fix(config): ensure empty mapping persists for oauth-model-alias deletions #1305

fix(logging): add API response timestamp and fix request timestamp timing

README.md

@@ -134,6 +134,10 @@ VSCode extension for quick switching between Claude Code models, featuring integ
 
 Windows desktop app built with Tauri + React for monitoring AI coding assistant quotas via CLIProxyAPI. Track usage across Gemini, Claude, OpenAI Codex, and Antigravity accounts with real-time dashboard, system tray integration, and one-click proxy control - no API keys needed.
 
+### [CPA-XXX Panel](https://github.com/ferretgeek/CPA-X)
+
+A lightweight web admin panel for CLIProxyAPI with health checks, resource monitoring, real-time logs, auto-update, request statistics and pricing display. Supports one-click installation and systemd service.
+
 > [!NOTE]  
 > If you developed a project based on CLIProxyAPI, please open a PR to add it to this list.
 

README_CN.md

@@ -133,6 +133,10 @@ CLI 封装器，用于通过 CLIProxyAPI OAuth 即时切换多个 Claude 账户
 
 Windows 桌面应用，基于 Tauri + React 构建，用于通过 CLIProxyAPI 监控 AI 编程助手配额。支持跨 Gemini、Claude、OpenAI Codex 和 Antigravity 账户的使用量追踪，提供实时仪表盘、系统托盘集成和一键代理控制，无需 API 密钥。
 
+### [CPA-XXX Panel](https://github.com/ferretgeek/CPA-X)
+
+面向 CLIProxyAPI 的 Web 管理面板，提供健康检查、资源监控、日志查看、自动更新、请求统计与定价展示，支持一键安装与 systemd 服务。
+
 > [!NOTE]  
 > 如果你开发了基于 CLIProxyAPI 的项目，请提交一个 PR（拉取请求）将其添加到此列表中。
 

go.mod

@@ -21,7 +21,6 @@ require (
 	golang.org/x/crypto v0.45.0
 	golang.org/x/net v0.47.0
 	golang.org/x/oauth2 v0.30.0
-	golang.org/x/text v0.31.0
 	gopkg.in/natefinch/lumberjack.v2 v2.2.1
 	gopkg.in/yaml.v3 v3.0.1
 )
@@ -71,6 +70,7 @@ require (
 	golang.org/x/arch v0.8.0 // indirect
 	golang.org/x/sync v0.18.0 // indirect
 	golang.org/x/sys v0.38.0 // indirect
+	golang.org/x/text v0.31.0 // indirect
 	google.golang.org/protobuf v1.34.1 // indirect
 	gopkg.in/ini.v1 v1.67.0 // indirect
 )

internal/api/handlers/management/model_definitions.go

@@ -0,0 +1,33 @@
+package management
+
+import (
+	"net/http"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+)
+
+// GetStaticModelDefinitions returns static model metadata for a given channel.
+// Channel is provided via path param (:channel) or query param (?channel=...).
+func (h *Handler) GetStaticModelDefinitions(c *gin.Context) {
+	channel := strings.TrimSpace(c.Param("channel"))
+	if channel == "" {
+		channel = strings.TrimSpace(c.Query("channel"))
+	}
+	if channel == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "channel is required"})
+		return
+	}
+
+	models := registry.GetStaticModelDefinitionsByChannel(channel)
+	if models == nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "unknown channel", "channel": channel})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"channel": strings.ToLower(strings.TrimSpace(channel)),
+		"models":  models,
+	})
+}

internal/api/middleware/request_logging.go

@@ -8,6 +8,7 @@ import (
 	"io"
 	"net/http"
 	"strings"
+	"time"
 
 	"github.com/gin-gonic/gin"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/logging"
@@ -103,6 +104,7 @@ func captureRequestInfo(c *gin.Context) (*RequestInfo, error) {
 		Headers:   headers,
 		Body:      body,
 		RequestID: logging.GetGinRequestID(c),
+		Timestamp: time.Now(),
 	}, nil
 }
 

internal/api/middleware/response_writer.go

@@ -7,6 +7,7 @@ import (
 	"bytes"
 	"net/http"
 	"strings"
+	"time"
 
 	"github.com/gin-gonic/gin"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
@@ -20,22 +21,24 @@ type RequestInfo struct {
 	Headers   map[string][]string // Headers contains the request headers.
 	Body      []byte              // Body is the raw request body.
 	RequestID string              // RequestID is the unique identifier for the request.
+	Timestamp time.Time           // Timestamp is when the request was received.
 }
 
 // ResponseWriterWrapper wraps the standard gin.ResponseWriter to intercept and log response data.
 // It is designed to handle both standard and streaming responses, ensuring that logging operations do not block the client response.
 type ResponseWriterWrapper struct {
 	gin.ResponseWriter
-	body           *bytes.Buffer              // body is a buffer to store the response body for non-streaming responses.
-	isStreaming    bool                       // isStreaming indicates whether the response is a streaming type (e.g., text/event-stream).
-	streamWriter   logging.StreamingLogWriter // streamWriter is a writer for handling streaming log entries.
-	chunkChannel   chan []byte                // chunkChannel is a channel for asynchronously passing response chunks to the logger.
-	streamDone     chan struct{}              // streamDone signals when the streaming goroutine completes.
-	logger         logging.RequestLogger      // logger is the instance of the request logger service.
-	requestInfo    *RequestInfo               // requestInfo holds the details of the original request.
-	statusCode     int                        // statusCode stores the HTTP status code of the response.
-	headers        map[string][]string        // headers stores the response headers.
-	logOnErrorOnly bool                       // logOnErrorOnly enables logging only when an error response is detected.
+	body                *bytes.Buffer              // body is a buffer to store the response body for non-streaming responses.
+	isStreaming         bool                       // isStreaming indicates whether the response is a streaming type (e.g., text/event-stream).
+	streamWriter        logging.StreamingLogWriter // streamWriter is a writer for handling streaming log entries.
+	chunkChannel        chan []byte                // chunkChannel is a channel for asynchronously passing response chunks to the logger.
+	streamDone          chan struct{}              // streamDone signals when the streaming goroutine completes.
+	logger              logging.RequestLogger      // logger is the instance of the request logger service.
+	requestInfo         *RequestInfo               // requestInfo holds the details of the original request.
+	statusCode          int                        // statusCode stores the HTTP status code of the response.
+	headers             map[string][]string        // headers stores the response headers.
+	logOnErrorOnly      bool                       // logOnErrorOnly enables logging only when an error response is detected.
+	firstChunkTimestamp time.Time                  // firstChunkTimestamp captures TTFB for streaming responses.
 }
 
 // NewResponseWriterWrapper creates and initializes a new ResponseWriterWrapper.
@@ -73,6 +76,10 @@ func (w *ResponseWriterWrapper) Write(data []byte) (int, error) {
 
 	// THEN: Handle logging based on response type
 	if w.isStreaming && w.chunkChannel != nil {
+		// Capture TTFB on first chunk (synchronous, before async channel send)
+		if w.firstChunkTimestamp.IsZero() {
+			w.firstChunkTimestamp = time.Now()
+		}
 		// For streaming responses: Send to async logging channel (non-blocking)
 		select {
 		case w.chunkChannel <- append([]byte(nil), data...): // Non-blocking send with copy
@@ -117,6 +124,10 @@ func (w *ResponseWriterWrapper) WriteString(data string) (int, error) {
 
 	// THEN: Capture for logging
 	if w.isStreaming && w.chunkChannel != nil {
+		// Capture TTFB on first chunk (synchronous, before async channel send)
+		if w.firstChunkTimestamp.IsZero() {
+			w.firstChunkTimestamp = time.Now()
+		}
 		select {
 		case w.chunkChannel <- []byte(data):
 		default:
@@ -280,6 +291,8 @@ func (w *ResponseWriterWrapper) Finalize(c *gin.Context) error {
 			w.streamDone = nil
 		}
 
+		w.streamWriter.SetFirstChunkTimestamp(w.firstChunkTimestamp)
+
 		// Write API Request and Response to the streaming log before closing
 		apiRequest := w.extractAPIRequest(c)
 		if len(apiRequest) > 0 {
@@ -297,7 +310,7 @@ func (w *ResponseWriterWrapper) Finalize(c *gin.Context) error {
 		return nil
 	}
 
-	return w.logRequest(finalStatusCode, w.cloneHeaders(), w.body.Bytes(), w.extractAPIRequest(c), w.extractAPIResponse(c), slicesAPIResponseError, forceLog)
+	return w.logRequest(finalStatusCode, w.cloneHeaders(), w.body.Bytes(), w.extractAPIRequest(c), w.extractAPIResponse(c), w.extractAPIResponseTimestamp(c), slicesAPIResponseError, forceLog)
 }
 
 func (w *ResponseWriterWrapper) cloneHeaders() map[string][]string {
@@ -337,7 +350,18 @@ func (w *ResponseWriterWrapper) extractAPIResponse(c *gin.Context) []byte {
 	return data
 }
 
-func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]string, body []byte, apiRequestBody, apiResponseBody []byte, apiResponseErrors []*interfaces.ErrorMessage, forceLog bool) error {
+func (w *ResponseWriterWrapper) extractAPIResponseTimestamp(c *gin.Context) time.Time {
+	ts, isExist := c.Get("API_RESPONSE_TIMESTAMP")
+	if !isExist {
+		return time.Time{}
+	}
+	if t, ok := ts.(time.Time); ok {
+		return t
+	}
+	return time.Time{}
+}
+
+func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]string, body []byte, apiRequestBody, apiResponseBody []byte, apiResponseTimestamp time.Time, apiResponseErrors []*interfaces.ErrorMessage, forceLog bool) error {
 	if w.requestInfo == nil {
 		return nil
 	}
@@ -348,7 +372,7 @@ func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]
 	}
 
 	if loggerWithOptions, ok := w.logger.(interface {
-		LogRequestWithOptions(string, string, map[string][]string, []byte, int, map[string][]string, []byte, []byte, []byte, []*interfaces.ErrorMessage, bool, string) error
+		LogRequestWithOptions(string, string, map[string][]string, []byte, int, map[string][]string, []byte, []byte, []byte, []*interfaces.ErrorMessage, bool, string, time.Time, time.Time) error
 	}); ok {
 		return loggerWithOptions.LogRequestWithOptions(
 			w.requestInfo.URL,
@@ -363,6 +387,8 @@ func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]
 			apiResponseErrors,
 			forceLog,
 			w.requestInfo.RequestID,
+			w.requestInfo.Timestamp,
+			apiResponseTimestamp,
 		)
 	}
 
