+		mgmt.PATCH("/ampcode/force-model-mappings", s.mgmt.PutAmpForceModelMappings)
+
 		mgmt.GET("/request-retry", s.mgmt.GetRequestRetry)
 		mgmt.PUT("/request-retry", s.mgmt.PutRequestRetry)
 		mgmt.PATCH("/request-retry", s.mgmt.PutRequestRetry)

internal/auth/gemini/gemini_auth.go

@@ -76,7 +76,8 @@ func (g *GeminiAuth) GetAuthenticatedClient(ctx context.Context, ts *GeminiToken
 			auth := &proxy.Auth{User: username, Password: password}
 			dialer, errSOCKS5 := proxy.SOCKS5("tcp", proxyURL.Host, auth, proxy.Direct)
 			if errSOCKS5 != nil {
-				log.Fatalf("create SOCKS5 dialer failed: %v", errSOCKS5)
+				log.Errorf("create SOCKS5 dialer failed: %v", errSOCKS5)
+				return nil, fmt.Errorf("create SOCKS5 dialer failed: %w", errSOCKS5)
 			}
 			transport = &http.Transport{
 				DialContext: func(ctx context.Context, network, addr string) (net.Conn, error) {
@@ -238,7 +239,11 @@ func (g *GeminiAuth) getTokenFromWeb(ctx context.Context, config *oauth2.Config,
 	// Start the server in a goroutine.
 	go func() {
 		if err := server.ListenAndServe(); !errors.Is(err, http.ErrServerClosed) {
-			log.Fatalf("ListenAndServe(): %v", err)
+			log.Errorf("ListenAndServe(): %v", err)
+			select {
+			case errChan <- err:
+			default:
+			}
 		}
 	}()
 

internal/auth/iflow/iflow_auth.go

@@ -309,17 +309,23 @@ func (ia *IFlowAuth) AuthenticateWithCookie(ctx context.Context, cookie string)
 		return nil, fmt.Errorf("iflow cookie authentication: cookie is empty")
 	}
 
-	// First, get initial API key information using GET request
+	// First, get initial API key information using GET request to obtain the name
 	keyInfo, err := ia.fetchAPIKeyInfo(ctx, cookie)
 	if err != nil {
 		return nil, fmt.Errorf("iflow cookie authentication: fetch initial API key info failed: %w", err)
 	}
 
-	// Convert to token data format
+	// Refresh the API key using POST request
+	refreshedKeyInfo, err := ia.RefreshAPIKey(ctx, cookie, keyInfo.Name)
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie authentication: refresh API key failed: %w", err)
+	}
+
+	// Convert to token data format using refreshed key
 	data := &IFlowTokenData{
-		APIKey: keyInfo.APIKey,
-		Expire: keyInfo.ExpireTime,
-		Email:  keyInfo.Name,
+		APIKey: refreshedKeyInfo.APIKey,
+		Expire: refreshedKeyInfo.ExpireTime,
+		Email:  refreshedKeyInfo.Name,
 		Cookie: cookie,
 	}
 

internal/cmd/login.go

@@ -65,20 +65,20 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 	authenticator := sdkAuth.NewGeminiAuthenticator()
 	record, errLogin := authenticator.Login(ctx, cfg, loginOpts)
 	if errLogin != nil {
-		log.Fatalf("Gemini authentication failed: %v", errLogin)
+		log.Errorf("Gemini authentication failed: %v", errLogin)
 		return
 	}
 
 	storage, okStorage := record.Storage.(*gemini.GeminiTokenStorage)
 	if !okStorage || storage == nil {
-		log.Fatal("Gemini authentication failed: unsupported token storage")
+		log.Error("Gemini authentication failed: unsupported token storage")
 		return
 	}
 
 	geminiAuth := gemini.NewGeminiAuth()
 	httpClient, errClient := geminiAuth.GetAuthenticatedClient(ctx, storage, cfg, options.NoBrowser)
 	if errClient != nil {
-		log.Fatalf("Gemini authentication failed: %v", errClient)
+		log.Errorf("Gemini authentication failed: %v", errClient)
 		return
 	}
 
@@ -86,7 +86,7 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 
 	projects, errProjects := fetchGCPProjects(ctx, httpClient)
 	if errProjects != nil {
-		log.Fatalf("Failed to get project list: %v", errProjects)
+		log.Errorf("Failed to get project list: %v", errProjects)
 		return
 	}
 
@@ -98,11 +98,11 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 	selectedProjectID := promptForProjectSelection(projects, strings.TrimSpace(projectID), promptFn)
 	projectSelections, errSelection := resolveProjectSelections(selectedProjectID, projects)
 	if errSelection != nil {
-		log.Fatalf("Invalid project selection: %v", errSelection)
+		log.Errorf("Invalid project selection: %v", errSelection)
 		return
 	}
 	if len(projectSelections) == 0 {
-		log.Fatal("No project selected; aborting login.")
+		log.Error("No project selected; aborting login.")
 		return
 	}
 
@@ -116,7 +116,7 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 				showProjectSelectionHelp(storage.Email, projects)
 				return
 			}
-			log.Fatalf("Failed to complete user setup: %v", errSetup)
+			log.Errorf("Failed to complete user setup: %v", errSetup)
 			return
 		}
 		finalID := strings.TrimSpace(storage.ProjectID)
@@ -133,11 +133,11 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 		for _, pid := range activatedProjects {
 			isChecked, errCheck := checkCloudAPIIsEnabled(ctx, httpClient, pid)
 			if errCheck != nil {
-				log.Fatalf("Failed to check if Cloud AI API is enabled for %s: %v", pid, errCheck)
+				log.Errorf("Failed to check if Cloud AI API is enabled for %s: %v", pid, errCheck)
 				return
 			}
 			if !isChecked {
-				log.Fatalf("Failed to check if Cloud AI API is enabled for project %s. If you encounter an error message, please create an issue.", pid)
+				log.Errorf("Failed to check if Cloud AI API is enabled for project %s. If you encounter an error message, please create an issue.", pid)
 				return
 			}
 		}
@@ -153,7 +153,7 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 
 	savedPath, errSave := store.Save(ctx, record)
 	if errSave != nil {
-		log.Fatalf("Failed to save token to file: %v", errSave)
+		log.Errorf("Failed to save token to file: %v", errSave)
 		return
 	}
 
@@ -555,6 +555,7 @@ func checkCloudAPIIsEnabled(ctx context.Context, httpClient *http.Client, projec
 				continue
 			}
 		}
+		_ = resp.Body.Close()
 		return false, fmt.Errorf("project activation required: %s", errMessage)
 	}
 	return true, nil

internal/cmd/run.go

@@ -45,12 +45,13 @@ func StartService(cfg *config.Config, configPath string, localPassword string) {
 
 	service, err := builder.Build()
 	if err != nil {
-		log.Fatalf("failed to build proxy service: %v", err)
+		log.Errorf("failed to build proxy service: %v", err)
+		return
 	}
 
 	err = service.Run(runCtx)
 	if err != nil && !errors.Is(err, context.Canceled) {
-		log.Fatalf("proxy service exited with error: %v", err)
+		log.Errorf("proxy service exited with error: %v", err)
 	}
 }
 

internal/cmd/vertex_import.go

@@ -29,30 +29,30 @@ func DoVertexImport(cfg *config.Config, keyPath string) {
 	}
 	rawPath := strings.TrimSpace(keyPath)
 	if rawPath == "" {
-		log.Fatalf("vertex-import: missing service account key path")
+		log.Errorf("vertex-import: missing service account key path")
 		return
 	}
 	data, errRead := os.ReadFile(rawPath)
 	if errRead != nil {
-		log.Fatalf("vertex-import: read file failed: %v", errRead)
+		log.Errorf("vertex-import: read file failed: %v", errRead)
 		return
 	}
 	var sa map[string]any
 	if errUnmarshal := json.Unmarshal(data, &sa); errUnmarshal != nil {
-		log.Fatalf("vertex-import: invalid service account json: %v", errUnmarshal)
+		log.Errorf("vertex-import: invalid service account json: %v", errUnmarshal)
 		return
 	}
 	// Validate and normalize private_key before saving
 	normalizedSA, errFix := vertex.NormalizeServiceAccountMap(sa)
 	if errFix != nil {
-		log.Fatalf("vertex-import: %v", errFix)
+		log.Errorf("vertex-import: %v", errFix)
 		return
 	}
 	sa = normalizedSA
 	email, _ := sa["client_email"].(string)
 	projectID, _ := sa["project_id"].(string)
 	if strings.TrimSpace(projectID) == "" {
-		log.Fatalf("vertex-import: project_id missing in service account json")
+		log.Errorf("vertex-import: project_id missing in service account json")
 		return
 	}
 	if strings.TrimSpace(email) == "" {
@@ -92,7 +92,7 @@ func DoVertexImport(cfg *config.Config, keyPath string) {
 	}
 	path, errSave := store.Save(context.Background(), record)
 	if errSave != nil {
-		log.Fatalf("vertex-import: save credential failed: %v", errSave)
+		log.Errorf("vertex-import: save credential failed: %v", errSave)
 		return
 	}
 	fmt.Printf("Vertex credentials imported: %s\n", path)

internal/config/config.go

@@ -20,6 +20,9 @@ import (
 // Config represents the application's configuration, loaded from a YAML file.
 type Config struct {
 	config.SDKConfig `yaml:",inline"`
+	// Host is the network host/interface on which the API server will bind.
+	// Default is empty ("") to bind all interfaces (IPv4 + IPv6). Use "127.0.0.1" or "localhost" for local-only access.
+	Host string `yaml:"host" json:"-"`
 	// Port is the network port on which the API server will listen.
 	Port int `yaml:"port" json:"-"`
 
@@ -143,6 +146,10 @@ type AmpCode struct {
 	// When Amp requests a model that isn't available locally, these mappings
 	// allow routing to an alternative model that IS available.
 	ModelMappings []AmpModelMapping `yaml:"model-mappings" json:"model-mappings"`
+
+	// ForceModelMappings when true, model mappings take precedence over local API keys.
+	// When false (default), local API keys are used first if available.
+	ForceModelMappings bool `yaml:"force-model-mappings" json:"force-model-mappings"`
 }
 
 // PayloadConfig defines default and override parameter rules applied to provider payloads.
@@ -316,6 +323,7 @@ func LoadConfigOptional(configFile string, optional bool) (*Config, error) {
 	// Unmarshal the YAML data into the Config struct.
 	var cfg Config
 	// Set defaults before unmarshal so that absent keys keep defaults.
+	cfg.Host = "" // Default empty: binds to all interfaces (IPv4 + IPv6)
 	cfg.LoggingToFile = false
 	cfg.UsageStatisticsEnabled = false
 	cfg.DisableCooling = false

internal/interfaces/client_models.go

@@ -56,6 +56,8 @@ type Content struct {
 // Part represents a distinct piece of content within a message.
 // A part can be text, inline data (like an image), a function call, or a function response.
 type Part struct {
+	Thought bool `json:"thought,omitempty"`
+
 	// Text contains plain text content.
 	Text string `json:"text,omitempty"`
 

internal/logging/request_logger.go

@@ -20,6 +20,7 @@ import (
 	"github.com/klauspost/compress/zstd"
 	log "github.com/sirupsen/logrus"
 
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/buildinfo"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 )
@@ -83,6 +84,26 @@ type StreamingLogWriter interface {
 	//   - error: An error if writing fails, nil otherwise
 	WriteStatus(status int, headers map[string][]string) error
 
+	// WriteAPIRequest writes the upstream API request details to the log.
+	// This should be called before WriteStatus to maintain proper log ordering.
+	//
+	// Parameters:
+	//   - apiRequest: The API request data (typically includes URL, headers, body sent upstream)
+	//
+	// Returns:
+	//   - error: An error if writing fails, nil otherwise
+	WriteAPIRequest(apiRequest []byte) error
+
+	// WriteAPIResponse writes the upstream API response details to the log.
+	// This should be called after the streaming response is complete.
+	//
+	// Parameters:
+	//   - apiResponse: The API response data
+	//
+	// Returns:
+	//   - error: An error if writing fails, nil otherwise
+	WriteAPIResponse(apiResponse []byte) error
+
 	// Close finalizes the log file and cleans up resources.
 	//
 	// Returns:
@@ -247,10 +268,11 @@ func (l *FileRequestLogger) LogStreamingRequest(url, method string, headers map[
 
 	// Create streaming writer
 	writer := &FileStreamingLogWriter{
-		file:      file,
-		chunkChan: make(chan []byte, 100), // Buffered channel for async writes
-		closeChan: make(chan struct{}),
-		errorChan: make(chan error, 1),
+		file:           file,
+		chunkChan:      make(chan []byte, 100), // Buffered channel for async writes
+		closeChan:      make(chan struct{}),
+		errorChan:      make(chan error, 1),
+		bufferedChunks: &bytes.Buffer{},
 	}
 
 	// Start async writer goroutine
@@ -603,6 +625,7 @@ func (l *FileRequestLogger) formatRequestInfo(url, method string, headers map[st
 	var content strings.Builder
 
 	content.WriteString("=== REQUEST INFO ===\n")
+	content.WriteString(fmt.Sprintf("Version: %s\n", buildinfo.Version))
 	content.WriteString(fmt.Sprintf("URL: %s\n", url))
 	content.WriteString(fmt.Sprintf("Method: %s\n", method))
 	content.WriteString(fmt.Sprintf("Timestamp: %s\n", time.Now().Format(time.RFC3339Nano)))
@@ -626,11 +649,12 @@ func (l *FileRequestLogger) formatRequestInfo(url, method string, headers map[st
 
 // FileStreamingLogWriter implements StreamingLogWriter for file-based streaming logs.
 // It handles asynchronous writing of streaming response chunks to a file.
+// All data is buffered and written in the correct order when Close is called.
 type FileStreamingLogWriter struct {
 	// file is the file where log data is written.
 	file *os.File
 
-	// chunkChan is a channel for receiving response chunks to write.
+	// chunkChan is a channel for receiving response chunks to buffer.
 	chunkChan chan []byte
 
 	// closeChan is a channel for signaling when the writer is closed.
@@ -639,8 +663,23 @@ type FileStreamingLogWriter struct {
 	// errorChan is a channel for reporting errors during writing.
 	errorChan chan error
 
-	// statusWritten indicates whether the response status has been written.
+	// bufferedChunks stores the response chunks in order.
+	bufferedChunks *bytes.Buffer
+
+	// responseStatus stores the HTTP status code.
+	responseStatus int
+
+	// statusWritten indicates whether a non-zero status was recorded.
 	statusWritten bool
+
+	// responseHeaders stores the response headers.
+	responseHeaders map[string][]string
+
+	// apiRequest stores the upstream API request data.
+	apiRequest []byte
+
+	// apiResponse stores the upstream API response data.
+	apiResponse []byte
 }
 
 // WriteChunkAsync writes a response chunk asynchronously (non-blocking).
@@ -664,39 +703,65 @@ func (w *FileStreamingLogWriter) WriteChunkAsync(chunk []byte) {
 	}
 }
 
-// WriteStatus writes the response status and headers to the log.
+// WriteStatus buffers the response status and headers for later writing.
 //
 // Parameters:
 //   - status: The response status code
 //   - headers: The response headers
 //
 // Returns:
-//   - error: An error if writing fails, nil otherwise
+//   - error: Always returns nil (buffering cannot fail)
 func (w *FileStreamingLogWriter) WriteStatus(status int, headers map[string][]string) error {
-	if w.file == nil || w.statusWritten {
+	if status == 0 {
 		return nil
 	}
 
-	var content strings.Builder
-	content.WriteString("========================================\n")
-	content.WriteString("=== RESPONSE ===\n")
-	content.WriteString(fmt.Sprintf("Status: %d\n", status))
-
-	for key, values := range headers {
-		for _, value := range values {
-			content.WriteString(fmt.Sprintf("%s: %s\n", key, value))
+	w.responseStatus = status
+	if headers != nil {
+		w.responseHeaders = make(map[string][]string, len(headers))
+		for key, values := range headers {
+			headerValues := make([]string, len(values))
+			copy(headerValues, values)
+			w.responseHeaders[key] = headerValues
 		}
 	}
-	content.WriteString("\n")
+	w.statusWritten = true
+	return nil
+}
 
-	_, err := w.file.WriteString(content.String())
-	if err == nil {
-		w.statusWritten = true
+// WriteAPIRequest buffers the upstream API request details for later writing.
+//
+// Parameters:
+//   - apiRequest: The API request data (typically includes URL, headers, body sent upstream)
+//
+// Returns:
+//   - error: Always returns nil (buffering cannot fail)
+func (w *FileStreamingLogWriter) WriteAPIRequest(apiRequest []byte) error {
+	if len(apiRequest) == 0 {
+		return nil
 	}
-	return err
+	w.apiRequest = bytes.Clone(apiRequest)
+	return nil
+}
+
+// WriteAPIResponse buffers the upstream API response details for later writing.
+//
+// Parameters:
+//   - apiResponse: The API response data
+//
+// Returns:
+//   - error: Always returns nil (buffering cannot fail)
+func (w *FileStreamingLogWriter) WriteAPIResponse(apiResponse []byte) error {
+	if len(apiResponse) == 0 {
+		return nil
+	}
+	w.apiResponse = bytes.Clone(apiResponse)
+	return nil
 }
 
 // Close finalizes the log file and cleans up resources.
+// It writes all buffered data to the file in the correct order:
+// API REQUEST -> API RESPONSE -> RESPONSE (status, headers, body chunks)
 //
 // Returns:
 //   - error: An error if closing fails, nil otherwise
@@ -705,27 +770,84 @@ func (w *FileStreamingLogWriter) Close() error {
 		close(w.chunkChan)
 	}
 
-	// Wait for async writer to finish
+	// Wait for async writer to finish buffering chunks
 	if w.closeChan != nil {
 		<-w.closeChan
 		w.chunkChan = nil
 	}
 
-	if w.file != nil {
-		return w.file.Close()
+	if w.file == nil {
+		return nil
 	}
 
-	return nil
+	// Write all content in the correct order
+	var content strings.Builder
+
+	// 1. Write API REQUEST section
+	if len(w.apiRequest) > 0 {
+		if bytes.HasPrefix(w.apiRequest, []byte("=== API REQUEST")) {
+			content.Write(w.apiRequest)
+			if !bytes.HasSuffix(w.apiRequest, []byte("\n")) {
+				content.WriteString("\n")
+			}
+		} else {
+			content.WriteString("=== API REQUEST ===\n")
+			content.Write(w.apiRequest)
+			content.WriteString("\n")
+		}
+		content.WriteString("\n")
+	}
+
+	// 2. Write API RESPONSE section
+	if len(w.apiResponse) > 0 {
+		if bytes.HasPrefix(w.apiResponse, []byte("=== API RESPONSE")) {
+			content.Write(w.apiResponse)
+			if !bytes.HasSuffix(w.apiResponse, []byte("\n")) {
+				content.WriteString("\n")
+			}
+		} else {
+			content.WriteString("=== API RESPONSE ===\n")
+			content.Write(w.apiResponse)
+			content.WriteString("\n")
+		}
+		content.WriteString("\n")
+	}
+
+	// 3. Write RESPONSE section (status, headers, buffered chunks)
+	content.WriteString("=== RESPONSE ===\n")
+	if w.statusWritten {
+		content.WriteString(fmt.Sprintf("Status: %d\n", w.responseStatus))
+	}
+
+	for key, values := range w.responseHeaders {
+		for _, value := range values {
+			content.WriteString(fmt.Sprintf("%s: %s\n", key, value))
+		}
+	}
+	content.WriteString("\n")
+
+	// Write buffered response body chunks
+	if w.bufferedChunks != nil && w.bufferedChunks.Len() > 0 {
+		content.Write(w.bufferedChunks.Bytes())
+	}
+
+	// Write the complete content to file
+	if _, err := w.file.WriteString(content.String()); err != nil {
+		_ = w.file.Close()
+		return err
+	}
+
+	return w.file.Close()
 }
 
-// asyncWriter runs in a goroutine to handle async chunk writing.
-// It continuously reads chunks from the channel and writes them to the file.
+// asyncWriter runs in a goroutine to buffer chunks from the channel.
+// It continuously reads chunks from the channel and buffers them for later writing.
 func (w *FileStreamingLogWriter) asyncWriter() {
 	defer close(w.closeChan)
 
 	for chunk := range w.chunkChan {
-		if w.file != nil {
-			_, _ = w.file.Write(chunk)
+		if w.bufferedChunks != nil {
+			w.bufferedChunks.Write(chunk)
 		}
 	}
 }
@@ -752,6 +874,28 @@ func (w *NoOpStreamingLogWriter) WriteStatus(_ int, _ map[string][]string) error
 	return nil
 }
 