@@ -378,5 +404,7 @@ func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]
 		apiResponseBody,
 		apiResponseErrors,
 		w.requestInfo.RequestID,
+		w.requestInfo.Timestamp,
+		apiResponseTimestamp,
 	)
 }

internal/api/server.go

@@ -12,6 +12,7 @@ import (
 	"net/http"
 	"os"
 	"path/filepath"
+	"reflect"
 	"strings"
 	"sync"
 	"sync/atomic"
@@ -607,6 +608,7 @@ func (s *Server) registerManagementRoutes() {
 
 		mgmt.GET("/auth-files", s.mgmt.ListAuthFiles)
 		mgmt.GET("/auth-files/models", s.mgmt.GetAuthFileModels)
+		mgmt.GET("/model-definitions/:channel", s.mgmt.GetStaticModelDefinitions)
 		mgmt.GET("/auth-files/download", s.mgmt.DownloadAuthFile)
 		mgmt.POST("/auth-files", s.mgmt.UploadAuthFile)
 		mgmt.DELETE("/auth-files", s.mgmt.DeleteAuthFile)
@@ -989,14 +991,17 @@ func (s *Server) UpdateClients(cfg *config.Config) {
 		s.mgmt.SetAuthManager(s.handlers.AuthManager)
 	}
 
-	// Notify Amp module of config changes (for model mapping hot-reload)
-	if s.ampModule != nil {
-		log.Debugf("triggering amp module config update")
-		if err := s.ampModule.OnConfigUpdated(cfg); err != nil {
-			log.Errorf("failed to update Amp module config: %v", err)
+	// Notify Amp module only when Amp config has changed.
+	ampConfigChanged := oldCfg == nil || !reflect.DeepEqual(oldCfg.AmpCode, cfg.AmpCode)
+	if ampConfigChanged {
+		if s.ampModule != nil {
+			log.Debugf("triggering amp module config update")
+			if err := s.ampModule.OnConfigUpdated(cfg); err != nil {
+				log.Errorf("failed to update Amp module config: %v", err)
+			}
+		} else {
+			log.Warnf("amp module is nil, skipping config update")
 		}
-	} else {
-		log.Warnf("amp module is nil, skipping config update")
 	}
 
 	// Count client sources from configuration and auth store.

internal/config/config.go

@@ -923,6 +923,7 @@ func SaveConfigPreserveComments(configFile string, cfg *Config) error {
 	removeLegacyGenerativeLanguageKeys(original.Content[0])
 
 	pruneMappingToGeneratedKeys(original.Content[0], generated.Content[0], "oauth-excluded-models")
+	pruneMappingToGeneratedKeys(original.Content[0], generated.Content[0], "oauth-model-alias")
 
 	// Merge generated into original in-place, preserving comments/order of existing nodes.
 	mergeMappingPreserve(original.Content[0], generated.Content[0])
@@ -1413,6 +1414,16 @@ func pruneMappingToGeneratedKeys(dstRoot, srcRoot *yaml.Node, key string) {
 	}
 	srcIdx := findMapKeyIndex(srcRoot, key)
 	if srcIdx < 0 {
+		// Keep an explicit empty mapping for oauth-model-alias when it was previously present.
+		//
+		// Rationale: LoadConfig runs MigrateOAuthModelAlias before unmarshalling. If the
+		// oauth-model-alias key is missing, migration will add the default antigravity aliases.
+		// When users delete the last channel from oauth-model-alias via the management API,
+		// we want that deletion to persist across hot reloads and restarts.
+		if key == "oauth-model-alias" {
+			dstRoot.Content[dstIdx+1] = &yaml.Node{Kind: yaml.MappingNode, Tag: "!!map"}
+			return
+		}
 		removeMapKey(dstRoot, key)
 		return
 	}

internal/logging/request_logger.go

@@ -44,10 +44,12 @@ type RequestLogger interface {
 	//   - apiRequest: The API request data
 	//   - apiResponse: The API response data
 	//   - requestID: Optional request ID for log file naming
+	//   - requestTimestamp: When the request was received
+	//   - apiResponseTimestamp: When the API response was received
 	//
 	// Returns:
 	//   - error: An error if logging fails, nil otherwise
-	LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, requestID string) error
+	LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, requestID string, requestTimestamp, apiResponseTimestamp time.Time) error
 
 	// LogStreamingRequest initiates logging for a streaming request and returns a writer for chunks.
 	//
@@ -109,6 +111,12 @@ type StreamingLogWriter interface {
 	//   - error: An error if writing fails, nil otherwise
 	WriteAPIResponse(apiResponse []byte) error
 
+	// SetFirstChunkTimestamp sets the TTFB timestamp captured when first chunk was received.
+	//
+	// Parameters:
+	//   - timestamp: The time when first response chunk was received
+	SetFirstChunkTimestamp(timestamp time.Time)
+
 	// Close finalizes the log file and cleans up resources.
 	//
 	// Returns:
@@ -180,20 +188,22 @@ func (l *FileRequestLogger) SetEnabled(enabled bool) {
 //   - apiRequest: The API request data
 //   - apiResponse: The API response data
 //   - requestID: Optional request ID for log file naming
+//   - requestTimestamp: When the request was received
+//   - apiResponseTimestamp: When the API response was received
 //
 // Returns:
 //   - error: An error if logging fails, nil otherwise
-func (l *FileRequestLogger) LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, requestID string) error {
-	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, false, requestID)
+func (l *FileRequestLogger) LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, requestID string, requestTimestamp, apiResponseTimestamp time.Time) error {
+	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, false, requestID, requestTimestamp, apiResponseTimestamp)
 }
 
 // LogRequestWithOptions logs a request with optional forced logging behavior.
 // The force flag allows writing error logs even when regular request logging is disabled.
-func (l *FileRequestLogger) LogRequestWithOptions(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool, requestID string) error {
-	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, force, requestID)
+func (l *FileRequestLogger) LogRequestWithOptions(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool, requestID string, requestTimestamp, apiResponseTimestamp time.Time) error {
+	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, force, requestID, requestTimestamp, apiResponseTimestamp)
 }
 
-func (l *FileRequestLogger) logRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool, requestID string) error {
+func (l *FileRequestLogger) logRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool, requestID string, requestTimestamp, apiResponseTimestamp time.Time) error {
 	if !l.enabled && !force {
 		return nil
 	}
@@ -247,6 +257,8 @@ func (l *FileRequestLogger) logRequest(url, method string, requestHeaders map[st
 		responseHeaders,
 		responseToWrite,
 		decompressErr,
+		requestTimestamp,
+		apiResponseTimestamp,
 	)
 	if errClose := logFile.Close(); errClose != nil {
 		log.WithError(errClose).Warn("failed to close request log file")
@@ -499,17 +511,22 @@ func (l *FileRequestLogger) writeNonStreamingLog(
 	responseHeaders map[string][]string,
 	response []byte,
 	decompressErr error,
+	requestTimestamp time.Time,
+	apiResponseTimestamp time.Time,
 ) error {
-	if errWrite := writeRequestInfoWithBody(w, url, method, requestHeaders, requestBody, requestBodyPath, time.Now()); errWrite != nil {
+	if requestTimestamp.IsZero() {
+		requestTimestamp = time.Now()
+	}
+	if errWrite := writeRequestInfoWithBody(w, url, method, requestHeaders, requestBody, requestBodyPath, requestTimestamp); errWrite != nil {
 		return errWrite
 	}
-	if errWrite := writeAPISection(w, "=== API REQUEST ===\n", "=== API REQUEST", apiRequest); errWrite != nil {
+	if errWrite := writeAPISection(w, "=== API REQUEST ===\n", "=== API REQUEST", apiRequest, time.Time{}); errWrite != nil {
 		return errWrite
 	}
 	if errWrite := writeAPIErrorResponses(w, apiResponseErrors); errWrite != nil {
 		return errWrite
 	}
-	if errWrite := writeAPISection(w, "=== API RESPONSE ===\n", "=== API RESPONSE", apiResponse); errWrite != nil {
+	if errWrite := writeAPISection(w, "=== API RESPONSE ===\n", "=== API RESPONSE", apiResponse, apiResponseTimestamp); errWrite != nil {
 		return errWrite
 	}
 	return writeResponseSection(w, statusCode, true, responseHeaders, bytes.NewReader(response), decompressErr, true)
@@ -583,7 +600,7 @@ func writeRequestInfoWithBody(
 	return nil
 }
 
-func writeAPISection(w io.Writer, sectionHeader string, sectionPrefix string, payload []byte) error {
+func writeAPISection(w io.Writer, sectionHeader string, sectionPrefix string, payload []byte, timestamp time.Time) error {
 	if len(payload) == 0 {
 		return nil
 	}
@@ -601,6 +618,11 @@ func writeAPISection(w io.Writer, sectionHeader string, sectionPrefix string, pa
 		if _, errWrite := io.WriteString(w, sectionHeader); errWrite != nil {
 			return errWrite
 		}
+		if !timestamp.IsZero() {
+			if _, errWrite := io.WriteString(w, fmt.Sprintf("Timestamp: %s\n", timestamp.Format(time.RFC3339Nano))); errWrite != nil {
+				return errWrite
+			}
+		}
 		if _, errWrite := w.Write(payload); errWrite != nil {
 			return errWrite
 		}
@@ -974,6 +996,9 @@ type FileStreamingLogWriter struct {
 
 	// apiResponse stores the upstream API response data.
 	apiResponse []byte
+
+	// apiResponseTimestamp captures when the API response was received.
+	apiResponseTimestamp time.Time
 }
 
 // WriteChunkAsync writes a response chunk asynchronously (non-blocking).
@@ -1053,6 +1078,12 @@ func (w *FileStreamingLogWriter) WriteAPIResponse(apiResponse []byte) error {
 	return nil
 }
 
+func (w *FileStreamingLogWriter) SetFirstChunkTimestamp(timestamp time.Time) {
+	if !timestamp.IsZero() {
+		w.apiResponseTimestamp = timestamp
+	}
+}
+
 // 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)
@@ -1140,10 +1171,10 @@ func (w *FileStreamingLogWriter) writeFinalLog(logFile *os.File) error {
 	if errWrite := writeRequestInfoWithBody(logFile, w.url, w.method, w.requestHeaders, nil, w.requestBodyPath, w.timestamp); errWrite != nil {
 		return errWrite
 	}
-	if errWrite := writeAPISection(logFile, "=== API REQUEST ===\n", "=== API REQUEST", w.apiRequest); errWrite != nil {
+	if errWrite := writeAPISection(logFile, "=== API REQUEST ===\n", "=== API REQUEST", w.apiRequest, time.Time{}); errWrite != nil {
 		return errWrite
 	}
-	if errWrite := writeAPISection(logFile, "=== API RESPONSE ===\n", "=== API RESPONSE", w.apiResponse); errWrite != nil {
+	if errWrite := writeAPISection(logFile, "=== API RESPONSE ===\n", "=== API RESPONSE", w.apiResponse, w.apiResponseTimestamp); errWrite != nil {
 		return errWrite
 	}
 
@@ -1220,6 +1251,8 @@ func (w *NoOpStreamingLogWriter) WriteAPIResponse(_ []byte) error {
 	return nil
 }
 
+func (w *NoOpStreamingLogWriter) SetFirstChunkTimestamp(_ time.Time) {}
+
 // Close is a no-op implementation that does nothing and always returns nil.
 //
 // Returns:

internal/registry/model_definitions.go

@@ -1,848 +1,69 @@
-// Package registry provides model definitions for various AI service providers.
-// This file contains static model definitions that can be used by clients
-// when registering their supported models.
+// Package registry provides model definitions and lookup helpers for various AI providers.
+// Static model metadata is stored in model_definitions_static_data.go.
 package registry
 
-// GetClaudeModels returns the standard Claude model definitions
-func GetClaudeModels() []*ModelInfo {
-	return []*ModelInfo{
+import (
+	"sort"
+	"strings"
+)
 
-		{
-			ID:                  "claude-haiku-4-5-20251001",
-			Object:              "model",
-			Created:             1759276800, // 2025-10-01
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Haiku",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			// Thinking: not supported for Haiku models
-		},
-		{
-			ID:                  "claude-sonnet-4-5-20250929",
-			Object:              "model",
-			Created:             1759104000, // 2025-09-29
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Sonnet",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-opus-4-5-20251101",
-			Object:              "model",
-			Created:             1761955200, // 2025-11-01
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Opus",
-			Description:         "Premium model combining maximum intelligence with practical performance",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-opus-4-1-20250805",
-			Object:              "model",
-			Created:             1722945600, // 2025-08-05
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.1 Opus",
-			ContextLength:       200000,
-			MaxCompletionTokens: 32000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-opus-4-20250514",
-			Object:              "model",
-			Created:             1715644800, // 2025-05-14
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4 Opus",
-			ContextLength:       200000,
-			MaxCompletionTokens: 32000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-sonnet-4-20250514",
-			Object:              "model",
-			Created:             1715644800, // 2025-05-14
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4 Sonnet",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-3-7-sonnet-20250219",
-			Object:              "model",
-			Created:             1708300800, // 2025-02-19
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 3.7 Sonnet",
-			ContextLength:       128000,
-			MaxCompletionTokens: 8192,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-3-5-haiku-20241022",
-			Object:              "model",
-			Created:             1729555200, // 2024-10-22
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 3.5 Haiku",
-			ContextLength:       128000,
-			MaxCompletionTokens: 8192,
-			// Thinking: not supported for Haiku models
-		},
-	}
-}
-
-// GetGeminiModels returns the standard Gemini model definitions
-func GetGeminiModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                         "gemini-2.5-pro",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-pro",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Pro",
-			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash",
-			Version:                    "001",
-			DisplayName:                "Gemini 2.5 Flash",
-			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-lite",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-lite",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Lite",
-			Description:                "Our smallest and most cost effective model, built for at scale usage.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-3-pro-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Preview",
-			Description:                "Gemini 3 Pro Preview",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
-		},
-		{
-			ID:                         "gemini-3-flash-preview",
-			Object:                     "model",
-			Created:                    1765929600,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-flash-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Flash Preview",
-			Description:                "Gemini 3 Flash Preview",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}},
-		},
-		{
-			ID:                         "gemini-3-pro-image-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-image-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Image Preview",
-			Description:                "Gemini 3 Pro Image Preview",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
-		},
-	}
-}
-
-func GetGeminiVertexModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                         "gemini-2.5-pro",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-pro",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Pro",
-			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash",
-			Version:                    "001",
-			DisplayName:                "Gemini 2.5 Flash",
-			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-lite",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-lite",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Lite",
-			Description:                "Our smallest and most cost effective model, built for at scale usage.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-3-pro-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Preview",
-			Description:                "Gemini 3 Pro Preview",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
-		},
-		{
-			ID:                         "gemini-3-flash-preview",
-			Object:                     "model",
-			Created:                    1765929600,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-flash-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Flash Preview",
-			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}},
-		},
-		{
-			ID:                         "gemini-3-pro-image-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-image-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Image Preview",
-			Description:                "Gemini 3 Pro Image Preview",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
-		},
-		// Imagen image generation models - use :predict action
-		{
-			ID:                         "imagen-4.0-generate-001",
-			Object:                     "model",
-			Created:                    1750000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-4.0-generate-001",
-			Version:                    "4.0",
-			DisplayName:                "Imagen 4.0 Generate",
-			Description:                "Imagen 4.0 image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-		{
-			ID:                         "imagen-4.0-ultra-generate-001",
-			Object:                     "model",
-			Created:                    1750000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-4.0-ultra-generate-001",
-			Version:                    "4.0",
-			DisplayName:                "Imagen 4.0 Ultra Generate",
-			Description:                "Imagen 4.0 Ultra high-quality image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-		{
-			ID:                         "imagen-3.0-generate-002",
-			Object:                     "model",
-			Created:                    1740000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-3.0-generate-002",
-			Version:                    "3.0",
-			DisplayName:                "Imagen 3.0 Generate",
-			Description:                "Imagen 3.0 image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-		{
-			ID:                         "imagen-3.0-fast-generate-001",
-			Object:                     "model",
-			Created:                    1740000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-3.0-fast-generate-001",
-			Version:                    "3.0",
-			DisplayName:                "Imagen 3.0 Fast Generate",
-			Description:                "Imagen 3.0 fast image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-		{
-			ID:                         "imagen-4.0-fast-generate-001",
-			Object:                     "model",
-			Created:                    1750000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-4.0-fast-generate-001",
-			Version:                    "4.0",
-			DisplayName:                "Imagen 4.0 Fast Generate",
-			Description:                "Imagen 4.0 fast image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-	}
-}
-
-// GetGeminiCLIModels returns the standard Gemini model definitions
-func GetGeminiCLIModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                         "gemini-2.5-pro",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-pro",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Pro",
-			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash",
-			Version:                    "001",
-			DisplayName:                "Gemini 2.5 Flash",
-			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-lite",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-lite",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Lite",
-			Description:                "Our smallest and most cost effective model, built for at scale usage.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-3-pro-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Preview",
-			Description:                "Our most intelligent model with SOTA reasoning and multimodal understanding, and powerful agentic and vibe coding capabilities",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
-		},
-		{
-			ID:                         "gemini-3-flash-preview",
-			Object:                     "model",
-			Created:                    1765929600,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-flash-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Flash Preview",
-			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}},
-		},
-	}
-}
-
-// GetAIStudioModels returns the Gemini model definitions for AI Studio integrations
-func GetAIStudioModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                         "gemini-2.5-pro",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-pro",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Pro",
-			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash",
-			Version:                    "001",