+// WriteAPIRequest is a no-op implementation that does nothing and always returns nil.
+//
+// Parameters:
+//   - apiRequest: The API request data (ignored)
+//
+// Returns:
+//   - error: Always returns nil
+func (w *NoOpStreamingLogWriter) WriteAPIRequest(_ []byte) error {
+	return nil
+}
+
+// WriteAPIResponse is a no-op implementation that does nothing and always returns nil.
+//
+// Parameters:
+//   - apiResponse: The API response data (ignored)
+//
+// Returns:
+//   - error: Always returns nil
+func (w *NoOpStreamingLogWriter) WriteAPIResponse(_ []byte) error {
+	return nil
+}
+
 // Close is a no-op implementation that does nothing and always returns nil.
 //
 // Returns:

internal/registry/model_definitions.go

@@ -16,6 +16,7 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 4.5 Haiku",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
+			// Thinking: not supported for Haiku models
 		},
 		{
 			ID:                  "claude-sonnet-4-5-20250929",
@@ -26,60 +27,6 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 4.5 Sonnet",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
-		},
-		{
-			ID:                  "claude-sonnet-4-5-thinking",
-			Object:              "model",
-			Created:             1759104000, // 2025-09-29
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Sonnet Thinking",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                  "claude-opus-4-5-thinking",
-			Object:              "model",
-			Created:             1761955200, // 2025-11-01
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Opus Thinking",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                  "claude-opus-4-5-thinking-low",
-			Object:              "model",
-			Created:             1761955200, // 2025-11-01
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Opus Thinking Low",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                  "claude-opus-4-5-thinking-medium",
-			Object:              "model",
-			Created:             1761955200, // 2025-11-01
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Opus Thinking Medium",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                  "claude-opus-4-5-thinking-high",
-			Object:              "model",
-			Created:             1761955200, // 2025-11-01
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Opus Thinking High",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
@@ -92,6 +39,7 @@ func GetClaudeModels() []*ModelInfo {
 			Description:         "Premium model combining maximum intelligence with practical performance",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
 			ID:                  "claude-opus-4-1-20250805",
@@ -102,6 +50,7 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 4.1 Opus",
 			ContextLength:       200000,
 			MaxCompletionTokens: 32000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
 			ID:                  "claude-opus-4-20250514",
@@ -112,6 +61,7 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 4 Opus",
 			ContextLength:       200000,
 			MaxCompletionTokens: 32000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
 			ID:                  "claude-sonnet-4-20250514",
@@ -122,6 +72,7 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 4 Sonnet",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
 			ID:                  "claude-3-7-sonnet-20250219",
@@ -132,6 +83,7 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 3.7 Sonnet",
 			ContextLength:       128000,
 			MaxCompletionTokens: 8192,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
 			ID:                  "claude-3-5-haiku-20241022",
@@ -142,6 +94,7 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 3.5 Haiku",
 			ContextLength:       128000,
 			MaxCompletionTokens: 8192,
+			// Thinking: not supported for Haiku models
 		},
 	}
 }
@@ -529,58 +482,7 @@ func GetOpenAIModels() []*ModelInfo {
 			ContextLength:       400000,
 			MaxCompletionTokens: 128000,
 			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5-minimal",
-			Object:              "model",
-			Created:             1754524800,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-08-07",
-			DisplayName:         "GPT 5 Minimal",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5-low",
-			Object:              "model",
-			Created:             1754524800,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-08-07",
-			DisplayName:         "GPT 5 Low",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5-medium",
-			Object:              "model",
-			Created:             1754524800,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-08-07",
-			DisplayName:         "GPT 5 Medium",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5-high",
-			Object:              "model",
-			Created:             1754524800,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-08-07",
-			DisplayName:         "GPT 5 High",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"minimal", "low", "medium", "high"}},
 		},
 		{
 			ID:                  "gpt-5-codex",
@@ -594,45 +496,7 @@ func GetOpenAIModels() []*ModelInfo {
 			ContextLength:       400000,
 			MaxCompletionTokens: 128000,
 			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5-codex-low",
-			Object:              "model",
-			Created:             1757894400,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-09-15",
-			DisplayName:         "GPT 5 Codex Low",
-			Description:         "Stable version of GPT 5 Codex, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5-codex-medium",
-			Object:              "model",
-			Created:             1757894400,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-09-15",
-			DisplayName:         "GPT 5 Codex Medium",
-			Description:         "Stable version of GPT 5 Codex, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5-codex-high",
-			Object:              "model",
-			Created:             1757894400,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-09-15",
-			DisplayName:         "GPT 5 Codex High",
-			Description:         "Stable version of GPT 5 Codex, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
 		},
 		{
 			ID:                  "gpt-5-codex-mini",
@@ -646,32 +510,7 @@ func GetOpenAIModels() []*ModelInfo {
 			ContextLength:       400000,
 			MaxCompletionTokens: 128000,
 			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5-codex-mini-medium",
-			Object:              "model",
-			Created:             1762473600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-11-07",
-			DisplayName:         "GPT 5 Codex Mini Medium",
-			Description:         "Stable version of GPT 5 Codex Mini: cheaper, faster, but less capable version of GPT 5 Codex.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5-codex-mini-high",
-			Object:              "model",
-			Created:             1762473600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-11-07",
-			DisplayName:         "GPT 5 Codex Mini High",
-			Description:         "Stable version of GPT 5 Codex Mini: cheaper, faster, but less capable version of GPT 5 Codex.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
 		},
 		{
 			ID:                  "gpt-5.1",
@@ -685,58 +524,7 @@ func GetOpenAIModels() []*ModelInfo {
 			ContextLength:       400000,
 			MaxCompletionTokens: 128000,
 			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-none",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Low",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-low",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Low",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-medium",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Medium",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-high",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 High",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"none", "low", "medium", "high"}},
 		},
 		{
 			ID:                  "gpt-5.1-codex",
@@ -745,50 +533,12 @@ func GetOpenAIModels() []*ModelInfo {
 			OwnedBy:             "openai",
 			Type:                "openai",
 			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Codex",
-			Description:         "Stable version of GPT 5 Codex, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-codex-low",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Codex Low",
-			Description:         "Stable version of GPT 5 Codex, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-codex-medium",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Codex Medium",
-			Description:         "Stable version of GPT 5 Codex, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-codex-high",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Codex High",
-			Description:         "Stable version of GPT 5 Codex, The best model for coding and agentic tasks across domains.",
+			DisplayName:         "GPT 5.1 Codex",
+			Description:         "Stable version of GPT 5.1 Codex, The best model for coding and agentic tasks across domains.",
 			ContextLength:       400000,
 			MaxCompletionTokens: 128000,
 			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
 		},
 		{
 			ID:                  "gpt-5.1-codex-mini",
@@ -797,39 +547,13 @@ func GetOpenAIModels() []*ModelInfo {
 			OwnedBy:             "openai",
 			Type:                "openai",
 			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Codex Mini",
-			Description:         "Stable version of GPT 5 Codex Mini: cheaper, faster, but less capable version of GPT 5 Codex.",
+			DisplayName:         "GPT 5.1 Codex Mini",
+			Description:         "Stable version of GPT 5.1 Codex Mini: cheaper, faster, but less capable version of GPT 5.1 Codex.",
 			ContextLength:       400000,
 			MaxCompletionTokens: 128000,
 			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
 		},
-		{
-			ID:                  "gpt-5.1-codex-mini-medium",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Codex Mini Medium",
-			Description:         "Stable version of GPT 5 Codex Mini: cheaper, faster, but less capable version of GPT 5 Codex.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-codex-mini-high",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5 Codex Mini High",
-			Description:         "Stable version of GPT 5 Codex Mini: cheaper, faster, but less capable version of GPT 5 Codex.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-
 		{
 			ID:                  "gpt-5.1-codex-max",
 			Object:              "model",
@@ -837,63 +561,12 @@ func GetOpenAIModels() []*ModelInfo {
 			OwnedBy:             "openai",
 			Type:                "openai",
 			Version:             "gpt-5.1-max",
-			DisplayName:         "GPT 5 Codex Max",
-			Description:         "Stable version of GPT 5 Codex Max",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-codex-max-low",
-			Object:              "model",
-			Created:             1763424000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-max",
-			DisplayName:         "GPT 5 Codex Max Low",
-			Description:         "Stable version of GPT 5 Codex Max Low",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-codex-max-medium",
-			Object:              "model",
-			Created:             1763424000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-max",
-			DisplayName:         "GPT 5 Codex Max Medium",
-			Description:         "Stable version of GPT 5 Codex Max Medium",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-codex-max-high",
-			Object:              "model",
-			Created:             1763424000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-max",
-			DisplayName:         "GPT 5 Codex Max High",
-			Description:         "Stable version of GPT 5 Codex Max High",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-		},
-		{
-			ID:                  "gpt-5.1-codex-max-xhigh",
-			Object:              "model",
-			Created:             1763424000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-max",
-			DisplayName:         "GPT 5 Codex Max XHigh",
-			Description:         "Stable version of GPT 5 Codex Max XHigh",
+			DisplayName:         "GPT 5.1 Codex Max",
+			Description:         "Stable version of GPT 5.1 Codex Max",
 			ContextLength:       400000,
 			MaxCompletionTokens: 128000,
 			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high", "xhigh"}},
 		},
 	}
 }
@@ -944,13 +617,13 @@ func GetQwenModels() []*ModelInfo {
 }
 
 // GetIFlowModels returns supported models for iFlow OAuth accounts.
-
 func GetIFlowModels() []*ModelInfo {
 	entries := []struct {
 		ID          string
 		DisplayName string
 		Description string
 		Created     int64
+		Thinking    *ThinkingSupport
 	}{
 		{ID: "tstars2.0", DisplayName: "TStars-2.0", Description: "iFlow TStars-2.0 multimodal assistant", Created: 1746489600},
 		{ID: "qwen3-coder-plus", DisplayName: "Qwen3-Coder-Plus", Description: "Qwen3 Coder Plus code generation", Created: 1753228800},
@@ -960,17 +633,17 @@ func GetIFlowModels() []*ModelInfo {
 		{ID: "kimi-k2-0905", DisplayName: "Kimi-K2-Instruct-0905", Description: "Moonshot Kimi K2 instruct 0905", Created: 1757030400},
 		{ID: "glm-4.6", DisplayName: "GLM-4.6", Description: "Zhipu GLM 4.6 general model", Created: 1759190400},
 		{ID: "kimi-k2", DisplayName: "Kimi-K2", Description: "Moonshot Kimi K2 general model", Created: 1752192000},
-		{ID: "kimi-k2-thinking", DisplayName: "Kimi-K2-Thinking", Description: "Moonshot Kimi K2 general model", Created: 1762387200},
+		{ID: "kimi-k2-thinking", DisplayName: "Kimi-K2-Thinking", Description: "Moonshot Kimi K2 thinking model", Created: 1762387200, Thinking: &ThinkingSupport{Levels: []string{"low", "medium", "high"}}},
 		{ID: "deepseek-v3.2-chat", DisplayName: "DeepSeek-V3.2", Description: "DeepSeek V3.2", Created: 1764576000},
 		{ID: "deepseek-v3.2", DisplayName: "DeepSeek-V3.2-Exp", Description: "DeepSeek V3.2 experimental", Created: 1759104000},
 		{ID: "deepseek-v3.1", DisplayName: "DeepSeek-V3.1-Terminus", Description: "DeepSeek V3.1 Terminus", Created: 1756339200},
-		{ID: "deepseek-r1", DisplayName: "DeepSeek-R1", Description: "DeepSeek reasoning model R1", Created: 1737331200},
+		{ID: "deepseek-r1", DisplayName: "DeepSeek-R1", Description: "DeepSeek reasoning model R1", Created: 1737331200, Thinking: &ThinkingSupport{Levels: []string{"low", "medium", "high"}}},
 		{ID: "deepseek-v3", DisplayName: "DeepSeek-V3-671B", Description: "DeepSeek V3 671B", Created: 1734307200},
 		{ID: "qwen3-32b", DisplayName: "Qwen3-32B", Description: "Qwen3 32B", Created: 1747094400},
-		{ID: "qwen3-235b-a22b-thinking-2507", DisplayName: "Qwen3-235B-A22B-Thinking", Description: "Qwen3 235B A22B Thinking (2507)", Created: 1753401600},
+		{ID: "qwen3-235b-a22b-thinking-2507", DisplayName: "Qwen3-235B-A22B-Thinking", Description: "Qwen3 235B A22B Thinking (2507)", Created: 1753401600, Thinking: &ThinkingSupport{Levels: []string{"low", "medium", "high"}}},
 		{ID: "qwen3-235b-a22b-instruct", DisplayName: "Qwen3-235B-A22B-Instruct", Description: "Qwen3 235B A22B Instruct", Created: 1753401600},
 		{ID: "qwen3-235b", DisplayName: "Qwen3-235B-A22B", Description: "Qwen3 235B A22B", Created: 1753401600},
-		{ID: "minimax-m2", DisplayName: "MiniMax-M2", Description: "MiniMax M2", Created: 1758672000},
+		{ID: "minimax-m2", DisplayName: "MiniMax-M2", Description: "MiniMax M2", Created: 1758672000, Thinking: &ThinkingSupport{Levels: []string{"low", "medium", "high"}}},
 	}
 	models := make([]*ModelInfo, 0, len(entries))
 	for _, entry := range entries {
@@ -982,7 +655,30 @@ func GetIFlowModels() []*ModelInfo {
 			Type:        "iflow",
 			DisplayName: entry.DisplayName,
 			Description: entry.Description,
+			Thinking:    entry.Thinking,
 		})
 	}
 	return models
 }
+
+// AntigravityModelConfig captures static antigravity model overrides, including
+// Thinking budget limits and provider max completion tokens.
+type AntigravityModelConfig struct {
+	Thinking            *ThinkingSupport
+	MaxCompletionTokens int
+	Name                string
+}
+
+// GetAntigravityModelConfig returns static configuration for antigravity models.
+// Keys use the ALIASED model names (after modelName2Alias conversion) for direct lookup.
+func GetAntigravityModelConfig() map[string]*AntigravityModelConfig {
+	return map[string]*AntigravityModelConfig{
+		"gemini-2.5-flash":                        {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}, Name: "models/gemini-2.5-flash"},
+		"gemini-2.5-flash-lite":                   {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}, Name: "models/gemini-2.5-flash-lite"},
+		"gemini-2.5-computer-use-preview-10-2025": {Name: "models/gemini-2.5-computer-use-preview-10-2025"},
+		"gemini-3-pro-preview":                    {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true}, Name: "models/gemini-3-pro-preview"},
+		"gemini-3-pro-image-preview":              {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true}, Name: "models/gemini-3-pro-image-preview"},
+		"gemini-claude-sonnet-4-5-thinking":       {Thinking: &ThinkingSupport{Min: 1024, Max: 200000, ZeroAllowed: false, DynamicAllowed: true}, MaxCompletionTokens: 64000},
+		"gemini-claude-opus-4-5-thinking":         {Thinking: &ThinkingSupport{Min: 1024, Max: 200000, ZeroAllowed: false, DynamicAllowed: true}, MaxCompletionTokens: 64000},
+	}
+}

internal/registry/model_registry.go

@@ -63,6 +63,9 @@ type ThinkingSupport struct {
 	ZeroAllowed bool `json:"zero_allowed,omitempty"`
 	// DynamicAllowed indicates whether -1 is a valid value (dynamic thinking budget).
 	DynamicAllowed bool `json:"dynamic_allowed,omitempty"`
+	// Levels defines discrete reasoning effort levels (e.g., "low", "medium", "high").
+	// When set, the model uses level-based reasoning instead of token budgets.
+	Levels []string `json:"levels,omitempty"`
 }
 
 // ModelRegistration tracks a model's availability

internal/runtime/executor/aistudio_executor.go

@@ -1,3 +1,6 @@
+// Package executor provides runtime execution capabilities for various AI service providers.
+// This file implements the AI Studio executor that routes requests through a websocket-backed
+// transport for the AI Studio provider.
 package executor
 
 import (
@@ -26,19 +29,28 @@ type AIStudioExecutor struct {
 	cfg      *config.Config
 }
 
-// NewAIStudioExecutor constructs a websocket executor for the provider name.
+// NewAIStudioExecutor creates a new AI Studio executor instance.
+//
+// Parameters:
+//   - cfg: The application configuration
+//   - provider: The provider name
+//   - relay: The websocket relay manager
+//
+// Returns:
+//   - *AIStudioExecutor: A new AI Studio executor instance
 func NewAIStudioExecutor(cfg *config.Config, provider string, relay *wsrelay.Manager) *AIStudioExecutor {
 	return &AIStudioExecutor{provider: strings.ToLower(provider), relay: relay, cfg: cfg}
 }
 
-// Identifier returns the logical provider key for routing.
+// Identifier returns the executor identifier.
 func (e *AIStudioExecutor) Identifier() string { return "aistudio" }
 
-// PrepareRequest is a no-op because websocket transport already injects headers.
+// PrepareRequest prepares the HTTP request for execution (no-op for AI Studio).
 func (e *AIStudioExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error {
 	return nil
 }
 
+// Execute performs a non-streaming request to the AI Studio API.
 func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
@@ -92,6 +104,7 @@ func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth,
 	return resp, nil
 }
 
+// ExecuteStream performs a streaming request to the AI Studio API.
 func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
@@ -239,6 +252,7 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 	return stream, nil
 }
 
+// CountTokens counts tokens for the given request using the AI Studio API.
 func (e *AIStudioExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
 	_, body, err := e.translateRequest(req, opts, false)
 	if err != nil {
@@ -293,8 +307,8 @@ func (e *AIStudioExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.A
 	return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
 }
 
-func (e *AIStudioExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
-	_ = ctx
+// Refresh refreshes the authentication credentials (no-op for AI Studio).
+func (e *AIStudioExecutor) Refresh(_ context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	return auth, nil
 }
 
@@ -309,7 +323,9 @@ func (e *AIStudioExecutor) translateRequest(req cliproxyexecutor.Request, opts c
 	to := sdktranslator.FromString("gemini")
 	payload := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), stream)
 	payload = applyThinkingMetadata(payload, req.Metadata, req.Model)
+	payload = util.ApplyDefaultThinkingIfNeeded(req.Model, payload)
 	payload = util.ConvertThinkingLevelToBudget(payload)
+	payload = util.NormalizeGeminiThinkingBudget(req.Model, payload)
 	payload = util.StripThinkingConfigIfUnsupported(req.Model, payload)
 	payload = fixGeminiImageAspectRatio(req.Model, payload)
 	payload = applyPayloadConfig(e.cfg, req.Model, payload)

internal/runtime/executor/antigravity_executor.go

@@ -1,3 +1,6 @@
+// Package executor provides runtime execution capabilities for various AI service providers.
+// This file implements the Antigravity executor that proxies requests to the antigravity
+// upstream using OAuth credentials.
 package executor
 
 import (
@@ -27,18 +30,17 @@ import (
 )
 
 const (
-	antigravityBaseURLDaily        = "https://daily-cloudcode-pa.sandbox.googleapis.com"
-	antigravityBaseURLAutopush     = "https://autopush-cloudcode-pa.sandbox.googleapis.com"
-	antigravityBaseURLProd         = "https://cloudcode-pa.googleapis.com"
-	antigravityStreamPath          = "/v1internal:streamGenerateContent"
-	antigravityGeneratePath        = "/v1internal:generateContent"
-	antigravityModelsPath          = "/v1internal:fetchAvailableModels"
-	antigravityClientID            = "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
-	antigravityClientSecret        = "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"
-	defaultAntigravityAgent        = "antigravity/1.11.5 windows/amd64"
-	antigravityAuthType            = "antigravity"
-	refreshSkew                    = 3000 * time.Second
-	streamScannerBuffer        int = 20_971_520
+	antigravityBaseURLDaily = "https://daily-cloudcode-pa.sandbox.googleapis.com"
+	// antigravityBaseURLAutopush     = "https://autopush-cloudcode-pa.sandbox.googleapis.com"
+	antigravityBaseURLProd  = "https://cloudcode-pa.googleapis.com"
+	antigravityStreamPath   = "/v1internal:streamGenerateContent"
+	antigravityGeneratePath = "/v1internal:generateContent"
+	antigravityModelsPath   = "/v1internal:fetchAvailableModels"
+	antigravityClientID     = "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
+	antigravityClientSecret = "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"
+	defaultAntigravityAgent = "antigravity/1.11.5 windows/amd64"
+	antigravityAuthType     = "antigravity"
+	refreshSkew             = 3000 * time.Second
 )
 
 var randSource = rand.New(rand.NewSource(time.Now().UnixNano()))
@@ -48,18 +50,24 @@ type AntigravityExecutor struct {
 	cfg *config.Config
 }
 
-// NewAntigravityExecutor constructs a new executor instance.
+// NewAntigravityExecutor creates a new Antigravity executor instance.
+//
+// Parameters:
+//   - cfg: The application configuration
+//
+// Returns:
+//   - *AntigravityExecutor: A new Antigravity executor instance
 func NewAntigravityExecutor(cfg *config.Config) *AntigravityExecutor {
 	return &AntigravityExecutor{cfg: cfg}
 }
 
-// Identifier implements ProviderExecutor.
+// Identifier returns the executor identifier.
 func (e *AntigravityExecutor) Identifier() string { return antigravityAuthType }
 
-// PrepareRequest implements ProviderExecutor.
+// PrepareRequest prepares the HTTP request for execution (no-op for Antigravity).
 func (e *AntigravityExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
 
-// Execute handles non-streaming requests via the antigravity generate endpoint.
+// Execute performs a non-streaming request to the Antigravity API.
 func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	token, updatedAuth, errToken := e.ensureAccessToken(ctx, auth)
 	if errToken != nil {
@@ -77,6 +85,8 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 	translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 
 	translated = applyThinkingMetadataCLI(translated, req.Metadata, req.Model)
+	translated = util.ApplyDefaultThinkingIfNeededCLI(req.Model, translated)
+	translated = normalizeAntigravityThinking(req.Model, translated)
 
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
@@ -150,7 +160,7 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 	return resp, err
 }
 
-// ExecuteStream handles streaming requests via the antigravity upstream.
+// ExecuteStream performs a streaming request to the Antigravity API.
 func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
 	ctx = context.WithValue(ctx, "alt", "")
 
@@ -170,6 +180,8 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 	translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
 	translated = applyThinkingMetadataCLI(translated, req.Metadata, req.Model)
+	translated = util.ApplyDefaultThinkingIfNeededCLI(req.Model, translated)
+	translated = normalizeAntigravityThinking(req.Model, translated)
 
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
@@ -288,7 +300,7 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 	return nil, err
 }
 
-// Refresh refreshes the OAuth token using the refresh token.
+// Refresh refreshes the authentication credentials using the refresh token.
 func (e *AntigravityExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	if auth == nil {
 		return auth, nil
@@ -300,7 +312,7 @@ func (e *AntigravityExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Au
 	return updated, nil
 }
 