-			DisplayName:                "Gemini 2.5 Flash",
-			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-lite",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-lite",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Lite",
-			Description:                "Our smallest and most cost effective model, built for at scale usage.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-3-pro-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Preview",
-			Description:                "Gemini 3 Pro Preview",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-3-flash-preview",
-			Object:                     "model",
-			Created:                    1765929600,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-flash-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Flash Preview",
-			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-pro-latest",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-pro-latest",
-			Version:                    "2.5",
-			DisplayName:                "Gemini Pro Latest",
-			Description:                "Latest release of Gemini Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-flash-latest",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-flash-latest",
-			Version:                    "2.5",
-			DisplayName:                "Gemini Flash Latest",
-			Description:                "Latest release of Gemini Flash",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-flash-lite-latest",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-flash-lite-latest",
-			Version:                    "2.5",
-			DisplayName:                "Gemini Flash-Lite Latest",
-			Description:                "Latest release of Gemini Flash-Lite",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 512, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-image-preview",
-			Object:                     "model",
-			Created:                    1756166400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-image-preview",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Image Preview",
-			Description:                "State-of-the-art image generation and editing model.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           8192,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			// image models don't support thinkingConfig; leave Thinking nil
-		},
-		{
-			ID:                         "gemini-2.5-flash-image",
-			Object:                     "model",
-			Created:                    1759363200,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-image",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Image",
-			Description:                "State-of-the-art image generation and editing model.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           8192,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			// image models don't support thinkingConfig; leave Thinking nil
-		},
-	}
-}
-
-// GetOpenAIModels returns the standard OpenAI model definitions
-func GetOpenAIModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                  "gpt-5",
-			Object:              "model",
-			Created:             1754524800,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-08-07",
-			DisplayName:         "GPT 5",
-			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",
-			Object:              "model",
-			Created:             1757894400,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-09-15",
-			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"},
-			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
-		},
-		{
-			ID:                  "gpt-5-codex-mini",
-			Object:              "model",
-			Created:             1762473600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-11-07",
-			DisplayName:         "GPT 5 Codex Mini",
-			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",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5",
-			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",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			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",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			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-max",
-			Object:              "model",
-			Created:             1763424000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-max",
-			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"}},
-		},
-		{
-			ID:                  "gpt-5.2",
-			Object:              "model",
-			Created:             1765440000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.2",
-			DisplayName:         "GPT 5.2",
-			Description:         "Stable version of GPT 5.2",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"none", "low", "medium", "high", "xhigh"}},
-		},
-		{
-			ID:                  "gpt-5.2-codex",
-			Object:              "model",
-			Created:             1765440000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.2",
-			DisplayName:         "GPT 5.2 Codex",
-			Description:         "Stable version of GPT 5.2 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", "xhigh"}},
-		},
-	}
-}
-
-// GetQwenModels returns the standard Qwen model definitions
-func GetQwenModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                  "qwen3-coder-plus",
-			Object:              "model",
-			Created:             1753228800,
-			OwnedBy:             "qwen",
-			Type:                "qwen",
-			Version:             "3.0",
-			DisplayName:         "Qwen3 Coder Plus",
-			Description:         "Advanced code generation and understanding model",
-			ContextLength:       32768,
-			MaxCompletionTokens: 8192,
-			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
-		},
-		{
-			ID:                  "qwen3-coder-flash",
-			Object:              "model",
-			Created:             1753228800,
-			OwnedBy:             "qwen",
-			Type:                "qwen",
-			Version:             "3.0",
-			DisplayName:         "Qwen3 Coder Flash",
-			Description:         "Fast code generation model",
-			ContextLength:       8192,
-			MaxCompletionTokens: 2048,
-			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
-		},
-		{
-			ID:                  "vision-model",
-			Object:              "model",
-			Created:             1758672000,
-			OwnedBy:             "qwen",
-			Type:                "qwen",
-			Version:             "3.0",
-			DisplayName:         "Qwen3 Vision Model",
-			Description:         "Vision model model",
-			ContextLength:       32768,
-			MaxCompletionTokens: 2048,
-			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
-		},
-	}
-}
-
-// iFlowThinkingSupport is a shared ThinkingSupport configuration for iFlow models
-// that support thinking mode via chat_template_kwargs.enable_thinking (boolean toggle).
-// Uses level-based configuration so standard normalization flows apply before conversion.
-var iFlowThinkingSupport = &ThinkingSupport{
-	Levels: []string{"none", "auto", "minimal", "low", "medium", "high", "xhigh"},
-}
-
-// 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},
-		{ID: "qwen3-max", DisplayName: "Qwen3-Max", Description: "Qwen3 flagship model", Created: 1758672000},
-		{ID: "qwen3-vl-plus", DisplayName: "Qwen3-VL-Plus", Description: "Qwen3 multimodal vision-language", Created: 1758672000},
-		{ID: "qwen3-max-preview", DisplayName: "Qwen3-Max-Preview", Description: "Qwen3 Max preview build", Created: 1757030400},
-		{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, Thinking: iFlowThinkingSupport},
-		{ID: "glm-4.7", DisplayName: "GLM-4.7", Description: "Zhipu GLM 4.7 general model", Created: 1766448000, Thinking: iFlowThinkingSupport},
-		{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 thinking model", Created: 1762387200},
-		{ID: "deepseek-v3.2-chat", DisplayName: "DeepSeek-V3.2", Description: "DeepSeek V3.2 Chat", Created: 1764576000},
-		{ID: "deepseek-v3.2-reasoner", DisplayName: "DeepSeek-V3.2", Description: "DeepSeek V3.2 Reasoner", 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-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-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, Thinking: iFlowThinkingSupport},
-		{ID: "minimax-m2.1", DisplayName: "MiniMax-M2.1", Description: "MiniMax M2.1", Created: 1766448000, Thinking: iFlowThinkingSupport},
-		{ID: "iflow-rome-30ba3b", DisplayName: "iFlow-ROME", Description: "iFlow Rome 30BA3B model", Created: 1736899200},
-	}
-	models := make([]*ModelInfo, 0, len(entries))
-	for _, entry := range entries {
-		models = append(models, &ModelInfo{
-			ID:          entry.ID,
-			Object:      "model",
-			Created:     entry.Created,
-			OwnedBy:     "iflow",
-			Type:        "iflow",
-			DisplayName: entry.DisplayName,
-			Description: entry.Description,
-			Thinking:    entry.Thinking,
+// GetStaticModelDefinitionsByChannel returns static model definitions for a given channel/provider.
+// It returns nil when the channel is unknown.
+//
+// Supported channels:
+//   - claude
+//   - gemini
+//   - vertex
+//   - gemini-cli
+//   - aistudio
+//   - codex
+//   - qwen
+//   - iflow
+//   - antigravity (returns static overrides only)
+func GetStaticModelDefinitionsByChannel(channel string) []*ModelInfo {
+	key := strings.ToLower(strings.TrimSpace(channel))
+	switch key {
+	case "claude":
+		return GetClaudeModels()
+	case "gemini":
+		return GetGeminiModels()
+	case "vertex":
+		return GetGeminiVertexModels()
+	case "gemini-cli":
+		return GetGeminiCLIModels()
+	case "aistudio":
+		return GetAIStudioModels()
+	case "codex":
+		return GetOpenAIModels()
+	case "qwen":
+		return GetQwenModels()
+	case "iflow":
+		return GetIFlowModels()
+	case "antigravity":
+		cfg := GetAntigravityModelConfig()
+		if len(cfg) == 0 {
+			return nil
+		}
+		models := make([]*ModelInfo, 0, len(cfg))
+		for modelID, entry := range cfg {
+			if modelID == "" || entry == nil {
+				continue
+			}
+			models = append(models, &ModelInfo{
+				ID:                  modelID,
+				Object:              "model",
+				OwnedBy:             "antigravity",
+				Type:                "antigravity",
+				Thinking:            entry.Thinking,
+				MaxCompletionTokens: entry.MaxCompletionTokens,
+			})
+		}
+		sort.Slice(models, func(i, j int) bool {
+			return strings.ToLower(models[i].ID) < strings.ToLower(models[j].ID)
 		})
-	}
-	return models
-}
-
-// AntigravityModelConfig captures static antigravity model overrides, including
-// Thinking budget limits and provider max completion tokens.
-type AntigravityModelConfig struct {
-	Thinking            *ThinkingSupport
-	MaxCompletionTokens int
-}
-
-// GetAntigravityModelConfig returns static configuration for antigravity models.
-// Keys use upstream model names returned by the Antigravity models endpoint.
-func GetAntigravityModelConfig() map[string]*AntigravityModelConfig {
-	return map[string]*AntigravityModelConfig{
-		"gemini-2.5-flash":           {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}},
-		"gemini-2.5-flash-lite":      {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}},
-		"rev19-uic3-1p":              {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true}},
-		"gemini-3-pro-high":          {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}}},
-		"gemini-3-pro-image":         {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}}},
-		"gemini-3-flash":             {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}}},
-		"claude-sonnet-4-5-thinking": {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
-		"claude-opus-4-5-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
-		"claude-sonnet-4-5":          {MaxCompletionTokens: 64000},
-		"gpt-oss-120b-medium":        {},
-		"tab_flash_lite_preview":     {},
+		return models
+	default:
+		return nil
 	}
 }
 