-// CountTokens is not supported for the antigravity provider.
+// CountTokens counts tokens for the given request (not supported for Antigravity).
 func (e *AntigravityExecutor) CountTokens(context.Context, *cliproxyauth.Auth, cliproxyexecutor.Request, cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
 	return cliproxyexecutor.Response{}, statusErr{code: http.StatusNotImplemented, msg: "count tokens not supported"}
 }
@@ -366,28 +378,34 @@ func FetchAntigravityModels(ctx context.Context, auth *cliproxyauth.Auth, cfg *c
 		}
 
 		now := time.Now().Unix()
+		modelConfig := registry.GetAntigravityModelConfig()
 		models := make([]*registry.ModelInfo, 0, len(result.Map()))
-		for id := range result.Map() {
-			id = modelName2Alias(id)
-			if id != "" {
+		for originalName := range result.Map() {
+			aliasName := modelName2Alias(originalName)
+			if aliasName != "" {
+				cfg := modelConfig[aliasName]
+				modelName := aliasName
+				if cfg != nil && cfg.Name != "" {
+					modelName = cfg.Name
+				}
 				modelInfo := &registry.ModelInfo{
-					ID:          id,
-					Name:        id,
-					Description: id,
-					DisplayName: id,
-					Version:     id,
+					ID:          aliasName,
+					Name:        modelName,
+					Description: aliasName,
+					DisplayName: aliasName,
+					Version:     aliasName,
 					Object:      "model",
 					Created:     now,
 					OwnedBy:     antigravityAuthType,
 					Type:        antigravityAuthType,
 				}
-				// Add Thinking support for thinking models
-				if strings.HasSuffix(id, "-thinking") || strings.Contains(id, "-thinking-") {
-					modelInfo.Thinking = &registry.ThinkingSupport{
-						Min:            1024,
-						Max:            100000,
-						ZeroAllowed:    false,
-						DynamicAllowed: true,
+				// Look up Thinking support from static config using alias name
+				if cfg != nil {
+					if cfg.Thinking != nil {
+						modelInfo.Thinking = cfg.Thinking
+					}
+					if cfg.MaxCompletionTokens > 0 {
+						modelInfo.MaxCompletionTokens = cfg.MaxCompletionTokens
 					}
 				}
 				models = append(models, modelInfo)
@@ -533,6 +551,9 @@ func (e *AntigravityExecutor) buildRequest(ctx context.Context, auth *cliproxyau
 		strJSON = util.DeleteKey(strJSON, "minLength")
 		strJSON = util.DeleteKey(strJSON, "maxLength")
 		strJSON = util.DeleteKey(strJSON, "exclusiveMinimum")
+		strJSON = util.DeleteKey(strJSON, "exclusiveMaximum")
+		strJSON = util.DeleteKey(strJSON, "$ref")
+		strJSON = util.DeleteKey(strJSON, "$defs")
 
 		paths = make([]string, 0)
 		util.Walk(gjson.Parse(strJSON), "", "anyOf", &paths)
@@ -648,7 +669,7 @@ func buildBaseURL(auth *cliproxyauth.Auth) string {
 	if baseURLs := antigravityBaseURLFallbackOrder(auth); len(baseURLs) > 0 {
 		return baseURLs[0]
 	}
-	return antigravityBaseURLAutopush
+	return antigravityBaseURLDaily
 }
 
 func resolveHost(base string) string {
@@ -684,7 +705,7 @@ func antigravityBaseURLFallbackOrder(auth *cliproxyauth.Auth) []string {
 	}
 	return []string{
 		antigravityBaseURLDaily,
-		antigravityBaseURLAutopush,
+		// antigravityBaseURLAutopush,
 		antigravityBaseURLProd,
 	}
 }
@@ -808,3 +829,65 @@ func alias2ModelName(modelName string) string {
 		return modelName
 	}
 }
+
+// normalizeAntigravityThinking clamps or removes thinking config based on model support.
+// For Claude models, it additionally ensures thinking budget < max_tokens.
+func normalizeAntigravityThinking(model string, payload []byte) []byte {
+	payload = util.StripThinkingConfigIfUnsupported(model, payload)
+	if !util.ModelSupportsThinking(model) {
+		return payload
+	}
+	budget := gjson.GetBytes(payload, "request.generationConfig.thinkingConfig.thinkingBudget")
+	if !budget.Exists() {
+		return payload
+	}
+	raw := int(budget.Int())
+	normalized := util.NormalizeThinkingBudget(model, raw)
+
+	isClaude := strings.Contains(strings.ToLower(model), "claude")
+	if isClaude {
+		effectiveMax, setDefaultMax := antigravityEffectiveMaxTokens(model, payload)
+		if effectiveMax > 0 && normalized >= effectiveMax {
+			normalized = effectiveMax - 1
+		}
+		minBudget := antigravityMinThinkingBudget(model)
+		if minBudget > 0 && normalized >= 0 && normalized < minBudget {
+			// Budget is below minimum, remove thinking config entirely
+			payload, _ = sjson.DeleteBytes(payload, "request.generationConfig.thinkingConfig")
+			return payload
+		}
+		if setDefaultMax {
+			if res, errSet := sjson.SetBytes(payload, "request.generationConfig.maxOutputTokens", effectiveMax); errSet == nil {
+				payload = res
+			}
+		}
+	}
+
+	updated, err := sjson.SetBytes(payload, "request.generationConfig.thinkingConfig.thinkingBudget", normalized)
+	if err != nil {
+		return payload
+	}
+	return updated
+}
+
+// antigravityEffectiveMaxTokens returns the max tokens to cap thinking:
+// prefer request-provided maxOutputTokens; otherwise fall back to model default.
+// The boolean indicates whether the value came from the model default (and thus should be written back).
+func antigravityEffectiveMaxTokens(model string, payload []byte) (max int, fromModel bool) {
+	if maxTok := gjson.GetBytes(payload, "request.generationConfig.maxOutputTokens"); maxTok.Exists() && maxTok.Int() > 0 {
+		return int(maxTok.Int()), false
+	}
+	if modelInfo := registry.GetGlobalRegistry().GetModelInfo(model); modelInfo != nil && modelInfo.MaxCompletionTokens > 0 {
+		return modelInfo.MaxCompletionTokens, true
+	}
+	return 0, false
+}
+
+// antigravityMinThinkingBudget returns the minimum thinking budget for a model.
+// Falls back to -1 if no model info is found.
+func antigravityMinThinkingBudget(model string) int {
+	if modelInfo := registry.GetGlobalRegistry().GetModelInfo(model); modelInfo != nil && modelInfo.Thinking != nil {
+		return modelInfo.Thinking.Min
+	}
+	return -1
+}

internal/runtime/executor/claude_executor.go

@@ -54,15 +54,22 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	// Use streaming translation to preserve function calling, except for claude.
 	stream := from != to
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), stream)
-	modelForUpstream := req.Model
-	if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
-		body, _ = sjson.SetBytes(body, "model", modelOverride)
-		modelForUpstream = modelOverride
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel == "" {
+		upstreamModel = req.Model
 	}
-	// Inject thinking config based on model suffix for thinking variants
-	body = e.injectThinkingConfig(req.Model, body)
+	if modelOverride := e.resolveUpstreamModel(upstreamModel, auth); modelOverride != "" {
+		upstreamModel = modelOverride
+	} else if !strings.EqualFold(upstreamModel, req.Model) {
+		if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
+			upstreamModel = modelOverride
+		}
+	}
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	// Inject thinking config based on model metadata for thinking variants
+	body = e.injectThinkingConfig(req.Model, req.Metadata, body)
 
-	if !strings.HasPrefix(modelForUpstream, "claude-3-5-haiku") {
+	if !strings.HasPrefix(upstreamModel, "claude-3-5-haiku") {
 		body = checkSystemInstructions(body)
 	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
@@ -161,11 +168,20 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("claude")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
-	if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
-		body, _ = sjson.SetBytes(body, "model", modelOverride)
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel == "" {
+		upstreamModel = req.Model
 	}
-	// Inject thinking config based on model suffix for thinking variants
-	body = e.injectThinkingConfig(req.Model, body)
+	if modelOverride := e.resolveUpstreamModel(upstreamModel, auth); modelOverride != "" {
+		upstreamModel = modelOverride
+	} else if !strings.EqualFold(upstreamModel, req.Model) {
+		if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
+			upstreamModel = modelOverride
+		}
+	}
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	// Inject thinking config based on model metadata for thinking variants
+	body = e.injectThinkingConfig(req.Model, req.Metadata, body)
 	body = checkSystemInstructions(body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 
@@ -238,7 +254,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 		// If from == to (Claude → Claude), directly forward the SSE stream without translation
 		if from == to {
 			scanner := bufio.NewScanner(decodedBody)
-			scanner.Buffer(nil, 20_971_520)
+			scanner.Buffer(nil, 52_428_800) // 50MB
 			for scanner.Scan() {
 				line := scanner.Bytes()
 				appendAPIResponseChunk(ctx, e.cfg, line)
@@ -261,7 +277,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 
 		// For other formats, use translation
 		scanner := bufio.NewScanner(decodedBody)
-		scanner.Buffer(nil, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -295,13 +311,20 @@ func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	// Use streaming translation to preserve function calling, except for claude.
 	stream := from != to
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), stream)
-	modelForUpstream := req.Model
-	if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
-		body, _ = sjson.SetBytes(body, "model", modelOverride)
-		modelForUpstream = modelOverride
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel == "" {
+		upstreamModel = req.Model
 	}
+	if modelOverride := e.resolveUpstreamModel(upstreamModel, auth); modelOverride != "" {
+		upstreamModel = modelOverride
+	} else if !strings.EqualFold(upstreamModel, req.Model) {
+		if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
+			upstreamModel = modelOverride
+		}
+	}
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
-	if !strings.HasPrefix(modelForUpstream, "claude-3-5-haiku") {
+	if !strings.HasPrefix(upstreamModel, "claude-3-5-haiku") {
 		body = checkSystemInstructions(body)
 	}
 
@@ -427,31 +450,15 @@ func extractAndRemoveBetas(body []byte) ([]string, []byte) {
 	return betas, body
 }
 
-// injectThinkingConfig adds thinking configuration based on model name suffix
-func (e *ClaudeExecutor) injectThinkingConfig(modelName string, body []byte) []byte {
-	// Only inject if thinking config is not already present
-	if gjson.GetBytes(body, "thinking").Exists() {
+// injectThinkingConfig adds thinking configuration based on metadata using the unified flow.
+// It uses util.ResolveClaudeThinkingConfig which internally calls ResolveThinkingConfigFromMetadata
+// and NormalizeThinkingBudget, ensuring consistency with other executors like Gemini.
+func (e *ClaudeExecutor) injectThinkingConfig(modelName string, metadata map[string]any, body []byte) []byte {
+	budget, ok := util.ResolveClaudeThinkingConfig(modelName, metadata)
+	if !ok {
 		return body
 	}
-
-	var budgetTokens int
-	switch {
-	case strings.HasSuffix(modelName, "-thinking-low"):
-		budgetTokens = 1024
-	case strings.HasSuffix(modelName, "-thinking-medium"):
-		budgetTokens = 8192
-	case strings.HasSuffix(modelName, "-thinking-high"):
-		budgetTokens = 24576
-	case strings.HasSuffix(modelName, "-thinking"):
-		// Default thinking without suffix uses medium budget
-		budgetTokens = 8192
-	default:
-		return body
-	}
-
-	body, _ = sjson.SetBytes(body, "thinking.type", "enabled")
-	body, _ = sjson.SetBytes(body, "thinking.budget_tokens", budgetTokens)
-	return body
+	return util.ApplyClaudeThinkingConfig(body, budget)
 }
 
 // ensureMaxTokensForThinking ensures max_tokens > thinking.budget_tokens when thinking is enabled.
@@ -491,35 +498,45 @@ func ensureMaxTokensForThinking(modelName string, body []byte) []byte {
 }
 
 func (e *ClaudeExecutor) resolveUpstreamModel(alias string, auth *cliproxyauth.Auth) string {
-	if alias == "" {
+	trimmed := strings.TrimSpace(alias)
+	if trimmed == "" {
 		return ""
 	}
-	// Hardcoded mappings for thinking models to actual Claude model names
-	switch alias {
-	case "claude-opus-4-5-thinking", "claude-opus-4-5-thinking-low", "claude-opus-4-5-thinking-medium", "claude-opus-4-5-thinking-high":
-		return "claude-opus-4-5-20251101"
-	case "claude-sonnet-4-5-thinking":
-		return "claude-sonnet-4-5-20250929"
-	}
+
 	entry := e.resolveClaudeConfig(auth)
 	if entry == nil {
 		return ""
 	}
+
+	normalizedModel, metadata := util.NormalizeThinkingModel(trimmed)
+
+	// Candidate names to match against configured aliases/names.
+	candidates := []string{strings.TrimSpace(normalizedModel)}
+	if !strings.EqualFold(normalizedModel, trimmed) {
+		candidates = append(candidates, trimmed)
+	}
+	if original := util.ResolveOriginalModel(normalizedModel, metadata); original != "" && !strings.EqualFold(original, normalizedModel) {
+		candidates = append(candidates, original)
+	}
+
 	for i := range entry.Models {
 		model := entry.Models[i]
 		name := strings.TrimSpace(model.Name)
 		modelAlias := strings.TrimSpace(model.Alias)
-		if modelAlias != "" {
-			if strings.EqualFold(modelAlias, alias) {
+
+		for _, candidate := range candidates {
+			if candidate == "" {
+				continue
+			}
+			if modelAlias != "" && strings.EqualFold(modelAlias, candidate) {
 				if name != "" {
 					return name
 				}
-				return alias
+				return candidate
+			}
+			if name != "" && strings.EqualFold(name, candidate) {
+				return name
 			}
-			continue
-		}
-		if name != "" && strings.EqualFold(name, alias) {
-			return name
 		}
 	}
 	return ""

internal/runtime/executor/codex_executor.go

@@ -49,14 +49,18 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-
-	body = e.setReasoningEffortByAlias(req.Model, body)
-
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning.effort")
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return resp, errValidate
+	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
-
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 	body, _ = sjson.SetBytes(body, "stream", true)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
 
@@ -142,13 +146,20 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
-	body = e.setReasoningEffortByAlias(req.Model, body)
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning.effort")
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return nil, errValidate
+	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	url := strings.TrimSuffix(baseURL, "/") + "/responses"
 	httpReq, err := e.cacheHelper(ctx, from, url, req, body)
@@ -205,7 +216,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		scanner.Buffer(nil, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -235,14 +246,16 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 }
 
 func (e *CodexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 
 	modelForCounting := req.Model
 
-	body = e.setReasoningEffortByAlias(req.Model, body)
-
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning.effort")
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
 	body, _ = sjson.SetBytes(body, "stream", false)
 
@@ -261,83 +274,6 @@ func (e *CodexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth
 	return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
 }
 
-func (e *CodexExecutor) setReasoningEffortByAlias(modelName string, payload []byte) []byte {
-	if util.InArray([]string{"gpt-5", "gpt-5-minimal", "gpt-5-low", "gpt-5-medium", "gpt-5-high"}, modelName) {
-		payload, _ = sjson.SetBytes(payload, "model", "gpt-5")
-		switch modelName {
-		case "gpt-5-minimal":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "minimal")
-		case "gpt-5-low":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "low")
-		case "gpt-5-medium":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "medium")
-		case "gpt-5-high":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5-codex", "gpt-5-codex-low", "gpt-5-codex-medium", "gpt-5-codex-high"}, modelName) {
-		payload, _ = sjson.SetBytes(payload, "model", "gpt-5-codex")
-		switch modelName {
-		case "gpt-5-codex-low":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "low")
-		case "gpt-5-codex-medium":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "medium")
-		case "gpt-5-codex-high":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5-codex-mini", "gpt-5-codex-mini-medium", "gpt-5-codex-mini-high"}, modelName) {
-		payload, _ = sjson.SetBytes(payload, "model", "gpt-5-codex-mini")
-		switch modelName {
-		case "gpt-5-codex-mini-medium":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "medium")
-		case "gpt-5-codex-mini-high":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5.1", "gpt-5.1-none", "gpt-5.1-low", "gpt-5.1-medium", "gpt-5.1-high"}, modelName) {
-		payload, _ = sjson.SetBytes(payload, "model", "gpt-5.1")
-		switch modelName {
-		case "gpt-5.1-none":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "none")
-		case "gpt-5.1-low":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "low")
-		case "gpt-5.1-medium":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "medium")
-		case "gpt-5.1-high":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5.1-codex", "gpt-5.1-codex-low", "gpt-5.1-codex-medium", "gpt-5.1-codex-high"}, modelName) {
-		payload, _ = sjson.SetBytes(payload, "model", "gpt-5.1-codex")
-		switch modelName {
-		case "gpt-5.1-codex-low":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "low")
-		case "gpt-5.1-codex-medium":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "medium")
-		case "gpt-5.1-codex-high":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5.1-codex-mini", "gpt-5.1-codex-mini-medium", "gpt-5.1-codex-mini-high"}, modelName) {
-		payload, _ = sjson.SetBytes(payload, "model", "gpt-5.1-codex-mini")
-		switch modelName {
-		case "gpt-5.1-codex-mini-medium":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "medium")
-		case "gpt-5.1-codex-mini-high":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5.1-codex-max", "gpt-5.1-codex-max-low", "gpt-5.1-codex-max-medium", "gpt-5.1-codex-max-high", "gpt-5.1-codex-max-xhigh"}, modelName) {
-		payload, _ = sjson.SetBytes(payload, "model", "gpt-5.1-codex-max")
-		switch modelName {
-		case "gpt-5.1-codex-max-low":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "low")
-		case "gpt-5.1-codex-max-medium":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "medium")
-		case "gpt-5.1-codex-max-high":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "high")
-		case "gpt-5.1-codex-max-xhigh":
-			payload, _ = sjson.SetBytes(payload, "reasoning.effort", "xhigh")
-		}
-	}
-	return payload
-}
-
 func tokenizerForCodexModel(model string) (tokenizer.Codec, error) {
 	sanitized := strings.ToLower(strings.TrimSpace(model))
 	switch {

internal/runtime/executor/gemini_cli_executor.go

@@ -1,3 +1,6 @@
+// Package executor provides runtime execution capabilities for various AI service providers.
+// This file implements the Gemini CLI executor that talks to Cloud Code Assist endpoints
+// using OAuth credentials from auth metadata.
 package executor
 
 import (
@@ -29,11 +32,11 @@ import (
 const (
 	codeAssistEndpoint      = "https://cloudcode-pa.googleapis.com"
 	codeAssistVersion       = "v1internal"
-	geminiOauthClientID     = "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com"
-	geminiOauthClientSecret = "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"
+	geminiOAuthClientID     = "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com"
+	geminiOAuthClientSecret = "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"
 )
 
-var geminiOauthScopes = []string{
+var geminiOAuthScopes = []string{
 	"https://www.googleapis.com/auth/cloud-platform",
 	"https://www.googleapis.com/auth/userinfo.email",
 	"https://www.googleapis.com/auth/userinfo.profile",
@@ -44,14 +47,24 @@ type GeminiCLIExecutor struct {
 	cfg *config.Config
 }
 
+// NewGeminiCLIExecutor creates a new Gemini CLI executor instance.
+//
+// Parameters:
+//   - cfg: The application configuration
+//
+// Returns:
+//   - *GeminiCLIExecutor: A new Gemini CLI executor instance
 func NewGeminiCLIExecutor(cfg *config.Config) *GeminiCLIExecutor {
 	return &GeminiCLIExecutor{cfg: cfg}
 }
 
+// Identifier returns the executor identifier.
 func (e *GeminiCLIExecutor) Identifier() string { return "gemini-cli" }
 
+// PrepareRequest prepares the HTTP request for execution (no-op for Gemini CLI).
 func (e *GeminiCLIExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
 
+// Execute performs a non-streaming request to the Gemini CLI API.
 func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	tokenSource, baseTokenData, err := prepareGeminiCLITokenSource(ctx, e.cfg, auth)
 	if err != nil {
@@ -64,6 +77,8 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 	to := sdktranslator.FromString("gemini-cli")
 	basePayload := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 	basePayload = applyThinkingMetadataCLI(basePayload, req.Metadata, req.Model)
+	basePayload = util.ApplyDefaultThinkingIfNeededCLI(req.Model, basePayload)
+	basePayload = util.NormalizeGeminiCLIThinkingBudget(req.Model, basePayload)
 	basePayload = util.StripThinkingConfigIfUnsupported(req.Model, basePayload)
 	basePayload = fixGeminiCLIImageAspectRatio(req.Model, basePayload)
 	basePayload = applyPayloadConfigWithRoot(e.cfg, req.Model, "gemini", "request", basePayload)
@@ -187,6 +202,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 	return resp, err
 }
 
+// ExecuteStream performs a streaming request to the Gemini CLI API.
 func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
 	tokenSource, baseTokenData, err := prepareGeminiCLITokenSource(ctx, e.cfg, auth)
 	if err != nil {
@@ -199,6 +215,8 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 	to := sdktranslator.FromString("gemini-cli")
 	basePayload := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 	basePayload = applyThinkingMetadataCLI(basePayload, req.Metadata, req.Model)
+	basePayload = util.ApplyDefaultThinkingIfNeededCLI(req.Model, basePayload)
+	basePayload = util.NormalizeGeminiCLIThinkingBudget(req.Model, basePayload)
 	basePayload = util.StripThinkingConfigIfUnsupported(req.Model, basePayload)
 	basePayload = fixGeminiCLIImageAspectRatio(req.Model, basePayload)
 	basePayload = applyPayloadConfigWithRoot(e.cfg, req.Model, "gemini", "request", basePayload)
@@ -305,7 +323,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 			}()
 			if opts.Alt == "" {
 				scanner := bufio.NewScanner(resp.Body)
-				scanner.Buffer(nil, 20_971_520)
+				scanner.Buffer(nil, streamScannerBuffer)
 				var param any
 				for scanner.Scan() {
 					line := scanner.Bytes()
@@ -367,6 +385,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 	return nil, err
 }
 
+// CountTokens counts tokens for the given request using the Gemini CLI API.
 func (e *GeminiCLIExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
 	tokenSource, baseTokenData, err := prepareGeminiCLITokenSource(ctx, e.cfg, auth)
 	if err != nil {
@@ -467,9 +486,8 @@ func (e *GeminiCLIExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.
 	return cliproxyexecutor.Response{}, newGeminiStatusErr(lastStatus, lastBody)
 }
 
-func (e *GeminiCLIExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
-	log.Debugf("gemini cli executor: refresh called")
-	_ = ctx
+// Refresh refreshes the authentication credentials (no-op for Gemini CLI).
+func (e *GeminiCLIExecutor) Refresh(_ context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	return auth, nil
 }
 
@@ -511,9 +529,9 @@ func prepareGeminiCLITokenSource(ctx context.Context, cfg *config.Config, auth *
 	}
 
 	conf := &oauth2.Config{
-		ClientID:     geminiOauthClientID,
-		ClientSecret: geminiOauthClientSecret,
-		Scopes:       geminiOauthScopes,
+		ClientID:     geminiOAuthClientID,
+		ClientSecret: geminiOAuthClientSecret,
+		Scopes:       geminiOAuthScopes,
 		Endpoint:     google.Endpoint,
 	}
 

internal/runtime/executor/gemini_executor.go

@@ -11,7 +11,6 @@ import (
 	"io"
 	"net/http"
 	"strings"
-	"time"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
@@ -21,8 +20,6 @@ import (
 	log "github.com/sirupsen/logrus"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
-	"golang.org/x/oauth2"
-	"golang.org/x/oauth2/google"
 )
 
 const (
@@ -31,6 +28,9 @@ const (
 
 	// glAPIVersion is the API version used for Gemini requests.
 	glAPIVersion = "v1beta"
+
+	// streamScannerBuffer is the buffer size for SSE stream scanning.
+	streamScannerBuffer = 52_428_800
 )
 
 // GeminiExecutor is a stateless executor for the official Gemini API using API keys.
@@ -48,9 +48,11 @@ type GeminiExecutor struct {
 //
 // Returns:
 //   - *GeminiExecutor: A new Gemini executor instance
-func NewGeminiExecutor(cfg *config.Config) *GeminiExecutor { return &GeminiExecutor{cfg: cfg} }
+func NewGeminiExecutor(cfg *config.Config) *GeminiExecutor {
+	return &GeminiExecutor{cfg: cfg}
+}
 
-// Identifier returns the executor identifier for Gemini.
+// Identifier returns the executor identifier.
 func (e *GeminiExecutor) Identifier() string { return "gemini" }
 
 // PrepareRequest prepares the HTTP request for execution (no-op for Gemini).
@@ -75,14 +77,19 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	// Official Gemini API via API key or OAuth bearer
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 	body = applyThinkingMetadata(body, req.Metadata, req.Model)
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	action := "generateContent"
 	if req.Metadata != nil {
@@ -91,7 +98,7 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 		}
 	}
 	baseURL := resolveGeminiBaseURL(auth)
-	url := fmt.Sprintf("%s/%s/models/%s:%s", baseURL, glAPIVersion, req.Model, action)
+	url := fmt.Sprintf("%s/%s/models/%s:%s", baseURL, glAPIVersion, upstreamModel, action)
 	if opts.Alt != "" && action != "countTokens" {
 		url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
 	}
@@ -159,22 +166,28 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	return resp, nil
 }
 
+// ExecuteStream performs a streaming request to the Gemini API.
 func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
 	apiKey, bearer := geminiCreds(auth)
 
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 	body = applyThinkingMetadata(body, req.Metadata, req.Model)
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	baseURL := resolveGeminiBaseURL(auth)
-	url := fmt.Sprintf("%s/%s/models/%s:%s", baseURL, glAPIVersion, req.Model, "streamGenerateContent")
+	url := fmt.Sprintf("%s/%s/models/%s:%s", baseURL, glAPIVersion, upstreamModel, "streamGenerateContent")
 	if opts.Alt == "" {
 		url = url + "?alt=sse"
 	} else {
@@ -239,7 +252,7 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		scanner.Buffer(nil, 20_971_520)
+		scanner.Buffer(nil, streamScannerBuffer)
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -270,6 +283,7 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	return stream, nil
 }
 
+// CountTokens counts tokens for the given request using the Gemini API.
 func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
 	apiKey, bearer := geminiCreds(auth)
 
@@ -343,106 +357,8 @@ func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
 }
 
-func (e *GeminiExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
-	log.Debugf("gemini executor: refresh called")
-	// OAuth bearer token refresh for official Gemini API.
-	if auth == nil {
-		return nil, fmt.Errorf("gemini executor: auth is nil")
-	}
-	if auth.Metadata == nil {
-		return auth, nil
-	}
-	// Token data is typically nested under "token" map in Gemini files.
-	tokenMap, _ := auth.Metadata["token"].(map[string]any)
-	var refreshToken, accessToken, clientID, clientSecret, tokenURI, expiryStr string
-	if tokenMap != nil {
-		if v, ok := tokenMap["refresh_token"].(string); ok {
-			refreshToken = v
-		}
-		if v, ok := tokenMap["access_token"].(string); ok {
-			accessToken = v
-		}
-		if v, ok := tokenMap["client_id"].(string); ok {
-			clientID = v
-		}
-		if v, ok := tokenMap["client_secret"].(string); ok {
-			clientSecret = v
-		}
-		if v, ok := tokenMap["token_uri"].(string); ok {
-			tokenURI = v
-		}
-		if v, ok := tokenMap["expiry"].(string); ok {
-			expiryStr = v
-		}
-	} else {
-		// Fallback to top-level keys if present
-		if v, ok := auth.Metadata["refresh_token"].(string); ok {
-			refreshToken = v
-		}
-		if v, ok := auth.Metadata["access_token"].(string); ok {
-			accessToken = v
-		}
-		if v, ok := auth.Metadata["client_id"].(string); ok {
-			clientID = v
-		}
-		if v, ok := auth.Metadata["client_secret"].(string); ok {
-			clientSecret = v
-		}
-		if v, ok := auth.Metadata["token_uri"].(string); ok {
-			tokenURI = v
-		}
-		if v, ok := auth.Metadata["expiry"].(string); ok {
-			expiryStr = v
-		}
-	}
-	if refreshToken == "" {
-		// Nothing to do for API key or cookie based entries
-		return auth, nil
-	}
-
-	// Prepare oauth2 config; default to Google endpoints
-	endpoint := google.Endpoint
-	if tokenURI != "" {
-		endpoint.TokenURL = tokenURI
-	}
-	conf := &oauth2.Config{ClientID: clientID, ClientSecret: clientSecret, Endpoint: endpoint}
-
-	// Ensure proxy-aware HTTP client for token refresh
-	httpClient := util.SetProxy(&e.cfg.SDKConfig, &http.Client{})
-	ctx = context.WithValue(ctx, oauth2.HTTPClient, httpClient)
-
-	// Build base token
-	tok := &oauth2.Token{AccessToken: accessToken, RefreshToken: refreshToken}
-	if t, err := time.Parse(time.RFC3339, expiryStr); err == nil {
-		tok.Expiry = t
-	}
-	newTok, err := conf.TokenSource(ctx, tok).Token()
-	if err != nil {
-		return nil, err
-	}
-
-	// Persist back to metadata; prefer nested token map if present
-	if tokenMap == nil {
-		tokenMap = make(map[string]any)
-	}
-	tokenMap["access_token"] = newTok.AccessToken
-	tokenMap["refresh_token"] = newTok.RefreshToken
-	tokenMap["expiry"] = newTok.Expiry.Format(time.RFC3339)
-	if clientID != "" {
-		tokenMap["client_id"] = clientID
-	}
-	if clientSecret != "" {
-		tokenMap["client_secret"] = clientSecret
-	}
-	if tokenURI != "" {
-		tokenMap["token_uri"] = tokenURI
-	}
-	auth.Metadata["token"] = tokenMap
-
-	// Also mirror top-level access_token for compatibility if previously present
-	if _, ok := auth.Metadata["access_token"]; ok {
-		auth.Metadata["access_token"] = newTok.AccessToken
-	}
+// Refresh refreshes the authentication credentials (no-op for Gemini API key).
+func (e *GeminiExecutor) Refresh(_ context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	return auth, nil
 }
 

internal/runtime/executor/gemini_vertex_executor.go

@@ -1,6 +1,6 @@
-// Package executor contains provider executors. This file implements the Vertex AI
-// Gemini executor that talks to Google Vertex AI endpoints using service account
-// credentials imported by the CLI.
+// Package executor provides runtime execution capabilities for various AI service providers.
+// This file implements the Vertex AI Gemini executor that talks to Google Vertex AI
+// endpoints using service account credentials or API keys.
 package executor
 
 import (
@@ -36,20 +36,26 @@ type GeminiVertexExecutor struct {
 	cfg *config.Config
 }
 
-// NewGeminiVertexExecutor constructs the Vertex executor.
+// NewGeminiVertexExecutor creates a new Vertex AI Gemini executor instance.
+//
+// Parameters:
+//   - cfg: The application configuration
+//
+// Returns:
+//   - *GeminiVertexExecutor: A new Vertex AI Gemini executor instance
 func NewGeminiVertexExecutor(cfg *config.Config) *GeminiVertexExecutor {
 	return &GeminiVertexExecutor{cfg: cfg}
 }
 
-// Identifier returns provider key for manager routing.
+// Identifier returns the executor identifier.
 func (e *GeminiVertexExecutor) Identifier() string { return "vertex" }
 
-// PrepareRequest is a no-op for Vertex.
+// PrepareRequest prepares the HTTP request for execution (no-op for Vertex).
 func (e *GeminiVertexExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error {
 	return nil
 }
 
-// Execute handles non-streaming requests.
+// Execute performs a non-streaming request to the Vertex AI API.
 func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	// Try API key authentication first
 	apiKey, baseURL := vertexAPICreds(auth)
@@ -67,7 +73,7 @@ func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	return e.executeWithAPIKey(ctx, auth, req, opts, apiKey, baseURL)
 }
 
-// ExecuteStream handles SSE streaming for Vertex.
+// ExecuteStream performs a streaming request to the Vertex AI API.
 func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
 	// Try API key authentication first
 	apiKey, baseURL := vertexAPICreds(auth)
@@ -85,7 +91,7 @@ func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 	return e.executeStreamWithAPIKey(ctx, auth, req, opts, apiKey, baseURL)
 }
 
-// CountTokens calls Vertex countTokens endpoint.
+// CountTokens counts tokens for the given request using the Vertex AI API.
 func (e *GeminiVertexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
 	// Try API key authentication first
 	apiKey, baseURL := vertexAPICreds(auth)
@@ -103,179 +109,7 @@ func (e *GeminiVertexExecutor) CountTokens(ctx context.Context, auth *cliproxyau
 	return e.countTokensWithAPIKey(ctx, auth, req, opts, apiKey, baseURL)
 }
 
-// countTokensWithServiceAccount handles token counting using service account credentials.
-func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (cliproxyexecutor.Response, error) {
-	from := opts.SourceFormat
-	to := sdktranslator.FromString("gemini")
-	translatedReq := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
-		if budgetOverride != nil {
-			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
-			budgetOverride = &norm
-		}
-		translatedReq = util.ApplyGeminiThinkingConfig(translatedReq, budgetOverride, includeOverride)
-	}
-	translatedReq = util.StripThinkingConfigIfUnsupported(req.Model, translatedReq)
-	translatedReq = fixGeminiImageAspectRatio(req.Model, translatedReq)
-	respCtx := context.WithValue(ctx, "alt", opts.Alt)
-	translatedReq, _ = sjson.DeleteBytes(translatedReq, "tools")
-	translatedReq, _ = sjson.DeleteBytes(translatedReq, "generationConfig")
-	translatedReq, _ = sjson.DeleteBytes(translatedReq, "safetySettings")
-
-	baseURL := vertexBaseURL(location)
-	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, req.Model, "countTokens")
-
-	httpReq, errNewReq := http.NewRequestWithContext(respCtx, http.MethodPost, url, bytes.NewReader(translatedReq))
-	if errNewReq != nil {
-		return cliproxyexecutor.Response{}, errNewReq
-	}
-	httpReq.Header.Set("Content-Type", "application/json")
-	if token, errTok := vertexAccessToken(ctx, e.cfg, auth, saJSON); errTok == nil && token != "" {
-		httpReq.Header.Set("Authorization", "Bearer "+token)
-	} else if errTok != nil {
-		log.Errorf("vertex executor: access token error: %v", errTok)
-		return cliproxyexecutor.Response{}, statusErr{code: 500, msg: "internal server error"}
-	}
-	applyGeminiHeaders(httpReq, auth)
-
-	var authID, authLabel, authType, authValue string
-	if auth != nil {
-		authID = auth.ID
-		authLabel = auth.Label
-		authType, authValue = auth.AccountInfo()
-	}
-	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
-		URL:       url,
-		Method:    http.MethodPost,
-		Headers:   httpReq.Header.Clone(),
-		Body:      translatedReq,
-		Provider:  e.Identifier(),
-		AuthID:    authID,
-		AuthLabel: authLabel,
-		AuthType:  authType,
-		AuthValue: authValue,
-	})
-
-	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
-	httpResp, errDo := httpClient.Do(httpReq)
-	if errDo != nil {
-		recordAPIResponseError(ctx, e.cfg, errDo)
-		return cliproxyexecutor.Response{}, errDo
-	}
-	defer func() {
-		if errClose := httpResp.Body.Close(); errClose != nil {
-			log.Errorf("vertex executor: close response body error: %v", errClose)
-		}
-	}()
-	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
-	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
-		b, _ := io.ReadAll(httpResp.Body)
-		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
-		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(b)}
-	}
-	data, errRead := io.ReadAll(httpResp.Body)
-	if errRead != nil {
-		recordAPIResponseError(ctx, e.cfg, errRead)
-		return cliproxyexecutor.Response{}, errRead
-	}
-	appendAPIResponseChunk(ctx, e.cfg, data)
-	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
-		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(data)}
-	}
-	count := gjson.GetBytes(data, "totalTokens").Int()
-	out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
-	return cliproxyexecutor.Response{Payload: []byte(out)}, nil
-}
-
-// countTokensWithAPIKey handles token counting using API key credentials.
-func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (cliproxyexecutor.Response, error) {
-	from := opts.SourceFormat
-	to := sdktranslator.FromString("gemini")
-	translatedReq := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
-		if budgetOverride != nil {
-			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
-			budgetOverride = &norm
-		}
-		translatedReq = util.ApplyGeminiThinkingConfig(translatedReq, budgetOverride, includeOverride)
-	}
-	translatedReq = util.StripThinkingConfigIfUnsupported(req.Model, translatedReq)
-	translatedReq = fixGeminiImageAspectRatio(req.Model, translatedReq)
-	respCtx := context.WithValue(ctx, "alt", opts.Alt)
-	translatedReq, _ = sjson.DeleteBytes(translatedReq, "tools")
-	translatedReq, _ = sjson.DeleteBytes(translatedReq, "generationConfig")
-	translatedReq, _ = sjson.DeleteBytes(translatedReq, "safetySettings")
-
-	// For API key auth, use simpler URL format without project/location
-	if baseURL == "" {
-		baseURL = "https://generativelanguage.googleapis.com"
-	}
-	url := fmt.Sprintf("%s/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, req.Model, "countTokens")
-
-	httpReq, errNewReq := http.NewRequestWithContext(respCtx, http.MethodPost, url, bytes.NewReader(translatedReq))
-	if errNewReq != nil {
-		return cliproxyexecutor.Response{}, errNewReq
-	}
-	httpReq.Header.Set("Content-Type", "application/json")
-	if apiKey != "" {
-		httpReq.Header.Set("x-goog-api-key", apiKey)
-	}
-	applyGeminiHeaders(httpReq, auth)
-
-	var authID, authLabel, authType, authValue string
-	if auth != nil {
-		authID = auth.ID
-		authLabel = auth.Label
-		authType, authValue = auth.AccountInfo()
-	}
-	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
-		URL:       url,
-		Method:    http.MethodPost,
-		Headers:   httpReq.Header.Clone(),
-		Body:      translatedReq,
-		Provider:  e.Identifier(),
-		AuthID:    authID,
-		AuthLabel: authLabel,
-		AuthType:  authType,
-		AuthValue: authValue,
-	})
-
-	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
-	httpResp, errDo := httpClient.Do(httpReq)
-	if errDo != nil {
-		recordAPIResponseError(ctx, e.cfg, errDo)
-		return cliproxyexecutor.Response{}, errDo
-	}
-	defer func() {
-		if errClose := httpResp.Body.Close(); errClose != nil {
-			log.Errorf("vertex executor: close response body error: %v", errClose)
-		}
-	}()
-	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
-	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
-		b, _ := io.ReadAll(httpResp.Body)
-		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
-		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(b)}
-	}
-	data, errRead := io.ReadAll(httpResp.Body)
-	if errRead != nil {
-		recordAPIResponseError(ctx, e.cfg, errRead)
-		return cliproxyexecutor.Response{}, errRead
-	}
-	appendAPIResponseChunk(ctx, e.cfg, data)
-	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
-		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(data)}
-	}
-	count := gjson.GetBytes(data, "totalTokens").Int()
-	out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
-	return cliproxyexecutor.Response{Payload: []byte(out)}, nil
-}
-
-// Refresh is a no-op for service account based credentials.
+// Refresh refreshes the authentication credentials (no-op for Vertex).
 func (e *GeminiVertexExecutor) Refresh(_ context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	return auth, nil
 }
@@ -286,19 +120,24 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
 		if budgetOverride != nil {
 			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
 			budgetOverride = &norm
 		}
 		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
 	}
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	action := "generateContent"
 	if req.Metadata != nil {
@@ -307,7 +146,7 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 		}
 	}
 	baseURL := vertexBaseURL(location)
-	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, req.Model, action)
+	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, upstreamModel, action)
 	if opts.Alt != "" && action != "countTokens" {
 		url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
 	}
@@ -381,19 +220,24 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
 		if budgetOverride != nil {
 			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
 			budgetOverride = &norm
 		}
 		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
 	}
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	action := "generateContent"
 	if req.Metadata != nil {
@@ -406,7 +250,7 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 	if baseURL == "" {
 		baseURL = "https://generativelanguage.googleapis.com"
 	}
-	url := fmt.Sprintf("%s/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, req.Model, action)
+	url := fmt.Sprintf("%s/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, upstreamModel, action)
 	if opts.Alt != "" && action != "countTokens" {
 		url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
 	}
@@ -477,22 +321,27 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
 		if budgetOverride != nil {
 			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
 			budgetOverride = &norm
 		}
 		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
 	}
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	baseURL := vertexBaseURL(location)
-	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, req.Model, "streamGenerateContent")
+	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, upstreamModel, "streamGenerateContent")
 	if opts.Alt == "" {
 		url = url + "?alt=sse"
 	} else {
@@ -558,7 +407,7 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		scanner.Buffer(nil, 20_971_520)
+		scanner.Buffer(nil, streamScannerBuffer)
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -589,25 +438,30 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
 		if budgetOverride != nil {
 			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
 			budgetOverride = &norm
 		}
 		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
 	}
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	// For API key auth, use simpler URL format without project/location
 	if baseURL == "" {
 		baseURL = "https://generativelanguage.googleapis.com"
 	}
-	url := fmt.Sprintf("%s/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, req.Model, "streamGenerateContent")
+	url := fmt.Sprintf("%s/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, upstreamModel, "streamGenerateContent")
 	if opts.Alt == "" {
 		url = url + "?alt=sse"
 	} else {
@@ -670,7 +524,7 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		scanner.Buffer(nil, 20_971_520)
+		scanner.Buffer(nil, streamScannerBuffer)
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -696,6 +550,184 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 	return stream, nil
 }
 
+// countTokensWithServiceAccount counts tokens using service account credentials.
+func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (cliproxyexecutor.Response, error) {
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
+	from := opts.SourceFormat
+	to := sdktranslator.FromString("gemini")
+	translatedReq := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+		if budgetOverride != nil {
+			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
+			budgetOverride = &norm
+		}
+		translatedReq = util.ApplyGeminiThinkingConfig(translatedReq, budgetOverride, includeOverride)
+	}
+	translatedReq = util.StripThinkingConfigIfUnsupported(req.Model, translatedReq)
+	translatedReq = fixGeminiImageAspectRatio(req.Model, translatedReq)
+	translatedReq, _ = sjson.SetBytes(translatedReq, "model", upstreamModel)
+	respCtx := context.WithValue(ctx, "alt", opts.Alt)
+	translatedReq, _ = sjson.DeleteBytes(translatedReq, "tools")
+	translatedReq, _ = sjson.DeleteBytes(translatedReq, "generationConfig")
+	translatedReq, _ = sjson.DeleteBytes(translatedReq, "safetySettings")
+
+	baseURL := vertexBaseURL(location)
+	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, upstreamModel, "countTokens")
+
+	httpReq, errNewReq := http.NewRequestWithContext(respCtx, http.MethodPost, url, bytes.NewReader(translatedReq))
+	if errNewReq != nil {
+		return cliproxyexecutor.Response{}, errNewReq
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+	if token, errTok := vertexAccessToken(ctx, e.cfg, auth, saJSON); errTok == nil && token != "" {
+		httpReq.Header.Set("Authorization", "Bearer "+token)
+	} else if errTok != nil {
+		log.Errorf("vertex executor: access token error: %v", errTok)
+		return cliproxyexecutor.Response{}, statusErr{code: 500, msg: "internal server error"}
+	}
+	applyGeminiHeaders(httpReq, auth)
+
+	var authID, authLabel, authType, authValue string
+	if auth != nil {
+		authID = auth.ID
+		authLabel = auth.Label
+		authType, authValue = auth.AccountInfo()
+	}
+	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
+		URL:       url,
+		Method:    http.MethodPost,
+		Headers:   httpReq.Header.Clone(),
+		Body:      translatedReq,
+		Provider:  e.Identifier(),
+		AuthID:    authID,
+		AuthLabel: authLabel,
+		AuthType:  authType,
+		AuthValue: authValue,
+	})
+
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	httpResp, errDo := httpClient.Do(httpReq)
+	if errDo != nil {
+		recordAPIResponseError(ctx, e.cfg, errDo)
+		return cliproxyexecutor.Response{}, errDo
+	}
+	defer func() {
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("vertex executor: close response body error: %v", errClose)
+		}
+	}()
+	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
+		b, _ := io.ReadAll(httpResp.Body)
+		appendAPIResponseChunk(ctx, e.cfg, b)
+		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(b)}
+	}
+	data, errRead := io.ReadAll(httpResp.Body)
+	if errRead != nil {
+		recordAPIResponseError(ctx, e.cfg, errRead)
+		return cliproxyexecutor.Response{}, errRead
+	}
+	appendAPIResponseChunk(ctx, e.cfg, data)
+	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
+		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(data)}
+	}
+	count := gjson.GetBytes(data, "totalTokens").Int()
+	out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
+	return cliproxyexecutor.Response{Payload: []byte(out)}, nil
+}
+
+// countTokensWithAPIKey handles token counting using API key credentials.
+func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (cliproxyexecutor.Response, error) {
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
+	from := opts.SourceFormat
+	to := sdktranslator.FromString("gemini")
+	translatedReq := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+		if budgetOverride != nil {
+			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
+			budgetOverride = &norm
+		}
+		translatedReq = util.ApplyGeminiThinkingConfig(translatedReq, budgetOverride, includeOverride)
+	}
+	translatedReq = util.StripThinkingConfigIfUnsupported(req.Model, translatedReq)
+	translatedReq = fixGeminiImageAspectRatio(req.Model, translatedReq)
+	translatedReq, _ = sjson.SetBytes(translatedReq, "model", upstreamModel)
+	respCtx := context.WithValue(ctx, "alt", opts.Alt)
+	translatedReq, _ = sjson.DeleteBytes(translatedReq, "tools")
+	translatedReq, _ = sjson.DeleteBytes(translatedReq, "generationConfig")
+	translatedReq, _ = sjson.DeleteBytes(translatedReq, "safetySettings")
+
+	// For API key auth, use simpler URL format without project/location
+	if baseURL == "" {
+		baseURL = "https://generativelanguage.googleapis.com"
+	}
+	url := fmt.Sprintf("%s/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, req.Model, "countTokens")
+
+	httpReq, errNewReq := http.NewRequestWithContext(respCtx, http.MethodPost, url, bytes.NewReader(translatedReq))
+	if errNewReq != nil {
+		return cliproxyexecutor.Response{}, errNewReq
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+	if apiKey != "" {
+		httpReq.Header.Set("x-goog-api-key", apiKey)
+	}
+	applyGeminiHeaders(httpReq, auth)
+
+	var authID, authLabel, authType, authValue string
+	if auth != nil {
+		authID = auth.ID
+		authLabel = auth.Label
+		authType, authValue = auth.AccountInfo()
+	}
+	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
+		URL:       url,
+		Method:    http.MethodPost,
+		Headers:   httpReq.Header.Clone(),
+		Body:      translatedReq,
+		Provider:  e.Identifier(),
+		AuthID:    authID,
+		AuthLabel: authLabel,
+		AuthType:  authType,
+		AuthValue: authValue,
+	})
+
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	httpResp, errDo := httpClient.Do(httpReq)
+	if errDo != nil {
+		recordAPIResponseError(ctx, e.cfg, errDo)
+		return cliproxyexecutor.Response{}, errDo
+	}
+	defer func() {
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("vertex executor: close response body error: %v", errClose)
+		}
+	}()
+	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
+		b, _ := io.ReadAll(httpResp.Body)
+		appendAPIResponseChunk(ctx, e.cfg, b)
+		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(b)}
+	}
+	data, errRead := io.ReadAll(httpResp.Body)
+	if errRead != nil {
+		recordAPIResponseError(ctx, e.cfg, errRead)
+		return cliproxyexecutor.Response{}, errRead
+	}
+	appendAPIResponseChunk(ctx, e.cfg, data)
+	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
+		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(data)}
+	}
+	count := gjson.GetBytes(data, "totalTokens").Int()
+	out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
+	return cliproxyexecutor.Response{Payload: []byte(out)}, nil
+}
+
 // vertexCreds extracts project, location and raw service account JSON from auth metadata.
 func vertexCreds(a *cliproxyauth.Auth) (projectID, location string, serviceAccountJSON []byte, err error) {
 	if a == nil || a.Metadata == nil {