internal/registry/model_definitions_static_data.go

@@ -0,0 +1,846 @@
+// Package registry provides model definitions for various AI service providers.
+// This file stores the static model metadata catalog.
+package registry
+
+// GetClaudeModels returns the standard Claude model definitions
+func GetClaudeModels() []*ModelInfo {
+	return []*ModelInfo{
+
+		{
+			ID:                  "claude-haiku-4-5-20251001",
+			Object:              "model",
+			Created:             1759276800, // 2025-10-01
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Haiku",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			// Thinking: not supported for Haiku models
+		},
+		{
+			ID:                  "claude-sonnet-4-5-20250929",
+			Object:              "model",
+			Created:             1759104000, // 2025-09-29
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Sonnet",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-opus-4-5-20251101",
+			Object:              "model",
+			Created:             1761955200, // 2025-11-01
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Opus",
+			Description:         "Premium model combining maximum intelligence with practical performance",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-opus-4-1-20250805",
+			Object:              "model",
+			Created:             1722945600, // 2025-08-05
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.1 Opus",
+			ContextLength:       200000,
+			MaxCompletionTokens: 32000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-opus-4-20250514",
+			Object:              "model",
+			Created:             1715644800, // 2025-05-14
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4 Opus",
+			ContextLength:       200000,
+			MaxCompletionTokens: 32000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-sonnet-4-20250514",
+			Object:              "model",
+			Created:             1715644800, // 2025-05-14
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4 Sonnet",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-3-7-sonnet-20250219",
+			Object:              "model",
+			Created:             1708300800, // 2025-02-19
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 3.7 Sonnet",
+			ContextLength:       128000,
+			MaxCompletionTokens: 8192,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-3-5-haiku-20241022",
+			Object:              "model",
+			Created:             1729555200, // 2024-10-22
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 3.5 Haiku",
+			ContextLength:       128000,
+			MaxCompletionTokens: 8192,
+			// Thinking: not supported for Haiku models
+		},
+	}
+}
+
+// GetGeminiModels returns the standard Gemini model definitions
+func GetGeminiModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                         "gemini-2.5-pro",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-pro",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Pro",
+			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash",
+			Version:                    "001",
+			DisplayName:                "Gemini 2.5 Flash",
+			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash-lite",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-lite",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Lite",
+			Description:                "Our smallest and most cost effective model, built for at scale usage.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-3-pro-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Preview",
+			Description:                "Gemini 3 Pro Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
+		},
+		{
+			ID:                         "gemini-3-flash-preview",
+			Object:                     "model",
+			Created:                    1765929600,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-flash-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Flash Preview",
+			Description:                "Gemini 3 Flash Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}},
+		},
+		{
+			ID:                         "gemini-3-pro-image-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-image-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Image Preview",
+			Description:                "Gemini 3 Pro Image Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
+		},
+	}
+}
+
+func GetGeminiVertexModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                         "gemini-2.5-pro",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-pro",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Pro",
+			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash",
+			Version:                    "001",
+			DisplayName:                "Gemini 2.5 Flash",
+			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash-lite",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-lite",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Lite",
+			Description:                "Our smallest and most cost effective model, built for at scale usage.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-3-pro-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Preview",
+			Description:                "Gemini 3 Pro Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
+		},
+		{
+			ID:                         "gemini-3-flash-preview",
+			Object:                     "model",
+			Created:                    1765929600,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-flash-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Flash Preview",
+			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}},
+		},
+		{
+			ID:                         "gemini-3-pro-image-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-image-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Image Preview",
+			Description:                "Gemini 3 Pro Image Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
+		},
+		// Imagen image generation models - use :predict action
+		{
+			ID:                         "imagen-4.0-generate-001",
+			Object:                     "model",
+			Created:                    1750000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-4.0-generate-001",
+			Version:                    "4.0",
+			DisplayName:                "Imagen 4.0 Generate",
+			Description:                "Imagen 4.0 image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+		{
+			ID:                         "imagen-4.0-ultra-generate-001",
+			Object:                     "model",
+			Created:                    1750000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-4.0-ultra-generate-001",
+			Version:                    "4.0",
+			DisplayName:                "Imagen 4.0 Ultra Generate",
+			Description:                "Imagen 4.0 Ultra high-quality image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+		{
+			ID:                         "imagen-3.0-generate-002",
+			Object:                     "model",
+			Created:                    1740000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-3.0-generate-002",
+			Version:                    "3.0",
+			DisplayName:                "Imagen 3.0 Generate",
+			Description:                "Imagen 3.0 image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+		{
+			ID:                         "imagen-3.0-fast-generate-001",
+			Object:                     "model",
+			Created:                    1740000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-3.0-fast-generate-001",
+			Version:                    "3.0",
+			DisplayName:                "Imagen 3.0 Fast Generate",
+			Description:                "Imagen 3.0 fast image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+		{
+			ID:                         "imagen-4.0-fast-generate-001",
+			Object:                     "model",
+			Created:                    1750000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-4.0-fast-generate-001",
+			Version:                    "4.0",
+			DisplayName:                "Imagen 4.0 Fast Generate",
+			Description:                "Imagen 4.0 fast image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+	}
+}
+
+// GetGeminiCLIModels returns the standard Gemini model definitions
+func GetGeminiCLIModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                         "gemini-2.5-pro",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-pro",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Pro",
+			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash",
+			Version:                    "001",
+			DisplayName:                "Gemini 2.5 Flash",
+			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash-lite",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-lite",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Lite",
+			Description:                "Our smallest and most cost effective model, built for at scale usage.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-3-pro-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Preview",
+			Description:                "Our most intelligent model with SOTA reasoning and multimodal understanding, and powerful agentic and vibe coding capabilities",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
+		},
+		{
+			ID:                         "gemini-3-flash-preview",
+			Object:                     "model",
+			Created:                    1765929600,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-flash-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Flash Preview",
+			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}},
+		},
+	}
+}
+
+// GetAIStudioModels returns the Gemini model definitions for AI Studio integrations
+func GetAIStudioModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                         "gemini-2.5-pro",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-pro",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Pro",
+			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash",
+			Version:                    "001",
+			DisplayName:                "Gemini 2.5 Flash",
+			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash-lite",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-lite",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Lite",
+			Description:                "Our smallest and most cost effective model, built for at scale usage.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-3-pro-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Preview",
+			Description:                "Gemini 3 Pro Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-3-flash-preview",
+			Object:                     "model",
+			Created:                    1765929600,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-flash-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Flash Preview",
+			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-pro-latest",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-pro-latest",
+			Version:                    "2.5",
+			DisplayName:                "Gemini Pro Latest",
+			Description:                "Latest release of Gemini Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-flash-latest",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-flash-latest",
+			Version:                    "2.5",
+			DisplayName:                "Gemini Flash Latest",
+			Description:                "Latest release of Gemini Flash",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-flash-lite-latest",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-flash-lite-latest",
+			Version:                    "2.5",
+			DisplayName:                "Gemini Flash-Lite Latest",
+			Description:                "Latest release of Gemini Flash-Lite",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 512, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		// {
+		// 	ID:                         "gemini-2.5-flash-image-preview",
+		// 	Object:                     "model",
+		// 	Created:                    1756166400,
+		// 	OwnedBy:                    "google",
+		// 	Type:                       "gemini",
+		// 	Name:                       "models/gemini-2.5-flash-image-preview",
+		// 	Version:                    "2.5",
+		// 	DisplayName:                "Gemini 2.5 Flash Image Preview",
+		// 	Description:                "State-of-the-art image generation and editing model.",
+		// 	InputTokenLimit:            1048576,
+		// 	OutputTokenLimit:           8192,
+		// 	SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+		// 	// image models don't support thinkingConfig; leave Thinking nil
+		// },
+		{
+			ID:                         "gemini-2.5-flash-image",
+			Object:                     "model",
+			Created:                    1759363200,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-image",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Image",
+			Description:                "State-of-the-art image generation and editing model.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           8192,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			// image models don't support thinkingConfig; leave Thinking nil
+		},
+	}
+}
+
+// GetOpenAIModels returns the standard OpenAI model definitions
+func GetOpenAIModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                  "gpt-5",
+			Object:              "model",
+			Created:             1754524800,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5-2025-08-07",
+			DisplayName:         "GPT 5",
+			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",
+			Object:              "model",
+			Created:             1757894400,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5-2025-09-15",
+			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"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
+		},
+		{
+			ID:                  "gpt-5-codex-mini",
+			Object:              "model",
+			Created:             1762473600,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5-2025-11-07",
+			DisplayName:         "GPT 5 Codex Mini",
+			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",
+			Object:              "model",
+			Created:             1762905600,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.1-2025-11-12",
+			DisplayName:         "GPT 5",
+			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",
+			Object:              "model",
+			Created:             1762905600,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.1-2025-11-12",
+			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",
+			Object:              "model",
+			Created:             1762905600,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.1-2025-11-12",
+			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-max",
+			Object:              "model",
+			Created:             1763424000,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.1-max",
+			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"}},
+		},
+		{
+			ID:                  "gpt-5.2",
+			Object:              "model",
+			Created:             1765440000,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.2",
+			DisplayName:         "GPT 5.2",
+			Description:         "Stable version of GPT 5.2",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"none", "low", "medium", "high", "xhigh"}},
+		},
+		{
+			ID:                  "gpt-5.2-codex",
+			Object:              "model",
+			Created:             1765440000,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.2",
+			DisplayName:         "GPT 5.2 Codex",
+			Description:         "Stable version of GPT 5.2 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", "xhigh"}},
+		},
+	}
+}
+
+// GetQwenModels returns the standard Qwen model definitions
+func GetQwenModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                  "qwen3-coder-plus",
+			Object:              "model",
+			Created:             1753228800,
+			OwnedBy:             "qwen",
+			Type:                "qwen",
+			Version:             "3.0",
+			DisplayName:         "Qwen3 Coder Plus",
+			Description:         "Advanced code generation and understanding model",
+			ContextLength:       32768,
+			MaxCompletionTokens: 8192,
+			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
+		},
+		{
+			ID:                  "qwen3-coder-flash",
+			Object:              "model",
+			Created:             1753228800,
+			OwnedBy:             "qwen",
+			Type:                "qwen",
+			Version:             "3.0",
+			DisplayName:         "Qwen3 Coder Flash",
+			Description:         "Fast code generation model",
+			ContextLength:       8192,
+			MaxCompletionTokens: 2048,
+			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
+		},
+		{
+			ID:                  "vision-model",
+			Object:              "model",
+			Created:             1758672000,
+			OwnedBy:             "qwen",
+			Type:                "qwen",
+			Version:             "3.0",
+			DisplayName:         "Qwen3 Vision Model",
+			Description:         "Vision model model",
+			ContextLength:       32768,
+			MaxCompletionTokens: 2048,
+			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
+		},
+	}
+}
+
+// iFlowThinkingSupport is a shared ThinkingSupport configuration for iFlow models
+// that support thinking mode via chat_template_kwargs.enable_thinking (boolean toggle).
+// Uses level-based configuration so standard normalization flows apply before conversion.
+var iFlowThinkingSupport = &ThinkingSupport{
+	Levels: []string{"none", "auto", "minimal", "low", "medium", "high", "xhigh"},
+}
+
+// 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},
+		{ID: "qwen3-max", DisplayName: "Qwen3-Max", Description: "Qwen3 flagship model", Created: 1758672000},
+		{ID: "qwen3-vl-plus", DisplayName: "Qwen3-VL-Plus", Description: "Qwen3 multimodal vision-language", Created: 1758672000},
+		{ID: "qwen3-max-preview", DisplayName: "Qwen3-Max-Preview", Description: "Qwen3 Max preview build", Created: 1757030400, Thinking: iFlowThinkingSupport},
+		{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, Thinking: iFlowThinkingSupport},
+		{ID: "glm-4.7", DisplayName: "GLM-4.7", Description: "Zhipu GLM 4.7 general model", Created: 1766448000, Thinking: iFlowThinkingSupport},
+		{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 thinking model", Created: 1762387200},
+		{ID: "deepseek-v3.2-chat", DisplayName: "DeepSeek-V3.2", Description: "DeepSeek V3.2 Chat", Created: 1764576000},
+		{ID: "deepseek-v3.2-reasoner", DisplayName: "DeepSeek-V3.2", Description: "DeepSeek V3.2 Reasoner", Created: 1764576000},
+		{ID: "deepseek-v3.2", DisplayName: "DeepSeek-V3.2-Exp", Description: "DeepSeek V3.2 experimental", Created: 1759104000, Thinking: iFlowThinkingSupport},
+		{ID: "deepseek-v3.1", DisplayName: "DeepSeek-V3.1-Terminus", Description: "DeepSeek V3.1 Terminus", Created: 1756339200, Thinking: iFlowThinkingSupport},
+		{ID: "deepseek-r1", DisplayName: "DeepSeek-R1", Description: "DeepSeek reasoning model R1", Created: 1737331200},
+		{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-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, Thinking: iFlowThinkingSupport},
+		{ID: "minimax-m2.1", DisplayName: "MiniMax-M2.1", Description: "MiniMax M2.1", Created: 1766448000, Thinking: iFlowThinkingSupport},
+		{ID: "iflow-rome-30ba3b", DisplayName: "iFlow-ROME", Description: "iFlow Rome 30BA3B model", Created: 1736899200},
+	}
+	models := make([]*ModelInfo, 0, len(entries))
+	for _, entry := range entries {
+		models = append(models, &ModelInfo{
+			ID:          entry.ID,
+			Object:      "model",
+			Created:     entry.Created,
+			OwnedBy:     "iflow",
+			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
+}
+
+// GetAntigravityModelConfig returns static configuration for antigravity models.
+// Keys use upstream model names returned by the Antigravity models endpoint.
+func GetAntigravityModelConfig() map[string]*AntigravityModelConfig {
+	return map[string]*AntigravityModelConfig{
+		// "rev19-uic3-1p":              {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true}},
+		"gemini-2.5-flash":           {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}},
+		"gemini-2.5-flash-lite":      {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}},
+		"gemini-3-pro-high":          {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}}},
+		"gemini-3-pro-image":         {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}}},
+		"gemini-3-flash":             {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}}},
+		"claude-sonnet-4-5-thinking": {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
+		"claude-opus-4-5-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
+		"claude-sonnet-4-5":          {MaxCompletionTokens: 64000},
+		"gpt-oss-120b-medium":        {},
+		"tab_flash_lite_preview":     {},
+	}
+}