internal/runtime/executor/iflow_executor.go

@@ -57,6 +57,15 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	}
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return resp, errValidate
+	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 
 	endpoint := strings.TrimSuffix(baseURL, "/") + iflowDefaultEndpoint
@@ -139,6 +148,15 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	to := sdktranslator.FromString("openai")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	}
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return nil, errValidate
+	}
 	// Ensure tools array exists to avoid provider quirks similar to Qwen's behaviour.
 	toolsResult := gjson.GetBytes(body, "tools")
 	if toolsResult.Exists() && toolsResult.IsArray() && len(toolsResult.Array()) == 0 {
@@ -201,7 +219,7 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 		}()
 
 		scanner := bufio.NewScanner(httpResp.Body)
-		scanner.Buffer(nil, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB 
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()

internal/runtime/executor/openai_compat_executor.go

@@ -58,6 +58,15 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 		translated = e.overrideModel(translated, modelOverride)
 	}
 	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", translated)
+	translated = applyReasoningEffortMetadata(translated, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		translated, _ = sjson.SetBytes(translated, "model", upstreamModel)
+	}
+	translated = normalizeThinkingConfig(translated, upstreamModel)
+	if errValidate := validateThinkingConfig(translated, upstreamModel); errValidate != nil {
+		return resp, errValidate
+	}
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(translated))
@@ -143,6 +152,15 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 		translated = e.overrideModel(translated, modelOverride)
 	}
 	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", translated)
+	translated = applyReasoningEffortMetadata(translated, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		translated, _ = sjson.SetBytes(translated, "model", upstreamModel)
+	}
+	translated = normalizeThinkingConfig(translated, upstreamModel)
+	if errValidate := validateThinkingConfig(translated, upstreamModel); errValidate != nil {
+		return nil, errValidate
+	}
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(translated))
@@ -206,7 +224,7 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		scanner.Buffer(nil, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()

internal/runtime/executor/payload_helpers.go

@@ -1,6 +1,8 @@
 package executor
 
 import (
+	"fmt"
+	"net/http"
 	"strings"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
@@ -9,11 +11,11 @@ import (
 	"github.com/tidwall/sjson"
 )
 
-// applyThinkingMetadata applies thinking config from model suffix metadata (e.g., -reasoning, -thinking-N)
+// applyThinkingMetadata applies thinking config from model suffix metadata (e.g., (high), (8192))
 // for standard Gemini format payloads. It normalizes the budget when the model supports thinking.
 func applyThinkingMetadata(payload []byte, metadata map[string]any, model string) []byte {
-	budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(metadata)
-	if !ok {
+	budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(model, metadata)
+	if !ok || (budgetOverride == nil && includeOverride == nil) {
 		return payload
 	}
 	if !util.ModelSupportsThinking(model) {
@@ -26,20 +28,44 @@ func applyThinkingMetadata(payload []byte, metadata map[string]any, model string
 	return util.ApplyGeminiThinkingConfig(payload, budgetOverride, includeOverride)
 }
 
-// applyThinkingMetadataCLI applies thinking config from model suffix metadata (e.g., -reasoning, -thinking-N)
+// applyThinkingMetadataCLI applies thinking config from model suffix metadata (e.g., (high), (8192))
 // for Gemini CLI format payloads (nested under "request"). It normalizes the budget when the model supports thinking.
 func applyThinkingMetadataCLI(payload []byte, metadata map[string]any, model string) []byte {
-	budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(metadata)
-	if !ok {
+	budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(model, metadata)
+	if !ok || (budgetOverride == nil && includeOverride == nil) {
 		return payload
 	}
-	if budgetOverride != nil && util.ModelSupportsThinking(model) {
+	if !util.ModelSupportsThinking(model) {
+		return payload
+	}
+	if budgetOverride != nil {
 		norm := util.NormalizeThinkingBudget(model, *budgetOverride)
 		budgetOverride = &norm
 	}
 	return util.ApplyGeminiCLIThinkingConfig(payload, budgetOverride, includeOverride)
 }
 
+// applyReasoningEffortMetadata applies reasoning effort overrides from metadata to the given JSON path.
+// Metadata values take precedence over any existing field when the model supports thinking, intentionally
+// overwriting caller-provided values to honor suffix/default metadata priority.
+func applyReasoningEffortMetadata(payload []byte, metadata map[string]any, model, field string) []byte {
+	if len(metadata) == 0 {
+		return payload
+	}
+	if !util.ModelSupportsThinking(model) {
+		return payload
+	}
+	if field == "" {
+		return payload
+	}
+	if effort, ok := util.ReasoningEffortFromMetadata(metadata); ok && effort != "" {
+		if updated, err := sjson.SetBytes(payload, field, effort); err == nil {
+			return updated
+		}
+	}
+	return payload
+}
+
 // applyPayloadConfig applies payload default and override rules from configuration
 // to the given JSON payload for the specified model.
 // Defaults only fill missing fields, while overrides always overwrite existing values.
@@ -189,3 +215,93 @@ func matchModelPattern(pattern, model string) bool {
 	}
 	return pi == len(pattern)
 }
+
+// normalizeThinkingConfig normalizes thinking-related fields in the payload
+// based on model capabilities. For models without thinking support, it strips
+// reasoning fields. For models with level-based thinking, it validates and
+// normalizes the reasoning effort level.
+func normalizeThinkingConfig(payload []byte, model string) []byte {
+	if len(payload) == 0 || model == "" {
+		return payload
+	}
+
+	if !util.ModelSupportsThinking(model) {
+		return stripThinkingFields(payload)
+	}
+
+	if util.ModelUsesThinkingLevels(model) {
+		return normalizeReasoningEffortLevel(payload, model)
+	}
+
+	return payload
+}
+
+// stripThinkingFields removes thinking-related fields from the payload for
+// models that do not support thinking.
+func stripThinkingFields(payload []byte) []byte {
+	fieldsToRemove := []string{
+		"reasoning",
+		"reasoning_effort",
+		"reasoning.effort",
+	}
+	out := payload
+	for _, field := range fieldsToRemove {
+		if gjson.GetBytes(out, field).Exists() {
+			out, _ = sjson.DeleteBytes(out, field)
+		}
+	}
+	return out
+}
+
+// normalizeReasoningEffortLevel validates and normalizes the reasoning_effort
+// or reasoning.effort field for level-based thinking models.
+func normalizeReasoningEffortLevel(payload []byte, model string) []byte {
+	out := payload
+
+	if effort := gjson.GetBytes(out, "reasoning_effort"); effort.Exists() {
+		if normalized, ok := util.NormalizeReasoningEffortLevel(model, effort.String()); ok {
+			out, _ = sjson.SetBytes(out, "reasoning_effort", normalized)
+		}
+	}
+
+	if effort := gjson.GetBytes(out, "reasoning.effort"); effort.Exists() {
+		if normalized, ok := util.NormalizeReasoningEffortLevel(model, effort.String()); ok {
+			out, _ = sjson.SetBytes(out, "reasoning.effort", normalized)
+		}
+	}
+
+	return out
+}
+
+// validateThinkingConfig checks for unsupported reasoning levels on level-based models.
+// Returns a statusErr with 400 when an unsupported level is supplied to avoid silently
+// downgrading requests.
+func validateThinkingConfig(payload []byte, model string) error {
+	if len(payload) == 0 || model == "" {
+		return nil
+	}
+	if !util.ModelSupportsThinking(model) || !util.ModelUsesThinkingLevels(model) {
+		return nil
+	}
+
+	levels := util.GetModelThinkingLevels(model)
+	checkField := func(path string) error {
+		if effort := gjson.GetBytes(payload, path); effort.Exists() {
+			if _, ok := util.NormalizeReasoningEffortLevel(model, effort.String()); !ok {
+				return statusErr{
+					code: http.StatusBadRequest,
+					msg:  fmt.Sprintf("unsupported reasoning effort level %q for model %s (supported: %s)", effort.String(), model, strings.Join(levels, ", ")),
+				}
+			}
+		}
+		return nil
+	}
+
+	if err := checkField("reasoning_effort"); err != nil {
+		return err
+	}
+	if err := checkField("reasoning.effort"); err != nil {
+		return err
+	}
+	return nil
+}

internal/runtime/executor/qwen_executor.go

@@ -12,6 +12,7 @@ import (
 
 	qwenauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/qwen"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
 	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
 	sdktranslator "github.com/router-for-me/CLIProxyAPI/v6/sdk/translator"
@@ -50,6 +51,15 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	}
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return resp, errValidate
+	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
@@ -121,6 +131,15 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 	to := sdktranslator.FromString("openai")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	}
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return nil, errValidate
+	}
 	toolsResult := gjson.GetBytes(body, "tools")
 	// I'm addressing the Qwen3 "poisoning" issue, which is caused by the model needing a tool to be defined. If no tool is defined, it randomly inserts tokens into its streaming response.
 	// This will have no real consequences. It's just to scare Qwen3.
@@ -181,7 +200,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		scanner.Buffer(nil, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()

internal/translator/antigravity/claude/antigravity_claude_request.go

@@ -83,7 +83,15 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 				for j := 0; j < len(contentResults); j++ {
 					contentResult := contentResults[j]
 					contentTypeResult := contentResult.Get("type")
-					if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "text" {
+					if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "thinking" {
+						prompt := contentResult.Get("thinking").String()
+						signatureResult := contentResult.Get("signature")
+						signature := geminiCLIClaudeThoughtSignature
+						if signatureResult.Exists() {
+							signature = signatureResult.String()
+						}
+						clientContent.Parts = append(clientContent.Parts, client.Part{Text: prompt, Thought: true, ThoughtSignature: signature})
+					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "text" {
 						prompt := contentResult.Get("text").String()
 						clientContent.Parts = append(clientContent.Parts, client.Part{Text: prompt})
 					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "tool_use" {
@@ -92,10 +100,16 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 						functionID := contentResult.Get("id").String()
 						var args map[string]any
 						if err := json.Unmarshal([]byte(functionArgs), &args); err == nil {
-							clientContent.Parts = append(clientContent.Parts, client.Part{
-								FunctionCall:     &client.FunctionCall{ID: functionID, Name: functionName, Args: args},
-								ThoughtSignature: geminiCLIClaudeThoughtSignature,
-							})
+							if strings.Contains(modelName, "claude") {
+								clientContent.Parts = append(clientContent.Parts, client.Part{
+									FunctionCall: &client.FunctionCall{ID: functionID, Name: functionName, Args: args},
+								})
+							} else {
+								clientContent.Parts = append(clientContent.Parts, client.Part{
+									FunctionCall:     &client.FunctionCall{ID: functionID, Name: functionName, Args: args},
+									ThoughtSignature: geminiCLIClaudeThoughtSignature,
+								})
+							}
 						}
 					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "tool_result" {
 						toolCallID := contentResult.Get("tool_use_id").String()
@@ -109,6 +123,15 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 							functionResponse := client.FunctionResponse{ID: toolCallID, Name: funcName, Response: map[string]interface{}{"result": responseData}}
 							clientContent.Parts = append(clientContent.Parts, client.Part{FunctionResponse: &functionResponse})
 						}
+					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "image" {
+						sourceResult := contentResult.Get("source")
+						if sourceResult.Get("type").String() == "base64" {
+							inlineData := &client.InlineData{
+								MimeType: sourceResult.Get("media_type").String(),
+								Data:     sourceResult.Get("data").String(),
+							}
+							clientContent.Parts = append(clientContent.Parts, client.Part{InlineData: inlineData})
+						}
 					}
 				}
 				contents = append(contents, clientContent)
@@ -166,7 +189,6 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 		if t.Get("type").String() == "enabled" {
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())
-				budget = util.NormalizeThinkingBudget(modelName, budget)
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 			}

internal/translator/antigravity/claude/antigravity_claude_response.go

@@ -12,6 +12,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"github.com/tidwall/gjson"
@@ -34,8 +35,12 @@ type Params struct {
 	TotalTokenCount      int64  // Cached total token count from usage metadata
 	HasSentFinalEvents   bool   // Indicates if final content/message events have been sent
 	HasToolUse           bool   // Indicates if tool use was observed in the stream
+	HasContent           bool   // Tracks whether any content (text, thinking, or tool use) has been output
 }
 
+// toolUseIDCounter provides a process-wide unique counter for tool use identifiers.
+var toolUseIDCounter uint64
+
 // ConvertAntigravityResponseToClaude performs sophisticated streaming response format conversion.
 // This function implements a complex state machine that translates backend client responses
 // into Claude Code-compatible Server-Sent Events (SSE) format. It manages different response types
@@ -65,11 +70,14 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 
 	if bytes.Equal(rawJSON, []byte("[DONE]")) {
 		output := ""
-		appendFinalEvents(params, &output, true)
-
-		return []string{
-			output + "event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+		// Only send final events if we have actually output content
+		if params.HasContent {
+			appendFinalEvents(params, &output, true)
+			return []string{
+				output + "event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+			}
 		}
+		return []string{}
 	}
 
 	output := ""
@@ -111,11 +119,16 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 			if partTextResult.Exists() {
 				// Process thinking content (internal reasoning)
 				if partResult.Get("thought").Bool() {
-					// Continue existing thinking block if already in thinking state
-					if params.ResponseType == 2 {
+					if thoughtSignature := partResult.Get("thoughtSignature"); thoughtSignature.Exists() && thoughtSignature.String() != "" {
+						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())
+						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
 						output = output + "event: content_block_delta\n"
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, params.ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						params.HasContent = true
 					} else {
 						// Transition from another state to thinking
 						// First, close any existing content block
@@ -139,6 +152,7 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, params.ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						params.ResponseType = 2 // Set state to thinking
+						params.HasContent = true
 					}
 				} else {
 					finishReasonResult := gjson.GetBytes(rawJSON, "response.candidates.0.finishReason")
@@ -149,6 +163,7 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 							output = output + "event: content_block_delta\n"
 							data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, params.ResponseIndex), "delta.text", partTextResult.String())
 							output = output + fmt.Sprintf("data: %s\n\n\n", data)
+							params.HasContent = true
 						} else {
 							// Transition from another state to text content
 							// First, close any existing content block
@@ -163,15 +178,17 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 								output = output + "\n\n\n"
 								params.ResponseIndex++
 							}
-
-							// Start a new text content block
-							output = output + "event: content_block_start\n"
-							output = output + fmt.Sprintf(`data: {"type":"content_block_start","index":%d,"content_block":{"type":"text","text":""}}`, params.ResponseIndex)
-							output = output + "\n\n\n"
-							output = output + "event: content_block_delta\n"
-							data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, params.ResponseIndex), "delta.text", partTextResult.String())
-							output = output + fmt.Sprintf("data: %s\n\n\n", data)
-							params.ResponseType = 1 // Set state to content
+							if partTextResult.String() != "" {
+								// Start a new text content block
+								output = output + "event: content_block_start\n"
+								output = output + fmt.Sprintf(`data: {"type":"content_block_start","index":%d,"content_block":{"type":"text","text":""}}`, params.ResponseIndex)
+								output = output + "\n\n\n"
+								output = output + "event: content_block_delta\n"
+								data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, params.ResponseIndex), "delta.text", partTextResult.String())
+								output = output + fmt.Sprintf("data: %s\n\n\n", data)
+								params.ResponseType = 1 // Set state to content
+								params.HasContent = true
+							}
 						}
 					}
 				}
@@ -212,7 +229,7 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 
 				// Create the tool use block with unique ID and function details
 				data := fmt.Sprintf(`{"type":"content_block_start","index":%d,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`, params.ResponseIndex)
-				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&toolUseIDCounter, 1)))
 				data, _ = sjson.Set(data, "content_block.name", fcName)
 				output = output + fmt.Sprintf("data: %s\n\n\n", data)
 
@@ -222,6 +239,7 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 					output = output + fmt.Sprintf("data: %s\n\n\n", data)
 				}
 				params.ResponseType = 3
+				params.HasContent = true
 			}
 		}
 	}
@@ -261,6 +279,11 @@ func appendFinalEvents(params *Params, output *string, force bool) {
 		return
 	}
 
+	// Only send final events if we have actually output content
+	if !params.HasContent {
+		return
+	}
+
 	if params.ResponseType != 0 {
 		*output = *output + "event: content_block_stop\n"
 		*output = *output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, params.ResponseIndex)

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

@@ -48,13 +48,13 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		case "low":
-			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 1024))
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 1024)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		case "medium":
-			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 8192))
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 8192)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		case "high":
-			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 32768))
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 32768)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		default:
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
@@ -66,15 +66,15 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 	if !hasOfficialThinking && util.ModelSupportsThinking(modelName) {
 		if tc := gjson.GetBytes(rawJSON, "extra_body.google.thinking_config"); tc.Exists() && tc.IsObject() {
 			var setBudget bool
-			var normalized int
+			var budget int
 
 			if v := tc.Get("thinkingBudget"); v.Exists() {
-				normalized = util.NormalizeThinkingBudget(modelName, int(v.Int()))
-				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", normalized)
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				setBudget = true
 			} else if v := tc.Get("thinking_budget"); v.Exists() {
-				normalized = util.NormalizeThinkingBudget(modelName, int(v.Int()))
-				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", normalized)
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				setBudget = true
 			}
 
@@ -82,22 +82,27 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", v.Bool())
 			} else if v := tc.Get("include_thoughts"); v.Exists() {
 				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", v.Bool())
-			} else if setBudget && normalized != 0 {
+			} else if setBudget && budget != 0 {
 				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 			}
 		}
 	}
 
-	// For gemini-3-pro-preview, always send default thinkingConfig when none specified.
-	// This matches the official Gemini CLI behavior which always sends:
-	// { thinkingBudget: -1, includeThoughts: true }
-	// See: ai-gemini-cli/packages/core/src/config/defaultModelConfigs.ts
-	if !gjson.GetBytes(out, "request.generationConfig.thinkingConfig").Exists() && modelName == "gemini-3-pro-preview" {
-		out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
-		out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+	// Claude/Anthropic API format: thinking.type == "enabled" with budget_tokens
+	// This allows Claude Code and other Claude API clients to pass thinking configuration
+	if !gjson.GetBytes(out, "request.generationConfig.thinkingConfig").Exists() && util.ModelSupportsThinking(modelName) {
+		if t := gjson.GetBytes(rawJSON, "thinking"); 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())
+					out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
+					out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+				}
+			}
+		}
 	}
 
-	// Temperature/top_p/top_k
+	// Temperature/top_p/top_k/max_tokens
 	if tr := gjson.GetBytes(rawJSON, "temperature"); tr.Exists() && tr.Type == gjson.Number {
 		out, _ = sjson.SetBytes(out, "request.generationConfig.temperature", tr.Num)
 	}
@@ -107,6 +112,9 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 	if tkr := gjson.GetBytes(rawJSON, "top_k"); tkr.Exists() && tkr.Type == gjson.Number {
 		out, _ = sjson.SetBytes(out, "request.generationConfig.topK", tkr.Num)
 	}
+	if maxTok := gjson.GetBytes(rawJSON, "max_tokens"); maxTok.Exists() && maxTok.Type == gjson.Number {
+		out, _ = sjson.SetBytes(out, "request.generationConfig.maxOutputTokens", maxTok.Num)
+	}
 
 	// Map OpenAI modalities -> Gemini CLI request.generationConfig.responseModalities
 	// e.g. "modalities": ["image", "text"] -> ["IMAGE", "TEXT"]

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

@@ -11,6 +11,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/openai/chat-completions"
@@ -24,6 +25,9 @@ type convertCliResponseToOpenAIChatParams struct {
 	FunctionIndex int
 }
 