internal/runtime/executor/antigravity_executor.go

@@ -148,87 +148,108 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
 
-	var lastStatus int
-	var lastBody []byte
-	var lastErr error
+	attempts := antigravityRetryAttempts(auth, e.cfg)
 
-	for idx, baseURL := range baseURLs {
-		httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, false, opts.Alt, baseURL)
-		if errReq != nil {
-			err = errReq
-			return resp, err
-		}
+attemptLoop:
+	for attempt := 0; attempt < attempts; attempt++ {
+		var lastStatus int
+		var lastBody []byte
+		var lastErr error
 
-		httpResp, errDo := httpClient.Do(httpReq)
-		if errDo != nil {
-			recordAPIResponseError(ctx, e.cfg, errDo)
-			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
-				return resp, errDo
+		for idx, baseURL := range baseURLs {
+			httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, false, opts.Alt, baseURL)
+			if errReq != nil {
+				err = errReq
+				return resp, err
 			}
-			lastStatus = 0
-			lastBody = nil
-			lastErr = errDo
-			if idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
+
+			httpResp, errDo := httpClient.Do(httpReq)
+			if errDo != nil {
+				recordAPIResponseError(ctx, e.cfg, errDo)
+				if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+					return resp, errDo
+				}
+				lastStatus = 0
+				lastBody = nil
+				lastErr = errDo
+				if idx+1 < len(baseURLs) {
+					log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					continue
+				}
+				err = errDo
+				return resp, err
 			}
-			err = errDo
-			return resp, err
+
+			recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+			bodyBytes, errRead := io.ReadAll(httpResp.Body)
+			if errClose := httpResp.Body.Close(); errClose != nil {
+				log.Errorf("antigravity executor: close response body error: %v", errClose)
+			}
+			if errRead != nil {
+				recordAPIResponseError(ctx, e.cfg, errRead)
+				err = errRead
+				return resp, err
+			}
+			appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
+
+			if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+				log.Debugf("antigravity executor: upstream error status: %d, body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), bodyBytes))
+				lastStatus = httpResp.StatusCode
+				lastBody = append([]byte(nil), bodyBytes...)
+				lastErr = nil
+				if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
+					log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					continue
+				}
+				if antigravityShouldRetryNoCapacity(httpResp.StatusCode, bodyBytes) {
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: no capacity on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					if attempt+1 < attempts {
+						delay := antigravityNoCapacityRetryDelay(attempt)
+						log.Debugf("antigravity executor: no capacity for model %s, retrying in %s (attempt %d/%d)", baseModel, delay, attempt+1, attempts)
+						if errWait := antigravityWait(ctx, delay); errWait != nil {
+							return resp, errWait
+						}
+						continue attemptLoop
+					}
+				}
+				sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+				if httpResp.StatusCode == http.StatusTooManyRequests {
+					if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+						sErr.retryAfter = retryAfter
+					}
+				}
+				err = sErr
+				return resp, err
+			}
+
+			reporter.publish(ctx, parseAntigravityUsage(bodyBytes))
+			var param any
+			converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bodyBytes, &param)
+			resp = cliproxyexecutor.Response{Payload: []byte(converted)}
+			reporter.ensurePublished(ctx)
+			return resp, nil
 		}
 
-		recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
-		bodyBytes, errRead := io.ReadAll(httpResp.Body)
-		if errClose := httpResp.Body.Close(); errClose != nil {
-			log.Errorf("antigravity executor: close response body error: %v", errClose)
-		}
-		if errRead != nil {
-			recordAPIResponseError(ctx, e.cfg, errRead)
-			err = errRead
-			return resp, err
-		}
-		appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
-
-		if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
-			log.Debugf("antigravity executor: upstream error status: %d, body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), bodyBytes))
-			lastStatus = httpResp.StatusCode
-			lastBody = append([]byte(nil), bodyBytes...)
-			lastErr = nil
-			if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
-			}
-			sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
-			if httpResp.StatusCode == http.StatusTooManyRequests {
-				if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+		switch {
+		case lastStatus != 0:
+			sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+			if lastStatus == http.StatusTooManyRequests {
+				if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
 					sErr.retryAfter = retryAfter
 				}
 			}
 			err = sErr
-			return resp, err
+		case lastErr != nil:
+			err = lastErr
+		default:
+			err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
 		}
-
-		reporter.publish(ctx, parseAntigravityUsage(bodyBytes))
-		var param any
-		converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bodyBytes, &param)
-		resp = cliproxyexecutor.Response{Payload: []byte(converted)}
-		reporter.ensurePublished(ctx)
-		return resp, nil
+		return resp, err
 	}
 
-	switch {
-	case lastStatus != 0:
-		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
-		if lastStatus == http.StatusTooManyRequests {
-			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
-				sErr.retryAfter = retryAfter
-			}
-		}
-		err = sErr
-	case lastErr != nil:
-		err = lastErr
-	default:
-		err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
-	}
 	return resp, err
 }
 
@@ -268,150 +289,171 @@ func (e *AntigravityExecutor) executeClaudeNonStream(ctx context.Context, auth *
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
 
-	var lastStatus int
-	var lastBody []byte
-	var lastErr error
+	attempts := antigravityRetryAttempts(auth, e.cfg)
 
-	for idx, baseURL := range baseURLs {
-		httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, true, opts.Alt, baseURL)
-		if errReq != nil {
-			err = errReq
-			return resp, err
-		}
+attemptLoop:
+	for attempt := 0; attempt < attempts; attempt++ {
+		var lastStatus int
+		var lastBody []byte
+		var lastErr error
 
-		httpResp, errDo := httpClient.Do(httpReq)
-		if errDo != nil {
-			recordAPIResponseError(ctx, e.cfg, errDo)
-			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
-				return resp, errDo
+		for idx, baseURL := range baseURLs {
+			httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, true, opts.Alt, baseURL)
+			if errReq != nil {
+				err = errReq
+				return resp, err
 			}
-			lastStatus = 0
-			lastBody = nil
-			lastErr = errDo
-			if idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
-			}
-			err = errDo
-			return resp, err
-		}
-		recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
-		if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
-			bodyBytes, errRead := io.ReadAll(httpResp.Body)
-			if errClose := httpResp.Body.Close(); errClose != nil {
-				log.Errorf("antigravity executor: close response body error: %v", errClose)
-			}
-			if errRead != nil {
-				recordAPIResponseError(ctx, e.cfg, errRead)
-				if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
-					err = errRead
-					return resp, err
-				}
-				if errCtx := ctx.Err(); errCtx != nil {
-					err = errCtx
-					return resp, err
+
+			httpResp, errDo := httpClient.Do(httpReq)
+			if errDo != nil {
+				recordAPIResponseError(ctx, e.cfg, errDo)
+				if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+					return resp, errDo
 				}
 				lastStatus = 0
 				lastBody = nil
-				lastErr = errRead
+				lastErr = errDo
 				if idx+1 < len(baseURLs) {
-					log.Debugf("antigravity executor: read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
 					continue
 				}
-				err = errRead
+				err = errDo
 				return resp, err
 			}
-			appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
-			lastStatus = httpResp.StatusCode
-			lastBody = append([]byte(nil), bodyBytes...)
-			lastErr = nil
-			if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
+			recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+			if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+				bodyBytes, errRead := io.ReadAll(httpResp.Body)
+				if errClose := httpResp.Body.Close(); errClose != nil {
+					log.Errorf("antigravity executor: close response body error: %v", errClose)
+				}
+				if errRead != nil {
+					recordAPIResponseError(ctx, e.cfg, errRead)
+					if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
+						err = errRead
+						return resp, err
+					}
+					if errCtx := ctx.Err(); errCtx != nil {
+						err = errCtx
+						return resp, err
+					}
+					lastStatus = 0
+					lastBody = nil
+					lastErr = errRead
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					err = errRead
+					return resp, err
+				}
+				appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
+				lastStatus = httpResp.StatusCode
+				lastBody = append([]byte(nil), bodyBytes...)
+				lastErr = nil
+				if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
+					log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					continue
+				}
+				if antigravityShouldRetryNoCapacity(httpResp.StatusCode, bodyBytes) {
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: no capacity on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					if attempt+1 < attempts {
+						delay := antigravityNoCapacityRetryDelay(attempt)
+						log.Debugf("antigravity executor: no capacity for model %s, retrying in %s (attempt %d/%d)", baseModel, delay, attempt+1, attempts)
+						if errWait := antigravityWait(ctx, delay); errWait != nil {
+							return resp, errWait
+						}
+						continue attemptLoop
+					}
+				}
+				sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+				if httpResp.StatusCode == http.StatusTooManyRequests {
+					if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+						sErr.retryAfter = retryAfter
+					}
+				}
+				err = sErr
+				return resp, err
 			}
-			sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
-			if httpResp.StatusCode == http.StatusTooManyRequests {
-				if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+
+			out := make(chan cliproxyexecutor.StreamChunk)
+			go func(resp *http.Response) {
+				defer close(out)
+				defer func() {
+					if errClose := resp.Body.Close(); errClose != nil {
+						log.Errorf("antigravity executor: close response body error: %v", errClose)
+					}
+				}()
+				scanner := bufio.NewScanner(resp.Body)
+				scanner.Buffer(nil, streamScannerBuffer)
+				for scanner.Scan() {
+					line := scanner.Bytes()
+					appendAPIResponseChunk(ctx, e.cfg, line)
+
+					// Filter usage metadata for all models
+					// Only retain usage statistics in the terminal chunk
+					line = FilterSSEUsageMetadata(line)
+
+					payload := jsonPayload(line)
+					if payload == nil {
+						continue
+					}
+
+					if detail, ok := parseAntigravityStreamUsage(payload); ok {
+						reporter.publish(ctx, detail)
+					}
+
+					out <- cliproxyexecutor.StreamChunk{Payload: payload}
+				}
+				if errScan := scanner.Err(); errScan != nil {
+					recordAPIResponseError(ctx, e.cfg, errScan)
+					reporter.publishFailure(ctx)
+					out <- cliproxyexecutor.StreamChunk{Err: errScan}
+				} else {
+					reporter.ensurePublished(ctx)
+				}
+			}(httpResp)
+
+			var buffer bytes.Buffer
+			for chunk := range out {
+				if chunk.Err != nil {
+					return resp, chunk.Err
+				}
+				if len(chunk.Payload) > 0 {
+					_, _ = buffer.Write(chunk.Payload)
+					_, _ = buffer.Write([]byte("\n"))
+				}
+			}
+			resp = cliproxyexecutor.Response{Payload: e.convertStreamToNonStream(buffer.Bytes())}
+
+			reporter.publish(ctx, parseAntigravityUsage(resp.Payload))
+			var param any
+			converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, resp.Payload, &param)
+			resp = cliproxyexecutor.Response{Payload: []byte(converted)}
+			reporter.ensurePublished(ctx)
+
+			return resp, nil
+		}
+
+		switch {
+		case lastStatus != 0:
+			sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+			if lastStatus == http.StatusTooManyRequests {
+				if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
 					sErr.retryAfter = retryAfter
 				}
 			}
 			err = sErr
-			return resp, err
+		case lastErr != nil:
+			err = lastErr
+		default:
+			err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
 		}
-
-		out := make(chan cliproxyexecutor.StreamChunk)
-		go func(resp *http.Response) {
-			defer close(out)
-			defer func() {
-				if errClose := resp.Body.Close(); errClose != nil {
-					log.Errorf("antigravity executor: close response body error: %v", errClose)
-				}
-			}()
-			scanner := bufio.NewScanner(resp.Body)
-			scanner.Buffer(nil, streamScannerBuffer)
-			for scanner.Scan() {
-				line := scanner.Bytes()
-				appendAPIResponseChunk(ctx, e.cfg, line)
-
-				// Filter usage metadata for all models
-				// Only retain usage statistics in the terminal chunk
-				line = FilterSSEUsageMetadata(line)
-
-				payload := jsonPayload(line)
-				if payload == nil {
-					continue
-				}
-
-				if detail, ok := parseAntigravityStreamUsage(payload); ok {
-					reporter.publish(ctx, detail)
-				}
-
-				out <- cliproxyexecutor.StreamChunk{Payload: payload}
-			}
-			if errScan := scanner.Err(); errScan != nil {
-				recordAPIResponseError(ctx, e.cfg, errScan)
-				reporter.publishFailure(ctx)
-				out <- cliproxyexecutor.StreamChunk{Err: errScan}
-			} else {
-				reporter.ensurePublished(ctx)
-			}
-		}(httpResp)
-
-		var buffer bytes.Buffer
-		for chunk := range out {
-			if chunk.Err != nil {
-				return resp, chunk.Err
-			}
-			if len(chunk.Payload) > 0 {
-				_, _ = buffer.Write(chunk.Payload)
-				_, _ = buffer.Write([]byte("\n"))
-			}
-		}
-		resp = cliproxyexecutor.Response{Payload: e.convertStreamToNonStream(buffer.Bytes())}
-
-		reporter.publish(ctx, parseAntigravityUsage(resp.Payload))
-		var param any
-		converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, resp.Payload, &param)
-		resp = cliproxyexecutor.Response{Payload: []byte(converted)}
-		reporter.ensurePublished(ctx)
-
-		return resp, nil
+		return resp, err
 	}
 
-	switch {
-	case lastStatus != 0:
-		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
-		if lastStatus == http.StatusTooManyRequests {
-			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
-				sErr.retryAfter = retryAfter
-			}
-		}
-		err = sErr
-	case lastErr != nil:
-		err = lastErr
-	default:
-		err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
-	}
 	return resp, err
 }
 
@@ -635,139 +677,160 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
 
-	var lastStatus int
-	var lastBody []byte
-	var lastErr error
+	attempts := antigravityRetryAttempts(auth, e.cfg)
 
-	for idx, baseURL := range baseURLs {
-		httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, true, opts.Alt, baseURL)
-		if errReq != nil {
-			err = errReq
-			return nil, err
-		}
-		httpResp, errDo := httpClient.Do(httpReq)
-		if errDo != nil {
-			recordAPIResponseError(ctx, e.cfg, errDo)
-			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
-				return nil, errDo
+attemptLoop:
+	for attempt := 0; attempt < attempts; attempt++ {
+		var lastStatus int
+		var lastBody []byte
+		var lastErr error
+
+		for idx, baseURL := range baseURLs {
+			httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, true, opts.Alt, baseURL)
+			if errReq != nil {
+				err = errReq
+				return nil, err
 			}
-			lastStatus = 0
-			lastBody = nil
-			lastErr = errDo
-			if idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
-			}
-			err = errDo
-			return nil, err
-		}
-		recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
-		if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
-			bodyBytes, errRead := io.ReadAll(httpResp.Body)
-			if errClose := httpResp.Body.Close(); errClose != nil {
-				log.Errorf("antigravity executor: close response body error: %v", errClose)
-			}
-			if errRead != nil {
-				recordAPIResponseError(ctx, e.cfg, errRead)
-				if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
-					err = errRead
-					return nil, err
-				}
-				if errCtx := ctx.Err(); errCtx != nil {
-					err = errCtx
-					return nil, err
+			httpResp, errDo := httpClient.Do(httpReq)
+			if errDo != nil {
+				recordAPIResponseError(ctx, e.cfg, errDo)
+				if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+					return nil, errDo
 				}
 				lastStatus = 0
 				lastBody = nil
-				lastErr = errRead
+				lastErr = errDo
 				if idx+1 < len(baseURLs) {
-					log.Debugf("antigravity executor: read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
 					continue
 				}
-				err = errRead
+				err = errDo
 				return nil, err
 			}
-			appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
-			lastStatus = httpResp.StatusCode
-			lastBody = append([]byte(nil), bodyBytes...)
-			lastErr = nil
-			if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
+			recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+			if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+				bodyBytes, errRead := io.ReadAll(httpResp.Body)
+				if errClose := httpResp.Body.Close(); errClose != nil {
+					log.Errorf("antigravity executor: close response body error: %v", errClose)
+				}
+				if errRead != nil {
+					recordAPIResponseError(ctx, e.cfg, errRead)
+					if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
+						err = errRead
+						return nil, err
+					}
+					if errCtx := ctx.Err(); errCtx != nil {
+						err = errCtx
+						return nil, err
+					}
+					lastStatus = 0
+					lastBody = nil
+					lastErr = errRead
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					err = errRead
+					return nil, err
+				}
+				appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
+				lastStatus = httpResp.StatusCode
+				lastBody = append([]byte(nil), bodyBytes...)
+				lastErr = nil
+				if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
+					log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					continue
+				}
+				if antigravityShouldRetryNoCapacity(httpResp.StatusCode, bodyBytes) {
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: no capacity on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					if attempt+1 < attempts {
+						delay := antigravityNoCapacityRetryDelay(attempt)
+						log.Debugf("antigravity executor: no capacity for model %s, retrying in %s (attempt %d/%d)", baseModel, delay, attempt+1, attempts)
+						if errWait := antigravityWait(ctx, delay); errWait != nil {
+							return nil, errWait
+						}
+						continue attemptLoop
+					}
+				}
+				sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+				if httpResp.StatusCode == http.StatusTooManyRequests {
+					if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+						sErr.retryAfter = retryAfter
+					}
+				}
+				err = sErr
+				return nil, err
 			}
-			sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
-			if httpResp.StatusCode == http.StatusTooManyRequests {
-				if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+
+			out := make(chan cliproxyexecutor.StreamChunk)
+			stream = out
+			go func(resp *http.Response) {
+				defer close(out)
+				defer func() {
+					if errClose := resp.Body.Close(); errClose != nil {
+						log.Errorf("antigravity executor: close response body error: %v", errClose)
+					}
+				}()
+				scanner := bufio.NewScanner(resp.Body)
+				scanner.Buffer(nil, streamScannerBuffer)
+				var param any
+				for scanner.Scan() {
+					line := scanner.Bytes()
+					appendAPIResponseChunk(ctx, e.cfg, line)
+
+					// Filter usage metadata for all models
+					// Only retain usage statistics in the terminal chunk
+					line = FilterSSEUsageMetadata(line)
+
+					payload := jsonPayload(line)
+					if payload == nil {
+						continue
+					}
+
+					if detail, ok := parseAntigravityStreamUsage(payload); ok {
+						reporter.publish(ctx, detail)
+					}
+
+					chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bytes.Clone(payload), &param)
+					for i := range chunks {
+						out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
+					}
+				}
+				tail := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, []byte("[DONE]"), &param)
+				for i := range tail {
+					out <- cliproxyexecutor.StreamChunk{Payload: []byte(tail[i])}
+				}
+				if errScan := scanner.Err(); errScan != nil {
+					recordAPIResponseError(ctx, e.cfg, errScan)
+					reporter.publishFailure(ctx)
+					out <- cliproxyexecutor.StreamChunk{Err: errScan}
+				} else {
+					reporter.ensurePublished(ctx)
+				}
+			}(httpResp)
+			return stream, nil
+		}
+
+		switch {
+		case lastStatus != 0:
+			sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+			if lastStatus == http.StatusTooManyRequests {
+				if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
 					sErr.retryAfter = retryAfter
 				}
 			}
 			err = sErr
-			return nil, err
+		case lastErr != nil:
+			err = lastErr
+		default:
+			err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
 		}
-
-		out := make(chan cliproxyexecutor.StreamChunk)
-		stream = out
-		go func(resp *http.Response) {
-			defer close(out)
-			defer func() {
-				if errClose := resp.Body.Close(); errClose != nil {
-					log.Errorf("antigravity executor: close response body error: %v", errClose)
-				}
-			}()
-			scanner := bufio.NewScanner(resp.Body)
-			scanner.Buffer(nil, streamScannerBuffer)
-			var param any
-			for scanner.Scan() {
-				line := scanner.Bytes()
-				appendAPIResponseChunk(ctx, e.cfg, line)
-
-				// Filter usage metadata for all models
-				// Only retain usage statistics in the terminal chunk
-				line = FilterSSEUsageMetadata(line)
-
-				payload := jsonPayload(line)
-				if payload == nil {
-					continue
-				}
-
-				if detail, ok := parseAntigravityStreamUsage(payload); ok {
-					reporter.publish(ctx, detail)
-				}
-
-				chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bytes.Clone(payload), &param)
-				for i := range chunks {
-					out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
-				}
-			}
-			tail := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, []byte("[DONE]"), &param)
-			for i := range tail {
-				out <- cliproxyexecutor.StreamChunk{Payload: []byte(tail[i])}
-			}
-			if errScan := scanner.Err(); errScan != nil {
-				recordAPIResponseError(ctx, e.cfg, errScan)
-				reporter.publishFailure(ctx)
-				out <- cliproxyexecutor.StreamChunk{Err: errScan}
-			} else {
-				reporter.ensurePublished(ctx)
-			}
-		}(httpResp)
-		return stream, nil
+		return nil, err
 	}
 