+// functionCallIDCounter provides a process-wide unique counter for function call identifiers.
+var functionCallIDCounter uint64
+
 // ConvertAntigravityResponseToOpenAI translates a single chunk of a streaming response from the
 // Gemini CLI API format to the OpenAI Chat Completions streaming format.
 // It processes various Gemini CLI event types and transforms them into OpenAI-compatible JSON responses.
@@ -146,7 +150,7 @@ func ConvertAntigravityResponseToOpenAI(_ context.Context, _ string, originalReq
 
 				functionCallTemplate := `{"id": "","index": 0,"type": "function","function": {"name": "","arguments": ""}}`
 				fcName := functionCallResult.Get("name").String()
-				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				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() {

internal/translator/claude/gemini/claude_gemini_response.go

@@ -331,8 +331,8 @@ func ConvertClaudeResponseToGeminiNonStream(_ context.Context, modelName string,
 	streamingEvents := make([][]byte, 0)
 
 	scanner := bufio.NewScanner(bytes.NewReader(rawJSON))
-	buffer := make([]byte, 20_971_520)
-	scanner.Buffer(buffer, 20_971_520)
+	buffer := make([]byte, 52_428_800) // 50MB
+	scanner.Buffer(buffer, 52_428_800)
 	for scanner.Scan() {
 		line := scanner.Bytes()
 		// log.Debug(string(line))

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

@@ -445,8 +445,8 @@ func ConvertClaudeResponseToOpenAIResponsesNonStream(_ context.Context, _ string
 		// Use a simple scanner to iterate through raw bytes
 		// Note: extremely large responses may require increasing the buffer
 		scanner := bufio.NewScanner(bytes.NewReader(rawJSON))
-		buf := make([]byte, 20_971_520)
-		scanner.Buffer(buf, 20_971_520)
+		buf := make([]byte, 52_428_800) // 50MB
+		scanner.Buffer(buf, 52_428_800)
 		for scanner.Scan() {
 			line := scanner.Bytes()
 			if !bytes.HasPrefix(line, dataTag) {

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

@@ -165,7 +165,6 @@ func ConvertClaudeRequestToCLI(modelName string, inputRawJSON []byte, _ bool) []
 		if t.Get("type").String() == "enabled" {
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())
-				budget = util.NormalizeThinkingBudget(modelName, budget)
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 			}

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

@@ -12,6 +12,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"github.com/tidwall/gjson"
@@ -25,8 +26,12 @@ type Params struct {
 	HasFirstResponse bool // Indicates if the initial message_start event has been sent
 	ResponseType     int  // Current response type: 0=none, 1=content, 2=thinking, 3=function
 	ResponseIndex    int  // Index counter for content blocks in the streaming response
+	HasContent       bool // Tracks whether any content (text, thinking, or tool use) has been output
 }
 
+// toolUseIDCounter provides a process-wide unique counter for tool use identifiers.
+var toolUseIDCounter uint64
+
 // ConvertGeminiCLIResponseToClaude performs sophisticated streaming response format conversion.
 // This function implements a complex state machine that translates backend client responses
 // into Claude Code-compatible Server-Sent Events (SSE) format. It manages different response types
@@ -53,9 +58,13 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 	}
 
 	if bytes.Equal(rawJSON, []byte("[DONE]")) {
-		return []string{
-			"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+		// Only send message_stop if we have actually output content
+		if (*param).(*Params).HasContent {
+			return []string{
+				"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+			}
 		}
+		return []string{}
 	}
 
 	// Track whether tools are being used in this response chunk
@@ -104,6 +113,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 						output = output + "event: content_block_delta\n"
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, (*param).(*Params).ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						(*param).(*Params).HasContent = true
 					} else {
 						// Transition from another state to thinking
 						// First, close any existing content block
@@ -127,6 +137,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, (*param).(*Params).ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						(*param).(*Params).ResponseType = 2 // Set state to thinking
+						(*param).(*Params).HasContent = true
 					}
 				} else {
 					// Process regular text content (user-visible output)
@@ -135,6 +146,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 						output = output + "event: content_block_delta\n"
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, (*param).(*Params).ResponseIndex), "delta.text", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						(*param).(*Params).HasContent = true
 					} else {
 						// Transition from another state to text content
 						// First, close any existing content block
@@ -158,6 +170,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, (*param).(*Params).ResponseIndex), "delta.text", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						(*param).(*Params).ResponseType = 1 // Set state to content
+						(*param).(*Params).HasContent = true
 					}
 				}
 			} else if functionCallResult.Exists() {
@@ -197,7 +210,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 
 				// Create the tool use block with unique ID and function details
 				data := fmt.Sprintf(`{"type":"content_block_start","index":%d,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`, (*param).(*Params).ResponseIndex)
-				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&toolUseIDCounter, 1)))
 				data, _ = sjson.Set(data, "content_block.name", fcName)
 				output = output + fmt.Sprintf("data: %s\n\n\n", data)
 
@@ -207,6 +220,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 					output = output + fmt.Sprintf("data: %s\n\n\n", data)
 				}
 				(*param).(*Params).ResponseType = 3
+				(*param).(*Params).HasContent = true
 			}
 		}
 	}
@@ -215,28 +229,31 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 	// Process usage metadata and finish reason when present in the response
 	if usageResult.Exists() && bytes.Contains(rawJSON, []byte(`"finishReason"`)) {
 		if candidatesTokenCountResult := usageResult.Get("candidatesTokenCount"); candidatesTokenCountResult.Exists() {
-			// Close the final content block
-			output = output + "event: content_block_stop\n"
-			output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, (*param).(*Params).ResponseIndex)
-			output = output + "\n\n\n"
+			// Only send final events if we have actually output content
+			if (*param).(*Params).HasContent {
+				// Close the final content block
+				output = output + "event: content_block_stop\n"
+				output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, (*param).(*Params).ResponseIndex)
+				output = output + "\n\n\n"
 
-			// Send the final message delta with usage information and stop reason
-			output = output + "event: message_delta\n"
-			output = output + `data: `
+				// Send the final message delta with usage information and stop reason
+				output = output + "event: message_delta\n"
+				output = output + `data: `
 
-			// Create the message delta template with appropriate stop reason
-			template := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
-			// Set tool_use stop reason if tools were used in this response
-			if usedTool {
-				template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				// Create the message delta template with appropriate stop reason
+				template := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				// Set tool_use stop reason if tools were used in this response
+				if usedTool {
+					template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				}
+
+				// Include thinking tokens in output token count if present
+				thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
+				template, _ = sjson.Set(template, "usage.output_tokens", candidatesTokenCountResult.Int()+thoughtsTokenCount)
+				template, _ = sjson.Set(template, "usage.input_tokens", usageResult.Get("promptTokenCount").Int())
+
+				output = output + template + "\n\n\n"
 			}
-
-			// Include thinking tokens in output token count if present
-			thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
-			template, _ = sjson.Set(template, "usage.output_tokens", candidatesTokenCountResult.Int()+thoughtsTokenCount)
-			template, _ = sjson.Set(template, "usage.input_tokens", usageResult.Get("promptTokenCount").Int())
-
-			output = output + template + "\n\n\n"
 		}
 	}
 

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

@@ -48,13 +48,13 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		case "low":
-			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 1024))
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 1024)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		case "medium":
-			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 8192))
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 8192)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		case "high":
-			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 32768))
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 32768)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		default:
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
@@ -66,15 +66,15 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 	if !hasOfficialThinking && util.ModelSupportsThinking(modelName) {
 		if tc := gjson.GetBytes(rawJSON, "extra_body.google.thinking_config"); tc.Exists() && tc.IsObject() {
 			var setBudget bool
-			var normalized int
+			var budget int
 
 			if v := tc.Get("thinkingBudget"); v.Exists() {
-				normalized = util.NormalizeThinkingBudget(modelName, int(v.Int()))
-				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", normalized)
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				setBudget = true
 			} else if v := tc.Get("thinking_budget"); v.Exists() {
-				normalized = util.NormalizeThinkingBudget(modelName, int(v.Int()))
-				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", normalized)
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				setBudget = true
 			}
 
@@ -82,21 +82,12 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", v.Bool())
 			} else if v := tc.Get("include_thoughts"); v.Exists() {
 				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", v.Bool())
-			} else if setBudget && normalized != 0 {
+			} else if setBudget && budget != 0 {
 				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 			}
 		}
 	}
 
-	// For gemini-3-pro-preview, always send default thinkingConfig when none specified.
-	// This matches the official Gemini CLI behavior which always sends:
-	// { thinkingBudget: -1, includeThoughts: true }
-	// See: ai-gemini-cli/packages/core/src/config/defaultModelConfigs.ts
-	if !gjson.GetBytes(out, "request.generationConfig.thinkingConfig").Exists() && modelName == "gemini-3-pro-preview" {
-		out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
-		out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
-	}
-
 	// Temperature/top_p/top_k
 	if tr := gjson.GetBytes(rawJSON, "temperature"); tr.Exists() && tr.Type == gjson.Number {
 		out, _ = sjson.SetBytes(out, "request.generationConfig.temperature", tr.Num)

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

@@ -11,6 +11,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/openai/chat-completions"
@@ -24,6 +25,9 @@ type convertCliResponseToOpenAIChatParams struct {
 	FunctionIndex int
 }
 
+// functionCallIDCounter provides a process-wide unique counter for function call identifiers.
+var functionCallIDCounter uint64
+
 // ConvertCliResponseToOpenAI translates a single chunk of a streaming response from the
 // Gemini CLI API format to the OpenAI Chat Completions streaming format.
 // It processes various Gemini CLI event types and transforms them into OpenAI-compatible JSON responses.
@@ -146,7 +150,7 @@ func ConvertCliResponseToOpenAI(_ context.Context, _ string, originalRequestRawJ
 
 				functionCallTemplate := `{"id": "","index": 0,"type": "function","function": {"name": "","arguments": ""}}`
 				fcName := functionCallResult.Get("name").String()
-				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				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() {

internal/translator/gemini/claude/gemini_claude_request.go

@@ -158,7 +158,6 @@ func ConvertClaudeRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 		if t.Get("type").String() == "enabled" {
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())
-				budget = util.NormalizeThinkingBudget(modelName, budget)
 				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", budget)
 				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
 			}

internal/translator/gemini/claude/gemini_claude_response.go

@@ -12,6 +12,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"github.com/tidwall/gjson"
@@ -24,8 +25,12 @@ type Params struct {
 	HasFirstResponse bool
 	ResponseType     int
 	ResponseIndex    int
+	HasContent       bool // Tracks whether any content (text, thinking, or tool use) has been output
 }
 
+// toolUseIDCounter provides a process-wide unique counter for tool use identifiers.
+var toolUseIDCounter uint64
+
 // ConvertGeminiResponseToClaude performs sophisticated streaming response format conversion.
 // This function implements a complex state machine that translates backend client responses
 // into Claude-compatible Server-Sent Events (SSE) format. It manages different response types
@@ -53,9 +58,13 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 	}
 
 	if bytes.Equal(rawJSON, []byte("[DONE]")) {
-		return []string{
-			"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+		// Only send message_stop if we have actually output content
+		if (*param).(*Params).HasContent {
+			return []string{
+				"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+			}
 		}
+		return []string{}
 	}
 
 	// Track whether tools are being used in this response chunk
@@ -104,6 +113,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 						output = output + "event: content_block_delta\n"
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, (*param).(*Params).ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						(*param).(*Params).HasContent = true
 					} else {
 						// Transition from another state to thinking
 						// First, close any existing content block
@@ -127,6 +137,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, (*param).(*Params).ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						(*param).(*Params).ResponseType = 2 // Set state to thinking
+						(*param).(*Params).HasContent = true
 					}
 				} else {
 					// Process regular text content (user-visible output)
@@ -135,6 +146,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 						output = output + "event: content_block_delta\n"
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, (*param).(*Params).ResponseIndex), "delta.text", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						(*param).(*Params).HasContent = true
 					} else {
 						// Transition from another state to text content
 						// First, close any existing content block
@@ -158,6 +170,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, (*param).(*Params).ResponseIndex), "delta.text", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						(*param).(*Params).ResponseType = 1 // Set state to content
+						(*param).(*Params).HasContent = true
 					}
 				}
 			} else if functionCallResult.Exists() {
@@ -197,7 +210,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 
 				// Create the tool use block with unique ID and function details
 				data := fmt.Sprintf(`{"type":"content_block_start","index":%d,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`, (*param).(*Params).ResponseIndex)
-				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&toolUseIDCounter, 1)))
 				data, _ = sjson.Set(data, "content_block.name", fcName)
 				output = output + fmt.Sprintf("data: %s\n\n\n", data)
 
@@ -207,6 +220,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 					output = output + fmt.Sprintf("data: %s\n\n\n", data)
 				}
 				(*param).(*Params).ResponseType = 3
+				(*param).(*Params).HasContent = true
 			}
 		}
 	}
@@ -214,23 +228,26 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 	usageResult := gjson.GetBytes(rawJSON, "usageMetadata")
 	if usageResult.Exists() && bytes.Contains(rawJSON, []byte(`"finishReason"`)) {
 		if candidatesTokenCountResult := usageResult.Get("candidatesTokenCount"); candidatesTokenCountResult.Exists() {
-			output = output + "event: content_block_stop\n"
-			output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, (*param).(*Params).ResponseIndex)
-			output = output + "\n\n\n"
+			// Only send final events if we have actually output content
+			if (*param).(*Params).HasContent {
+				output = output + "event: content_block_stop\n"
+				output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, (*param).(*Params).ResponseIndex)
+				output = output + "\n\n\n"
 
-			output = output + "event: message_delta\n"
-			output = output + `data: `
+				output = output + "event: message_delta\n"
+				output = output + `data: `
 
-			template := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
-			if usedTool {
-				template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				template := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				if usedTool {
+					template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				}
+
+				thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
+				template, _ = sjson.Set(template, "usage.output_tokens", candidatesTokenCountResult.Int()+thoughtsTokenCount)
+				template, _ = sjson.Set(template, "usage.input_tokens", usageResult.Get("promptTokenCount").Int())
+
+				output = output + template + "\n\n\n"
 			}
-
-			thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
-			template, _ = sjson.Set(template, "usage.output_tokens", candidatesTokenCountResult.Int()+thoughtsTokenCount)
-			template, _ = sjson.Set(template, "usage.input_tokens", usageResult.Get("promptTokenCount").Int())
-
-			output = output + template + "\n\n\n"
 		}
 	}
 

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

@@ -48,13 +48,13 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", -1)
 			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.include_thoughts", true)
 		case "low":
-			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 1024))
+			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", 1024)
 			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.include_thoughts", true)
 		case "medium":
-			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 8192))
+			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", 8192)
 			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.include_thoughts", true)
 		case "high":
-			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 32768))
+			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", 32768)
 			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.include_thoughts", true)
 		default:
 			out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", -1)
@@ -66,15 +66,15 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 	if !hasOfficialThinking && util.ModelSupportsThinking(modelName) {
 		if tc := gjson.GetBytes(rawJSON, "extra_body.google.thinking_config"); tc.Exists() && tc.IsObject() {
 			var setBudget bool
-			var normalized int
+			var budget int
 
 			if v := tc.Get("thinkingBudget"); v.Exists() {
-				normalized = util.NormalizeThinkingBudget(modelName, int(v.Int()))
-				out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", normalized)
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", budget)
 				setBudget = true
 			} else if v := tc.Get("thinking_budget"); v.Exists() {
-				normalized = util.NormalizeThinkingBudget(modelName, int(v.Int()))
-				out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", normalized)
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.thinkingBudget", budget)
 				setBudget = true
 			}
 
@@ -82,7 +82,7 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 				out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.include_thoughts", v.Bool())
 			} else if v := tc.Get("include_thoughts"); v.Exists() {
 				out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.include_thoughts", v.Bool())
-			} else if setBudget && normalized != 0 {
+			} else if setBudget && budget != 0 {
 				out, _ = sjson.SetBytes(out, "generationConfig.thinkingConfig.include_thoughts", true)
 			}
 		}

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

@@ -11,6 +11,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"github.com/tidwall/gjson"
@@ -23,6 +24,9 @@ type convertGeminiResponseToOpenAIChatParams struct {
 	FunctionIndex int
 }
 
+// functionCallIDCounter provides a process-wide unique counter for function call identifiers.
+var functionCallIDCounter uint64
+
 // ConvertGeminiResponseToOpenAI translates a single chunk of a streaming response from the
 // Gemini API format to the OpenAI Chat Completions streaming format.
 // It processes various Gemini event types and transforms them into OpenAI-compatible JSON responses.
@@ -148,7 +152,7 @@ func ConvertGeminiResponseToOpenAI(_ context.Context, _ string, originalRequestR
 
 				functionCallTemplate := `{"id": "","index": 0,"type": "function","function": {"name": "","arguments": ""}}`
 				fcName := functionCallResult.Get("name").String()
-				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				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() {
@@ -281,7 +285,7 @@ func ConvertGeminiResponseToOpenAINonStream(_ context.Context, _ string, origina
 				}
 				functionCallItemTemplate := `{"id": "","type": "function","function": {"name": "","arguments": ""}}`
 				fcName := functionCallResult.Get("name").String()
-				functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				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)

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

@@ -400,16 +400,16 @@ func ConvertOpenAIResponsesRequestToGemini(modelName string, inputRawJSON []byte
 			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", -1)
 			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
 		case "minimal":
-			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 1024))
+			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", 1024)
 			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
 		case "low":
-			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 4096))
+			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", 4096)
 			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
 		case "medium":
-			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 8192))
+			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", 8192)
 			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
 		case "high":
-			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 32768))
+			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", 32768)
 			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
 		default:
 			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", -1)
@@ -421,32 +421,22 @@ func ConvertOpenAIResponsesRequestToGemini(modelName string, inputRawJSON []byte
 	if !hasOfficialThinking && util.ModelSupportsThinking(modelName) {
 		if tc := root.Get("extra_body.google.thinking_config"); tc.Exists() && tc.IsObject() {
 			var setBudget bool
-			var normalized int
+			var budget int
 			if v := tc.Get("thinking_budget"); v.Exists() {
-				normalized = util.NormalizeThinkingBudget(modelName, int(v.Int()))
-				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", normalized)
+				budget = int(v.Int())
+				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", budget)
 				setBudget = true
 			}
 			if v := tc.Get("include_thoughts"); v.Exists() {
 				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", v.Bool())
 			} else if setBudget {
-				if normalized != 0 {
+				if budget != 0 {
 					out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
 				}
 			}
 		}
 	}
 
-	// For gemini-3-pro-preview, always send default thinkingConfig when none specified.
-	// This matches the official Gemini CLI behavior which always sends:
-	// { thinkingBudget: -1, includeThoughts: true }
-	// See: ai-gemini-cli/packages/core/src/config/defaultModelConfigs.ts
-	if !gjson.Get(out, "generationConfig.thinkingConfig").Exists() && modelName == "gemini-3-pro-preview" {
-		out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", -1)
-		out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
-		// log.Debugf("Applied default thinkingConfig for gemini-3-pro-preview (matches Gemini CLI): thinkingBudget=-1, include_thoughts=true")
-	}
-
 	result := []byte(out)
 	result = common.AttachDefaultSafetySettings(result, "safetySettings")
 	return result

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

@@ -5,6 +5,7 @@ import (
 	"context"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"github.com/tidwall/gjson"
@@ -37,6 +38,12 @@ type geminiToResponsesState struct {
 	FuncCallIDs map[int]string
 }
 
+// responseIDCounter provides a process-wide unique counter for synthesized response identifiers.
+var responseIDCounter uint64
+
+// funcCallIDCounter provides a process-wide unique counter for function call identifiers.
+var funcCallIDCounter uint64
+
 func emitEvent(event string, payload string) string {
 	return fmt.Sprintf("event: %s\ndata: %s", event, payload)
 }
@@ -205,7 +212,7 @@ func ConvertGeminiResponseToOpenAIResponses(_ context.Context, modelName string,
 					st.FuncArgsBuf[idx] = &strings.Builder{}
 				}
 				if st.FuncCallIDs[idx] == "" {
-					st.FuncCallIDs[idx] = fmt.Sprintf("call_%d", time.Now().UnixNano())
+					st.FuncCallIDs[idx] = fmt.Sprintf("call_%d_%d", time.Now().UnixNano(), atomic.AddUint64(&funcCallIDCounter, 1))
 				}
 				st.FuncNames[idx] = name
 
@@ -464,7 +471,7 @@ func ConvertGeminiResponseToOpenAIResponsesNonStream(_ context.Context, _ string
 	// id: prefer provider responseId, otherwise synthesize
 	id := root.Get("responseId").String()
 	if id == "" {
-		id = fmt.Sprintf("resp_%x", time.Now().UnixNano())
+		id = fmt.Sprintf("resp_%x_%d", time.Now().UnixNano(), atomic.AddUint64(&responseIDCounter, 1))
 	}
 	// Normalize to response-style id (prefix resp_ if missing)
 	if !strings.HasPrefix(id, "resp_") {
@@ -575,7 +582,7 @@ func ConvertGeminiResponseToOpenAIResponsesNonStream(_ context.Context, _ string
 			if fc := p.Get("functionCall"); fc.Exists() {
 				name := fc.Get("name").String()
 				args := fc.Get("args")
-				callID := fmt.Sprintf("call_%x", time.Now().UnixNano())
+				callID := fmt.Sprintf("call_%x_%d", time.Now().UnixNano(), atomic.AddUint64(&funcCallIDCounter, 1))
 				outputs = append(outputs, map[string]interface{}{
 					"id":     fmt.Sprintf("fc_%s", callID),
 					"type":   "function_call",

internal/translator/openai/claude/openai_claude_request.go

@@ -8,6 +8,7 @@ package claude
 import (
 	"bytes"
 	"encoding/json"
+	"strings"
 
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -242,11 +243,12 @@ func convertClaudeContentPart(part gjson.Result) (string, bool) {
 
 	switch partType {
 	case "text":
-		if !part.Get("text").Exists() {
+		text := part.Get("text").String()
+		if strings.TrimSpace(text) == "" {
 			return "", false
 		}
 		textContent := `{"type":"text","text":""}`
-		textContent, _ = sjson.Set(textContent, "text", part.Get("text").String())
+		textContent, _ = sjson.Set(textContent, "text", text)
 		return textContent, true
 
 	case "image":

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

@@ -5,6 +5,7 @@ import (
 	"context"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"github.com/tidwall/gjson"
@@ -41,6 +42,9 @@ type oaiToResponsesState struct {
 	UsageSeen        bool
 }
 
+// responseIDCounter provides a process-wide unique counter for synthesized response identifiers.
+var responseIDCounter uint64
+
 func emitRespEvent(event string, payload string) string {
 	return fmt.Sprintf("event: %s\ndata: %s", event, payload)
 }
@@ -590,7 +594,7 @@ func ConvertOpenAIChatCompletionsResponseToOpenAIResponsesNonStream(_ context.Co
 	// id: use provider id if present, otherwise synthesize
 	id := root.Get("id").String()
 	if id == "" {
-		id = fmt.Sprintf("resp_%x", time.Now().UnixNano())
+		id = fmt.Sprintf("resp_%x_%d", time.Now().UnixNano(), atomic.AddUint64(&responseIDCounter, 1))
 	}
 	resp, _ = sjson.Set(resp, "id", id)
 

internal/util/claude_thinking.go

@@ -0,0 +1,46 @@
+package util
+
+import (
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// ApplyClaudeThinkingConfig applies thinking configuration to a Claude API request payload.
+// It sets the thinking.type to "enabled" and thinking.budget_tokens to the specified budget.
+// If budget is nil or the payload already has thinking config, it returns the payload unchanged.
+func ApplyClaudeThinkingConfig(body []byte, budget *int) []byte {
+	if budget == nil {
+		return body
+	}
+	if gjson.GetBytes(body, "thinking").Exists() {
+		return body
+	}
+	if *budget <= 0 {
+		return body
+	}
+	updated := body
+	updated, _ = sjson.SetBytes(updated, "thinking.type", "enabled")
+	updated, _ = sjson.SetBytes(updated, "thinking.budget_tokens", *budget)
+	return updated
+}
+
+// ResolveClaudeThinkingConfig resolves thinking configuration from metadata for Claude models.
+// It uses the unified ResolveThinkingConfigFromMetadata and normalizes the budget.
+// Returns the normalized budget (nil if thinking should not be enabled) and whether it matched.
+func ResolveClaudeThinkingConfig(modelName string, metadata map[string]any) (*int, bool) {
+	budget, include, matched := ResolveThinkingConfigFromMetadata(modelName, metadata)
+	if !matched {
+		return nil, false
+	}
+	if include != nil && !*include {
+		return nil, true
+	}
+	if budget == nil {
+		return nil, true
+	}
+	normalized := NormalizeThinkingBudget(modelName, *budget)
+	if normalized <= 0 {
+		return nil, true
+	}
+	return &normalized, true
+}

internal/util/gemini_thinking.go

@@ -1,8 +1,6 @@
 package util
 
 import (
-	"encoding/json"
-	"strconv"
 	"strings"
 
 	"github.com/tidwall/gjson"
@@ -15,80 +13,6 @@ const (
 	GeminiOriginalModelMetadataKey   = "gemini_original_model"
 )
 
-func ParseGeminiThinkingSuffix(model string) (string, *int, *bool, bool) {
-	if model == "" {
-		return model, nil, nil, false
-	}
-	lower := strings.ToLower(model)
-	if !strings.HasPrefix(lower, "gemini-") {
-		return model, nil, nil, false
-	}
-
-	if strings.HasSuffix(lower, "-nothinking") {
-		base := model[:len(model)-len("-nothinking")]
-		budgetValue := 0
-		if strings.HasPrefix(lower, "gemini-2.5-pro") {
-			budgetValue = 128
-		}
-		include := false
-		return base, &budgetValue, &include, true
-	}
-
-	// Handle "-reasoning" suffix: enables thinking with dynamic budget (-1)
-	// Maps: gemini-2.5-flash-reasoning -> gemini-2.5-flash with thinkingBudget=-1
-	if strings.HasSuffix(lower, "-reasoning") {
-		base := model[:len(model)-len("-reasoning")]
-		budgetValue := -1 // Dynamic budget
-		include := true
-		return base, &budgetValue, &include, true
-	}
-
-	idx := strings.LastIndex(lower, "-thinking-")
-	if idx == -1 {
-		return model, nil, nil, false
-	}
-
-	digits := model[idx+len("-thinking-"):]
-	if digits == "" {
-		return model, nil, nil, false
-	}
-	end := len(digits)
-	for i := 0; i < len(digits); i++ {
-		if digits[i] < '0' || digits[i] > '9' {
-			end = i
-			break
-		}
-	}
-	if end == 0 {
-		return model, nil, nil, false
-	}
-	valueStr := digits[:end]
-	value, err := strconv.Atoi(valueStr)
-	if err != nil {
-		return model, nil, nil, false
-	}
-	base := model[:idx]
-	budgetValue := value
-	return base, &budgetValue, nil, true
-}
-
-func NormalizeGeminiThinkingModel(modelName string) (string, map[string]any) {
-	baseModel, budget, include, matched := ParseGeminiThinkingSuffix(modelName)
-	if !matched {
-		return baseModel, nil
-	}
-	metadata := map[string]any{
-		GeminiOriginalModelMetadataKey: modelName,
-	}
-	if budget != nil {
-		metadata[GeminiThinkingBudgetMetadataKey] = *budget
-	}
-	if include != nil {
-		metadata[GeminiIncludeThoughtsMetadataKey] = *include
-	}
-	return baseModel, metadata
-}
-
 func ApplyGeminiThinkingConfig(body []byte, budget *int, includeThoughts *bool) []byte {
 	if budget == nil && includeThoughts == nil {
 		return body
@@ -133,78 +57,45 @@ func ApplyGeminiCLIThinkingConfig(body []byte, budget *int, includeThoughts *boo
 	return updated
 }
 
-func GeminiThinkingFromMetadata(metadata map[string]any) (*int, *bool, bool) {
-	if len(metadata) == 0 {
-		return nil, nil, false
+// modelsWithDefaultThinking lists models that should have thinking enabled by default
+// when no explicit thinkingConfig is provided.
+var modelsWithDefaultThinking = map[string]bool{
+	"gemini-3-pro-preview": true,
+}
+
+// ModelHasDefaultThinking returns true if the model should have thinking enabled by default.
+func ModelHasDefaultThinking(model string) bool {
+	return modelsWithDefaultThinking[model]
+}
+
+// ApplyDefaultThinkingIfNeeded injects default thinkingConfig for models that require it.
+// For standard Gemini API format (generationConfig.thinkingConfig path).
+// Returns the modified body if thinkingConfig was added, otherwise returns the original.
+func ApplyDefaultThinkingIfNeeded(model string, body []byte) []byte {
+	if !ModelHasDefaultThinking(model) {
+		return body
 	}
-	var (
-		budgetPtr  *int
-		includePtr *bool
-		matched    bool
-	)
-	if rawBudget, ok := metadata[GeminiThinkingBudgetMetadataKey]; ok {
-		switch v := rawBudget.(type) {
-		case int:
-			budget := v
-			budgetPtr = &budget
-			matched = true
-		case int32:
-			budget := int(v)
-			budgetPtr = &budget
-			matched = true
-		case int64:
-			budget := int(v)
-			budgetPtr = &budget
-			matched = true
-		case float64:
-			budget := int(v)
-			budgetPtr = &budget
-			matched = true
-		case json.Number:
-			if val, err := v.Int64(); err == nil {
-				budget := int(val)
-				budgetPtr = &budget
-				matched = true
-			}
-		}
+	if gjson.GetBytes(body, "generationConfig.thinkingConfig").Exists() {
+		return body
 	}
-	if rawInclude, ok := metadata[GeminiIncludeThoughtsMetadataKey]; ok {
-		switch v := rawInclude.(type) {
-		case bool:
-			include := v
-			includePtr = &include
-			matched = true
-		case string:
-			if parsed, err := strconv.ParseBool(v); err == nil {
-				include := parsed
-				includePtr = &include
-				matched = true
-			}
-		case json.Number:
-			if val, err := v.Int64(); err == nil {
-				include := val != 0
-				includePtr = &include
-				matched = true
-			}
-		case int:
-			include := v != 0
-			includePtr = &include
-			matched = true
-		case int32:
-			include := v != 0
-			includePtr = &include
-			matched = true
-		case int64:
-			include := v != 0
-			includePtr = &include
-			matched = true
-		case float64:
-			include := v != 0
-			includePtr = &include
-			matched = true
-		}
+	updated, _ := sjson.SetBytes(body, "generationConfig.thinkingConfig.thinkingBudget", -1)
+	updated, _ = sjson.SetBytes(updated, "generationConfig.thinkingConfig.include_thoughts", true)
+	return updated
+}
+
+// ApplyDefaultThinkingIfNeededCLI injects default thinkingConfig for models that require it.
+// For Gemini CLI API format (request.generationConfig.thinkingConfig path).
+// Returns the modified body if thinkingConfig was added, otherwise returns the original.
+func ApplyDefaultThinkingIfNeededCLI(model string, body []byte) []byte {
+	if !ModelHasDefaultThinking(model) {
+		return body
 	}
-	return budgetPtr, includePtr, matched
+	if gjson.GetBytes(body, "request.generationConfig.thinkingConfig").Exists() {
+		return body
+	}
+	updated, _ := sjson.SetBytes(body, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
+	updated, _ = sjson.SetBytes(updated, "request.generationConfig.thinkingConfig.include_thoughts", true)
+	return updated
 }
 
 // StripThinkingConfigIfUnsupported removes thinkingConfig from the request body
@@ -223,6 +114,32 @@ func StripThinkingConfigIfUnsupported(model string, body []byte) []byte {
 	return updated
 }
 
+// NormalizeGeminiThinkingBudget normalizes the thinkingBudget value in a standard Gemini
+// request body (generationConfig.thinkingConfig.thinkingBudget path).
+func NormalizeGeminiThinkingBudget(model string, body []byte) []byte {
+	const budgetPath = "generationConfig.thinkingConfig.thinkingBudget"
+	budget := gjson.GetBytes(body, budgetPath)
+	if !budget.Exists() {
+		return body
+	}
+	normalized := NormalizeThinkingBudget(model, int(budget.Int()))
+	updated, _ := sjson.SetBytes(body, budgetPath, normalized)
+	return updated
+}
+
+// NormalizeGeminiCLIThinkingBudget normalizes the thinkingBudget value in a Gemini CLI
+// request body (request.generationConfig.thinkingConfig.thinkingBudget path).
+func NormalizeGeminiCLIThinkingBudget(model string, body []byte) []byte {
+	const budgetPath = "request.generationConfig.thinkingConfig.thinkingBudget"
+	budget := gjson.GetBytes(body, budgetPath)
+	if !budget.Exists() {
+		return body
+	}
+	normalized := NormalizeThinkingBudget(model, int(budget.Int()))
+	updated, _ := sjson.SetBytes(body, budgetPath, normalized)
+	return updated
+}
+
 // ConvertThinkingLevelToBudget checks for "generationConfig.thinkingConfig.thinkingLevel"
 // and converts it to "thinkingBudget".
 // "high" -> 32768

internal/util/thinking.go

@@ -1,6 +1,8 @@
 package util
 
 import (
+	"strings"
+
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 )
 
@@ -67,3 +69,39 @@ func thinkingRangeFromRegistry(model string) (found bool, min int, max int, zero
 	}
 	return true, info.Thinking.Min, info.Thinking.Max, info.Thinking.ZeroAllowed, info.Thinking.DynamicAllowed
 }
+
+// GetModelThinkingLevels returns the discrete reasoning effort levels for the model.
+// Returns nil if the model has no thinking support or no levels defined.
+func GetModelThinkingLevels(model string) []string {
+	if model == "" {
+		return nil
+	}
+	info := registry.GetGlobalRegistry().GetModelInfo(model)
+	if info == nil || info.Thinking == nil {
+		return nil
+	}
+	return info.Thinking.Levels
+}
+
+// ModelUsesThinkingLevels reports whether the model uses discrete reasoning
+// effort levels instead of numeric budgets.
+func ModelUsesThinkingLevels(model string) bool {
+	levels := GetModelThinkingLevels(model)
+	return len(levels) > 0
+}
+
+// NormalizeReasoningEffortLevel validates and normalizes a reasoning effort
+// level for the given model. Returns false when the level is not supported.
+func NormalizeReasoningEffortLevel(model, effort string) (string, bool) {
+	levels := GetModelThinkingLevels(model)
+	if len(levels) == 0 {
+		return "", false
+	}
+	loweredEffort := strings.ToLower(strings.TrimSpace(effort))
+	for _, lvl := range levels {
+		if strings.ToLower(lvl) == loweredEffort {
+			return lvl, true
+		}
+	}
+	return "", false
+}

internal/util/thinking_suffix.go

@@ -0,0 +1,313 @@
+package util
+
+import (
+	"encoding/json"
+	"strconv"
+	"strings"
+)
+
+const (
+	ThinkingBudgetMetadataKey          = "thinking_budget"
+	ThinkingIncludeThoughtsMetadataKey = "thinking_include_thoughts"
+	ReasoningEffortMetadataKey         = "reasoning_effort"
+	ThinkingOriginalModelMetadataKey   = "thinking_original_model"
+)
+
+// NormalizeThinkingModel parses dynamic thinking suffixes on model names and returns
+// the normalized base model with extracted metadata. Supported pattern:
+//   - "(<value>)" where value can be:
+//   - A numeric budget (e.g., "(8192)", "(16384)")
+//   - A reasoning effort level (e.g., "(high)", "(medium)", "(low)")
+//
+// Examples:
+//   - "claude-sonnet-4-5-20250929(16384)" → budget=16384
+//   - "gpt-5.1(high)" → reasoning_effort="high"
+//   - "gemini-2.5-pro(32768)" → budget=32768
+//
+// Note: Empty parentheses "()" are not supported and will be ignored.
+func NormalizeThinkingModel(modelName string) (string, map[string]any) {
+	if modelName == "" {
+		return modelName, nil
+	}
+
+	baseModel := modelName
+
+	var (
+		budgetOverride  *int
+		reasoningEffort *string
+		matched         bool
+	)
+
+	// Match "(<value>)" pattern at the end of the model name
+	if idx := strings.LastIndex(modelName, "("); idx != -1 {
+		if !strings.HasSuffix(modelName, ")") {
+			// Incomplete parenthesis, ignore
+			return baseModel, nil
+		}
+
+		value := modelName[idx+1 : len(modelName)-1] // Extract content between ( and )
+		if value == "" {
+			// Empty parentheses not supported
+			return baseModel, nil
+		}
+
+		candidateBase := modelName[:idx]
+
+		// Auto-detect: pure numeric → budget, string → reasoning effort level
+		if parsed, ok := parseIntPrefix(value); ok {
+			// Numeric value: treat as thinking budget
+			baseModel = candidateBase
+			budgetOverride = &parsed
+			matched = true
+		} else {
+			// String value: treat as reasoning effort level
+			baseModel = candidateBase
+			raw := strings.ToLower(strings.TrimSpace(value))
+			if raw != "" {
+				reasoningEffort = &raw
+				matched = true
+			}
+		}
+	}
+
+	if !matched {
+		return baseModel, nil
+	}
+
+	metadata := map[string]any{
+		ThinkingOriginalModelMetadataKey: modelName,
+	}
+	if budgetOverride != nil {
+		metadata[ThinkingBudgetMetadataKey] = *budgetOverride
+	}
+	if reasoningEffort != nil {
+		metadata[ReasoningEffortMetadataKey] = *reasoningEffort
+	}
+	return baseModel, metadata
+}
+
+// ThinkingFromMetadata extracts thinking overrides from metadata produced by NormalizeThinkingModel.
+// It accepts both the new generic keys and legacy Gemini-specific keys.
+func ThinkingFromMetadata(metadata map[string]any) (*int, *bool, *string, bool) {
+	if len(metadata) == 0 {
+		return nil, nil, nil, false
+	}
+
+	var (
+		budgetPtr  *int
+		includePtr *bool
+		effortPtr  *string
+		matched    bool
+	)
+
+	readBudget := func(key string) {
+		if budgetPtr != nil {
+			return
+		}
+		if raw, ok := metadata[key]; ok {
+			if v, okNumber := parseNumberToInt(raw); okNumber {
+				budget := v
+				budgetPtr = &budget
+				matched = true
+			}
+		}
+	}
+
+	readInclude := func(key string) {
+		if includePtr != nil {
+			return
+		}
+		if raw, ok := metadata[key]; ok {
+			switch v := raw.(type) {
+			case bool:
+				val := v
+				includePtr = &val
+				matched = true
+			case *bool:
+				if v != nil {
+					val := *v
+					includePtr = &val
+					matched = true
+				}
+			}
+		}
+	}
+
+	readEffort := func(key string) {
+		if effortPtr != nil {
+			return
+		}
+		if raw, ok := metadata[key]; ok {
+			if val, okStr := raw.(string); okStr && strings.TrimSpace(val) != "" {
+				normalized := strings.ToLower(strings.TrimSpace(val))
+				effortPtr = &normalized
+				matched = true
+			}
+		}
+	}
+
+	readBudget(ThinkingBudgetMetadataKey)
+	readBudget(GeminiThinkingBudgetMetadataKey)
+	readInclude(ThinkingIncludeThoughtsMetadataKey)
+	readInclude(GeminiIncludeThoughtsMetadataKey)
+	readEffort(ReasoningEffortMetadataKey)
+	readEffort("reasoning.effort")
+
+	return budgetPtr, includePtr, effortPtr, matched
+}
+
+// ResolveThinkingConfigFromMetadata derives thinking budget/include overrides,
+// converting reasoning effort strings into budgets when possible.
+func ResolveThinkingConfigFromMetadata(model string, metadata map[string]any) (*int, *bool, bool) {
+	budget, include, effort, matched := ThinkingFromMetadata(metadata)
+	if !matched {
+		return nil, nil, false
+	}
+
+	if budget == nil && effort != nil {
+		if derived, ok := ThinkingEffortToBudget(model, *effort); ok {
+			budget = &derived
+		}
+	}
+	return budget, include, budget != nil || include != nil || effort != nil
+}
+
+// ReasoningEffortFromMetadata resolves a reasoning effort string from metadata,
+// inferring "auto" and "none" when budgets request dynamic or disabled thinking.
+func ReasoningEffortFromMetadata(metadata map[string]any) (string, bool) {
+	budget, include, effort, matched := ThinkingFromMetadata(metadata)
+	if !matched {
+		return "", false
+	}
+	if effort != nil && *effort != "" {
+		return strings.ToLower(strings.TrimSpace(*effort)), true
+	}
+	if budget != nil {
+		switch *budget {
+		case -1:
+			return "auto", true
+		case 0:
+			return "none", true
+		}
+	}
+	if include != nil && !*include {
+		return "none", true
+	}
+	return "", true
+}
+
+// ThinkingEffortToBudget maps reasoning effort levels to approximate budgets,
+// clamping the result to the model's supported range.
+func ThinkingEffortToBudget(model, effort string) (int, bool) {
+	if effort == "" {
+		return 0, false
+	}
+	normalized, ok := NormalizeReasoningEffortLevel(model, effort)
+	if !ok {
+		normalized = strings.ToLower(strings.TrimSpace(effort))
+	}
+	switch normalized {
+	case "none":
+		return 0, true
+	case "auto":
+		return NormalizeThinkingBudget(model, -1), true
+	case "minimal":
+		return NormalizeThinkingBudget(model, 512), true
+	case "low":
+		return NormalizeThinkingBudget(model, 1024), true
+	case "medium":
+		return NormalizeThinkingBudget(model, 8192), true
+	case "high":
+		return NormalizeThinkingBudget(model, 24576), true
+	case "xhigh":
+		return NormalizeThinkingBudget(model, 32768), true
+	default:
+		return 0, false
+	}
+}
+
+// ResolveOriginalModel returns the original model name stored in metadata (if present),
+// otherwise falls back to the provided model.
+func ResolveOriginalModel(model string, metadata map[string]any) string {
+	normalize := func(name string) string {
+		if name == "" {
+			return ""
+		}
+		if base, _ := NormalizeThinkingModel(name); base != "" {
+			return base
+		}
+		return strings.TrimSpace(name)
+	}
+
+	if metadata != nil {
+		if v, ok := metadata[ThinkingOriginalModelMetadataKey]; ok {
+			if s, okStr := v.(string); okStr && strings.TrimSpace(s) != "" {
+				if base := normalize(s); base != "" {
+					return base
+				}
+			}
+		}
+		if v, ok := metadata[GeminiOriginalModelMetadataKey]; ok {
+			if s, okStr := v.(string); okStr && strings.TrimSpace(s) != "" {
+				if base := normalize(s); base != "" {
+					return base
+				}
+			}
+		}
+	}
+	// Fallback: try to re-normalize the model name when metadata was dropped.
+	if base := normalize(model); base != "" {
+		return base
+	}
+	return model
+}
+
+func parseIntPrefix(value string) (int, bool) {
+	if value == "" {
+		return 0, false
+	}
+	digits := strings.TrimLeft(value, "-")
+	if digits == "" {
+		return 0, false
+	}
+	end := len(digits)
+	for i := 0; i < len(digits); i++ {
+		if digits[i] < '0' || digits[i] > '9' {
+			end = i
+			break
+		}
+	}
+	if end == 0 {
+		return 0, false
+	}
+	val, err := strconv.Atoi(digits[:end])
+	if err != nil {
+		return 0, false
+	}
+	return val, true
+}
+
+func parseNumberToInt(raw any) (int, bool) {
+	switch v := raw.(type) {
+	case int:
+		return v, true
+	case int32:
+		return int(v), true
+	case int64:
+		return int(v), true
+	case float64:
+		return int(v), true
+	case json.Number:
+		if val, err := v.Int64(); err == nil {
+			return int(val), true
+		}
+	case string:
+		if strings.TrimSpace(v) == "" {
+			return 0, false
+		}
+		if parsed, err := strconv.Atoi(strings.TrimSpace(v)); err == nil {
+			return parsed, true
+		}
+	}
+	return 0, false
+}

sdk/api/handlers/claude/code_handlers.go

@@ -271,6 +271,11 @@ func (h *ClaudeCodeAPIHandler) forwardClaudeStream(c *gin.Context, flusher http.
 				continue
 			}
 			if errMsg != nil {
+				status := http.StatusInternalServerError
+				if errMsg.StatusCode > 0 {
+					status = errMsg.StatusCode
+				}
+				c.Status(status)
 				// An error occurred: emit as a proper SSE error event
 				errorBytes, _ := json.Marshal(h.toClaudeError(errMsg))
 				_, _ = writer.WriteString("event: error\n")
@@ -278,6 +283,7 @@ func (h *ClaudeCodeAPIHandler) forwardClaudeStream(c *gin.Context, flusher http.
 				_, _ = writer.Write(errorBytes)
 				_, _ = writer.WriteString("\n\n")
 				_ = writer.Flush()
+				flusher.Flush()
 			}
 			var execErr error
 			if errMsg != nil {

sdk/api/handlers/gemini/gemini_handlers.go

@@ -48,8 +48,24 @@ func (h *GeminiAPIHandler) Models() []map[string]any {
 // GeminiModels handles the Gemini models listing endpoint.
 // It returns a JSON response containing available Gemini models and their specifications.
 func (h *GeminiAPIHandler) GeminiModels(c *gin.Context) {
+	rawModels := h.Models()
+	normalizedModels := make([]map[string]any, 0, len(rawModels))
+	defaultMethods := []string{"generateContent"}
+	for _, model := range rawModels {
+		normalizedModel := make(map[string]any, len(model))
+		for k, v := range model {
+			normalizedModel[k] = v
+		}
+		if name, ok := normalizedModel["name"].(string); ok && name != "" && !strings.HasPrefix(name, "models/") {
+			normalizedModel["name"] = "models/" + name
+		}
+		if _, ok := normalizedModel["supportedGenerationMethods"]; !ok {
+			normalizedModel["supportedGenerationMethods"] = defaultMethods
+		}
+		normalizedModels = append(normalizedModels, normalizedModel)
+	}
 	c.JSON(http.StatusOK, gin.H{
-		"models": h.Models(),
+		"models": normalizedModels,
 	})
 }
 

sdk/api/handlers/handlers.go

@@ -323,18 +323,32 @@ func (h *BaseAPIHandler) getRequestDetails(modelName string) (providers []string
 
 	providerName, extractedModelName, isDynamic := h.parseDynamicModel(resolvedModelName)
 
-	// First, normalize the model name to handle suffixes like "-thinking-128"
-	// This needs to happen before determining the provider for non-dynamic models.
-	normalizedModel, metadata = normalizeModelMetadata(resolvedModelName)
+	targetModelName := resolvedModelName
+	if isDynamic {
+		targetModelName = extractedModelName
+	}
+
+	// Normalize the model name to handle dynamic thinking suffixes before determining the provider.
+	normalizedModel, metadata = normalizeModelMetadata(targetModelName)
 
 	if isDynamic {
 		providers = []string{providerName}
-		// For dynamic models, the extractedModelName is already normalized by parseDynamicModel
-		// so we use it as the final normalizedModel.
-		normalizedModel = extractedModelName
 	} else {
 		// For non-dynamic models, use the normalizedModel to get the provider name.
 		providers = util.GetProviderName(normalizedModel)
+		if len(providers) == 0 && metadata != nil {
+			if originalRaw, ok := metadata[util.ThinkingOriginalModelMetadataKey]; ok {
+				if originalModel, okStr := originalRaw.(string); okStr {
+					originalModel = strings.TrimSpace(originalModel)
+					if originalModel != "" && !strings.EqualFold(originalModel, normalizedModel) {
+						if altProviders := util.GetProviderName(originalModel); len(altProviders) > 0 {
+							providers = altProviders
+							normalizedModel = originalModel
+						}
+					}
+				}
+			}
+		}
 	}
 
 	if len(providers) == 0 {
@@ -382,7 +396,7 @@ func cloneBytes(src []byte) []byte {
 }
 
 func normalizeModelMetadata(modelName string) (string, map[string]any) {
-	return util.NormalizeGeminiThinkingModel(modelName)
+	return util.NormalizeThinkingModel(modelName)
 }
 
 func cloneMetadata(src map[string]any) map[string]any {

sdk/cliproxy/service.go

@@ -498,7 +498,7 @@ func (s *Service) Run(ctx context.Context) error {
 	}()
 
 	time.Sleep(100 * time.Millisecond)
-	fmt.Printf("API server started successfully on: %d\n", s.cfg.Port)
+	fmt.Printf("API server started successfully on: %s:%d\n", s.cfg.Host, s.cfg.Port)
 
 	if s.hooks.OnAfterStart != nil {
 		s.hooks.OnAfterStart(s)
@@ -779,7 +779,7 @@ func (s *Service) registerModelsForAuth(a *coreauth.Auth) {
 							Created:     time.Now().Unix(),
 							OwnedBy:     compat.Name,
 							Type:        "openai-compatibility",
-							DisplayName: m.Name,
+							DisplayName: modelID,
 						})
 					}
 					// Register and return

test/amp_management_test.go

@@ -0,0 +1,827 @@
+package test
+
+import (
+	"bytes"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/api/handlers/management"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+)
+
+func init() {
+	gin.SetMode(gin.TestMode)
+}
+
+// newAmpTestHandler creates a test handler with default ampcode configuration.
+func newAmpTestHandler(t *testing.T) (*management.Handler, string) {
+	t.Helper()
+	tmpDir := t.TempDir()
+	configPath := filepath.Join(tmpDir, "config.yaml")
+
+	cfg := &config.Config{
+		AmpCode: config.AmpCode{
+			UpstreamURL:                   "https://example.com",
+			UpstreamAPIKey:                "test-api-key-12345",
+			RestrictManagementToLocalhost: true,
+			ForceModelMappings:            false,
+			ModelMappings: []config.AmpModelMapping{
+				{From: "gpt-4", To: "gemini-pro"},
+			},
+		},
+	}
+
+	if err := os.WriteFile(configPath, []byte("port: 8080\n"), 0644); err != nil {
+		t.Fatalf("failed to write config file: %v", err)
+	}
+
+	h := management.NewHandler(cfg, configPath, nil)
+	return h, configPath
+}
+
+// setupAmpRouter creates a test router with all ampcode management endpoints.
+func setupAmpRouter(h *management.Handler) *gin.Engine {
+	r := gin.New()
+	mgmt := r.Group("/v0/management")
+	{
+		mgmt.GET("/ampcode", h.GetAmpCode)
+		mgmt.GET("/ampcode/upstream-url", h.GetAmpUpstreamURL)
+		mgmt.PUT("/ampcode/upstream-url", h.PutAmpUpstreamURL)
+		mgmt.DELETE("/ampcode/upstream-url", h.DeleteAmpUpstreamURL)
+		mgmt.GET("/ampcode/upstream-api-key", h.GetAmpUpstreamAPIKey)
+		mgmt.PUT("/ampcode/upstream-api-key", h.PutAmpUpstreamAPIKey)
+		mgmt.DELETE("/ampcode/upstream-api-key", h.DeleteAmpUpstreamAPIKey)
+		mgmt.GET("/ampcode/restrict-management-to-localhost", h.GetAmpRestrictManagementToLocalhost)
+		mgmt.PUT("/ampcode/restrict-management-to-localhost", h.PutAmpRestrictManagementToLocalhost)
+		mgmt.GET("/ampcode/model-mappings", h.GetAmpModelMappings)
+		mgmt.PUT("/ampcode/model-mappings", h.PutAmpModelMappings)
+		mgmt.PATCH("/ampcode/model-mappings", h.PatchAmpModelMappings)
+		mgmt.DELETE("/ampcode/model-mappings", h.DeleteAmpModelMappings)
+		mgmt.GET("/ampcode/force-model-mappings", h.GetAmpForceModelMappings)
+		mgmt.PUT("/ampcode/force-model-mappings", h.PutAmpForceModelMappings)
+	}
+	return r
+}
+
+// TestGetAmpCode verifies GET /v0/management/ampcode returns full ampcode config.
+func TestGetAmpCode(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodGet, "/v0/management/ampcode", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var resp map[string]config.AmpCode
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal response: %v", err)
+	}
+
+	ampcode := resp["ampcode"]
+	if ampcode.UpstreamURL != "https://example.com" {
+		t.Errorf("expected upstream-url %q, got %q", "https://example.com", ampcode.UpstreamURL)
+	}
+	if len(ampcode.ModelMappings) != 1 {
+		t.Errorf("expected 1 model mapping, got %d", len(ampcode.ModelMappings))
+	}
+}
+
+// TestGetAmpUpstreamURL verifies GET /v0/management/ampcode/upstream-url returns the upstream URL.
+func TestGetAmpUpstreamURL(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/upstream-url", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var resp map[string]string
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal response: %v", err)
+	}
+
+	if resp["upstream-url"] != "https://example.com" {
+		t.Errorf("expected %q, got %q", "https://example.com", resp["upstream-url"])
+	}
+}
+
+// TestPutAmpUpstreamURL verifies PUT /v0/management/ampcode/upstream-url updates the upstream URL.
+func TestPutAmpUpstreamURL(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": "https://new-upstream.com"}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/upstream-url", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d: %s", http.StatusOK, w.Code, w.Body.String())
+	}
+}
+
+// TestDeleteAmpUpstreamURL verifies DELETE /v0/management/ampcode/upstream-url clears the upstream URL.
+func TestDeleteAmpUpstreamURL(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodDelete, "/v0/management/ampcode/upstream-url", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+}
+
+// TestGetAmpUpstreamAPIKey verifies GET /v0/management/ampcode/upstream-api-key returns the API key.
+func TestGetAmpUpstreamAPIKey(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/upstream-api-key", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var resp map[string]any
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal response: %v", err)
+	}
+
+	key := resp["upstream-api-key"].(string)
+	if key != "test-api-key-12345" {
+		t.Errorf("expected key %q, got %q", "test-api-key-12345", key)
+	}
+}
+
+// TestPutAmpUpstreamAPIKey verifies PUT /v0/management/ampcode/upstream-api-key updates the API key.
+func TestPutAmpUpstreamAPIKey(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": "new-secret-key"}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/upstream-api-key", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+}
+
+// TestDeleteAmpUpstreamAPIKey verifies DELETE /v0/management/ampcode/upstream-api-key clears the API key.
+func TestDeleteAmpUpstreamAPIKey(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodDelete, "/v0/management/ampcode/upstream-api-key", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+}
+
+// TestGetAmpRestrictManagementToLocalhost verifies GET returns the localhost restriction setting.
+func TestGetAmpRestrictManagementToLocalhost(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/restrict-management-to-localhost", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var resp map[string]bool
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal response: %v", err)
+	}
+
+	if resp["restrict-management-to-localhost"] != true {
+		t.Error("expected restrict-management-to-localhost to be true")
+	}
+}
+
+// TestPutAmpRestrictManagementToLocalhost verifies PUT updates the localhost restriction setting.
+func TestPutAmpRestrictManagementToLocalhost(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": false}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/restrict-management-to-localhost", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+}
+
+// TestGetAmpModelMappings verifies GET /v0/management/ampcode/model-mappings returns all mappings.
+func TestGetAmpModelMappings(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/model-mappings", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var resp map[string][]config.AmpModelMapping
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal response: %v", err)
+	}
+
+	mappings := resp["model-mappings"]
+	if len(mappings) != 1 {
+		t.Fatalf("expected 1 mapping, got %d", len(mappings))
+	}
+	if mappings[0].From != "gpt-4" || mappings[0].To != "gemini-pro" {
+		t.Errorf("unexpected mapping: %+v", mappings[0])
+	}
+}
+
+// TestPutAmpModelMappings verifies PUT /v0/management/ampcode/model-mappings replaces all mappings.
+func TestPutAmpModelMappings(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": [{"from": "claude-3", "to": "gpt-4o"}, {"from": "gemini", "to": "claude"}]}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d: %s", http.StatusOK, w.Code, w.Body.String())
+	}
+}
+
+// TestPatchAmpModelMappings verifies PATCH updates existing mappings and adds new ones.
+func TestPatchAmpModelMappings(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": [{"from": "gpt-4", "to": "updated-model"}, {"from": "new-model", "to": "target"}]}`
+	req := httptest.NewRequest(http.MethodPatch, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d: %s", http.StatusOK, w.Code, w.Body.String())
+	}
+}
+
+// TestDeleteAmpModelMappings_Specific verifies DELETE removes specified mappings by "from" field.
+func TestDeleteAmpModelMappings_Specific(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": ["gpt-4"]}`
+	req := httptest.NewRequest(http.MethodDelete, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+}
+
+// TestDeleteAmpModelMappings_All verifies DELETE with empty body removes all mappings.
+func TestDeleteAmpModelMappings_All(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodDelete, "/v0/management/ampcode/model-mappings", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+}
+
+// TestGetAmpForceModelMappings verifies GET returns the force-model-mappings setting.
+func TestGetAmpForceModelMappings(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/force-model-mappings", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var resp map[string]bool
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal response: %v", err)
+	}
+
+	if resp["force-model-mappings"] != false {
+		t.Error("expected force-model-mappings to be false")
+	}
+}
+
+// TestPutAmpForceModelMappings verifies PUT updates the force-model-mappings setting.
+func TestPutAmpForceModelMappings(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": true}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/force-model-mappings", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+}
+
+// TestPutAmpModelMappings_VerifyState verifies PUT replaces mappings and state is persisted.
+func TestPutAmpModelMappings_VerifyState(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": [{"from": "model-a", "to": "model-b"}, {"from": "model-c", "to": "model-d"}, {"from": "model-e", "to": "model-f"}]}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("PUT failed: status %d, body: %s", w.Code, w.Body.String())
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/model-mappings", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string][]config.AmpModelMapping
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	mappings := resp["model-mappings"]
+	if len(mappings) != 3 {
+		t.Fatalf("expected 3 mappings, got %d", len(mappings))
+	}
+
+	expected := map[string]string{"model-a": "model-b", "model-c": "model-d", "model-e": "model-f"}
+	for _, m := range mappings {
+		if expected[m.From] != m.To {
+			t.Errorf("mapping %q -> expected %q, got %q", m.From, expected[m.From], m.To)
+		}
+	}
+}
+
+// TestPatchAmpModelMappings_VerifyState verifies PATCH merges mappings correctly.
+func TestPatchAmpModelMappings_VerifyState(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": [{"from": "gpt-4", "to": "updated-target"}, {"from": "new-model", "to": "new-target"}]}`
+	req := httptest.NewRequest(http.MethodPatch, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("PATCH failed: status %d", w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/model-mappings", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string][]config.AmpModelMapping
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	mappings := resp["model-mappings"]
+	if len(mappings) != 2 {
+		t.Fatalf("expected 2 mappings (1 updated + 1 new), got %d", len(mappings))
+	}
+
+	found := make(map[string]string)
+	for _, m := range mappings {
+		found[m.From] = m.To
+	}
+
+	if found["gpt-4"] != "updated-target" {
+		t.Errorf("gpt-4 should map to updated-target, got %q", found["gpt-4"])
+	}
+	if found["new-model"] != "new-target" {
+		t.Errorf("new-model should map to new-target, got %q", found["new-model"])
+	}
+}
+
+// TestDeleteAmpModelMappings_VerifyState verifies DELETE removes specific mappings and keeps others.
+func TestDeleteAmpModelMappings_VerifyState(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	putBody := `{"value": [{"from": "a", "to": "1"}, {"from": "b", "to": "2"}, {"from": "c", "to": "3"}]}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(putBody))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	delBody := `{"value": ["a", "c"]}`
+	req = httptest.NewRequest(http.MethodDelete, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(delBody))
+	req.Header.Set("Content-Type", "application/json")
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("DELETE failed: status %d", w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/model-mappings", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string][]config.AmpModelMapping
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	mappings := resp["model-mappings"]
+	if len(mappings) != 1 {
+		t.Fatalf("expected 1 mapping remaining, got %d", len(mappings))
+	}
+	if mappings[0].From != "b" || mappings[0].To != "2" {
+		t.Errorf("expected b->2, got %s->%s", mappings[0].From, mappings[0].To)
+	}
+}
+
+// TestDeleteAmpModelMappings_NonExistent verifies DELETE with non-existent mapping doesn't affect existing ones.
+func TestDeleteAmpModelMappings_NonExistent(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	delBody := `{"value": ["non-existent-model"]}`
+	req := httptest.NewRequest(http.MethodDelete, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(delBody))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/model-mappings", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string][]config.AmpModelMapping
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	if len(resp["model-mappings"]) != 1 {
+		t.Errorf("original mapping should remain, got %d mappings", len(resp["model-mappings"]))
+	}
+}
+
+// TestPutAmpModelMappings_Empty verifies PUT with empty array clears all mappings.
+func TestPutAmpModelMappings_Empty(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": []}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/model-mappings", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string][]config.AmpModelMapping
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	if len(resp["model-mappings"]) != 0 {
+		t.Errorf("expected 0 mappings, got %d", len(resp["model-mappings"]))
+	}
+}
+
+// TestPutAmpUpstreamURL_VerifyState verifies PUT updates upstream URL and persists state.
+func TestPutAmpUpstreamURL_VerifyState(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": "https://new-api.example.com"}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/upstream-url", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("PUT failed: status %d", w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/upstream-url", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string]string
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	if resp["upstream-url"] != "https://new-api.example.com" {
+		t.Errorf("expected %q, got %q", "https://new-api.example.com", resp["upstream-url"])
+	}
+}
+
+// TestDeleteAmpUpstreamURL_VerifyState verifies DELETE clears upstream URL.
+func TestDeleteAmpUpstreamURL_VerifyState(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodDelete, "/v0/management/ampcode/upstream-url", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("DELETE failed: status %d", w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/upstream-url", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string]string
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	if resp["upstream-url"] != "" {
+		t.Errorf("expected empty string, got %q", resp["upstream-url"])
+	}
+}
+
+// TestPutAmpUpstreamAPIKey_VerifyState verifies PUT updates API key and persists state.
+func TestPutAmpUpstreamAPIKey_VerifyState(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": "new-secret-api-key-xyz"}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/upstream-api-key", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("PUT failed: status %d", w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/upstream-api-key", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string]string
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	if resp["upstream-api-key"] != "new-secret-api-key-xyz" {
+		t.Errorf("expected %q, got %q", "new-secret-api-key-xyz", resp["upstream-api-key"])
+	}
+}
+
+// TestDeleteAmpUpstreamAPIKey_VerifyState verifies DELETE clears API key.
+func TestDeleteAmpUpstreamAPIKey_VerifyState(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodDelete, "/v0/management/ampcode/upstream-api-key", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("DELETE failed: status %d", w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/upstream-api-key", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string]string
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	if resp["upstream-api-key"] != "" {
+		t.Errorf("expected empty string, got %q", resp["upstream-api-key"])
+	}
+}
+
+// TestPutAmpRestrictManagementToLocalhost_VerifyState verifies PUT updates localhost restriction.
+func TestPutAmpRestrictManagementToLocalhost_VerifyState(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": false}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/restrict-management-to-localhost", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("PUT failed: status %d", w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/restrict-management-to-localhost", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string]bool
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	if resp["restrict-management-to-localhost"] != false {
+		t.Error("expected false after update")
+	}
+}
+
+// TestPutAmpForceModelMappings_VerifyState verifies PUT updates force-model-mappings setting.
+func TestPutAmpForceModelMappings_VerifyState(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{"value": true}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/force-model-mappings", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("PUT failed: status %d", w.Code)
+	}
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/force-model-mappings", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string]bool
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	if resp["force-model-mappings"] != true {
+		t.Error("expected true after update")
+	}
+}
+
+// TestPutBoolField_EmptyObject verifies PUT with empty object returns 400.
+func TestPutBoolField_EmptyObject(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	body := `{}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/force-model-mappings", bytes.NewBufferString(body))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Fatalf("expected status %d for empty object, got %d", http.StatusBadRequest, w.Code)
+	}
+}
+
+// TestComplexMappingsWorkflow tests a full workflow: PUT, PATCH, DELETE, and GET.
+func TestComplexMappingsWorkflow(t *testing.T) {
+	h, _ := newAmpTestHandler(t)
+	r := setupAmpRouter(h)
+
+	putBody := `{"value": [{"from": "m1", "to": "t1"}, {"from": "m2", "to": "t2"}, {"from": "m3", "to": "t3"}, {"from": "m4", "to": "t4"}]}`
+	req := httptest.NewRequest(http.MethodPut, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(putBody))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	patchBody := `{"value": [{"from": "m2", "to": "t2-updated"}, {"from": "m5", "to": "t5"}]}`
+	req = httptest.NewRequest(http.MethodPatch, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(patchBody))
+	req.Header.Set("Content-Type", "application/json")
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	delBody := `{"value": ["m1", "m3"]}`
+	req = httptest.NewRequest(http.MethodDelete, "/v0/management/ampcode/model-mappings", bytes.NewBufferString(delBody))
+	req.Header.Set("Content-Type", "application/json")
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	req = httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/model-mappings", nil)
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	var resp map[string][]config.AmpModelMapping
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	mappings := resp["model-mappings"]
+	if len(mappings) != 3 {
+		t.Fatalf("expected 3 mappings (m2, m4, m5), got %d", len(mappings))
+	}
+
+	expected := map[string]string{"m2": "t2-updated", "m4": "t4", "m5": "t5"}
+	found := make(map[string]string)
+	for _, m := range mappings {
+		found[m.From] = m.To
+	}
+
+	for from, to := range expected {
+		if found[from] != to {
+			t.Errorf("mapping %s: expected %q, got %q", from, to, found[from])
+		}
+	}
+}
+
+// TestNilHandlerGetAmpCode verifies handler works with empty config.
+func TestNilHandlerGetAmpCode(t *testing.T) {
+	cfg := &config.Config{}
+	h := management.NewHandler(cfg, "", nil)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodGet, "/v0/management/ampcode", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+}
+
+// TestEmptyConfigGetAmpModelMappings verifies GET returns empty array for fresh config.
+func TestEmptyConfigGetAmpModelMappings(t *testing.T) {
+	cfg := &config.Config{}
+	tmpDir := t.TempDir()
+	configPath := filepath.Join(tmpDir, "config.yaml")
+	if err := os.WriteFile(configPath, []byte("port: 8080\n"), 0644); err != nil {
+		t.Fatalf("failed to write config: %v", err)
+	}
+
+	h := management.NewHandler(cfg, configPath, nil)
+	r := setupAmpRouter(h)
+
+	req := httptest.NewRequest(http.MethodGet, "/v0/management/ampcode/model-mappings", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var resp map[string][]config.AmpModelMapping
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+
+	if len(resp["model-mappings"]) != 0 {
+		t.Errorf("expected 0 mappings, got %d", len(resp["model-mappings"]))
+	}
+}