-	switch {
-	case lastStatus != 0:
-		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
-		if lastStatus == http.StatusTooManyRequests {
-			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
-				sErr.retryAfter = retryAfter
-			}
-		}
-		err = sErr
-	case lastErr != nil:
-		err = lastErr
-	default:
-		err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
-	}
 	return nil, err
 }
 
@@ -1384,14 +1447,70 @@ func resolveUserAgent(auth *cliproxyauth.Auth) string {
 	return defaultAntigravityAgent
 }
 
+func antigravityRetryAttempts(auth *cliproxyauth.Auth, cfg *config.Config) int {
+	retry := 0
+	if cfg != nil {
+		retry = cfg.RequestRetry
+	}
+	if auth != nil {
+		if override, ok := auth.RequestRetryOverride(); ok {
+			retry = override
+		}
+	}
+	if retry < 0 {
+		retry = 0
+	}
+	attempts := retry + 1
+	if attempts < 1 {
+		return 1
+	}
+	return attempts
+}
+
+func antigravityShouldRetryNoCapacity(statusCode int, body []byte) bool {
+	if statusCode != http.StatusServiceUnavailable {
+		return false
+	}
+	if len(body) == 0 {
+		return false
+	}
+	msg := strings.ToLower(string(body))
+	return strings.Contains(msg, "no capacity available")
+}
+
+func antigravityNoCapacityRetryDelay(attempt int) time.Duration {
+	if attempt < 0 {
+		attempt = 0
+	}
+	delay := time.Duration(attempt+1) * 250 * time.Millisecond
+	if delay > 2*time.Second {
+		delay = 2 * time.Second
+	}
+	return delay
+}
+
+func antigravityWait(ctx context.Context, wait time.Duration) error {
+	if wait <= 0 {
+		return nil
+	}
+	timer := time.NewTimer(wait)
+	defer timer.Stop()
+	select {
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-timer.C:
+		return nil
+	}
+}
+
 func antigravityBaseURLFallbackOrder(auth *cliproxyauth.Auth) []string {
 	if base := resolveCustomAntigravityBaseURL(auth); base != "" {
 		return []string{base}
 	}
 	return []string{
-		antigravitySandboxBaseURLDaily,
 		antigravityBaseURLDaily,
-		antigravityBaseURLProd,
+		antigravitySandboxBaseURLDaily,
+		// antigravityBaseURLProd,
 	}
 }
 

internal/runtime/executor/claude_executor.go

@@ -163,7 +163,7 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
 		if errClose := httpResp.Body.Close(); errClose != nil {
 			log.Errorf("response body close error: %v", errClose)
@@ -295,7 +295,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		if errClose := httpResp.Body.Close(); errClose != nil {
 			log.Errorf("response body close error: %v", errClose)
 		}

internal/runtime/executor/codex_executor.go

@@ -150,7 +150,7 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
 		return resp, err
 	}
@@ -265,7 +265,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 			return nil, readErr
 		}
 		appendAPIResponseChunk(ctx, e.cfg, data)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
 		err = statusErr{code: httpResp.StatusCode, msg: string(data)}
 		return nil, err
 	}

internal/runtime/executor/gemini_cli_executor.go

@@ -227,7 +227,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 
 		lastStatus = httpResp.StatusCode
 		lastBody = append([]byte(nil), data...)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
 		if httpResp.StatusCode == 429 {
 			if idx+1 < len(models) {
 				log.Debugf("gemini cli executor: rate limited, retrying with next model: %s", models[idx+1])
@@ -360,7 +360,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 			appendAPIResponseChunk(ctx, e.cfg, data)
 			lastStatus = httpResp.StatusCode
 			lastBody = append([]byte(nil), data...)
-			log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+			logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
 			if httpResp.StatusCode == 429 {
 				if idx+1 < len(models) {
 					log.Debugf("gemini cli executor: rate limited, retrying with next model: %s", models[idx+1])

internal/runtime/executor/gemini_executor.go

@@ -188,7 +188,7 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
 		return resp, err
 	}
@@ -282,7 +282,7 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		if errClose := httpResp.Body.Close(); errClose != nil {
 			log.Errorf("gemini executor: close response body error: %v", errClose)
 		}
@@ -402,7 +402,7 @@ func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	}
 	appendAPIResponseChunk(ctx, e.cfg, data)
 	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
-		log.Debugf("request error, error status: %d, error body: %s", resp.StatusCode, summarizeErrorBody(resp.Header.Get("Content-Type"), data))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", resp.StatusCode, summarizeErrorBody(resp.Header.Get("Content-Type"), data))
 		return cliproxyexecutor.Response{}, statusErr{code: resp.StatusCode, msg: string(data)}
 	}
 

internal/runtime/executor/gemini_vertex_executor.go

@@ -389,7 +389,7 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
 		return resp, err
 	}
@@ -503,7 +503,7 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
 		return resp, err
 	}
@@ -601,7 +601,7 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		if errClose := httpResp.Body.Close(); errClose != nil {
 			log.Errorf("vertex executor: close response body error: %v", errClose)
 		}
@@ -725,7 +725,7 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		if errClose := httpResp.Body.Close(); errClose != nil {
 			log.Errorf("vertex executor: close response body error: %v", errClose)
 		}
@@ -838,7 +838,7 @@ func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %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)
@@ -922,7 +922,7 @@ func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %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)

internal/runtime/executor/iflow_executor.go

@@ -142,7 +142,7 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("iflow request error: status %d body %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		logWithRequestID(ctx).Debugf("request error, error status: %d error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
 		return resp, err
 	}
@@ -244,7 +244,7 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 			log.Errorf("iflow executor: close response body error: %v", errClose)
 		}
 		appendAPIResponseChunk(ctx, e.cfg, data)
-		log.Debugf("iflow streaming error: status %d body %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+		logWithRequestID(ctx).Debugf("request error, error status: %d error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
 		err = statusErr{code: httpResp.StatusCode, msg: string(data)}
 		return nil, err
 	}

internal/runtime/executor/logging_helpers.go

@@ -12,7 +12,10 @@ import (
 
 	"github.com/gin-gonic/gin"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/logging"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	log "github.com/sirupsen/logrus"
+	"github.com/tidwall/gjson"
 )
 
 const (
@@ -332,6 +335,12 @@ func summarizeErrorBody(contentType string, body []byte) string {
 		}
 		return "[html body omitted]"
 	}
+
+	// Try to extract error message from JSON response
+	if message := extractJSONErrorMessage(body); message != "" {
+		return message
+	}
+
 	return string(body)
 }
 
@@ -358,3 +367,25 @@ func extractHTMLTitle(body []byte) string {
 	}
 	return strings.Join(strings.Fields(title), " ")
 }
+
+// extractJSONErrorMessage attempts to extract error.message from JSON error responses
+func extractJSONErrorMessage(body []byte) string {
+	result := gjson.GetBytes(body, "error.message")
+	if result.Exists() && result.String() != "" {
+		return result.String()
+	}
+	return ""
+}
+
+// logWithRequestID returns a logrus Entry with request_id field populated from context.
+// If no request ID is found in context, it returns the standard logger.
+func logWithRequestID(ctx context.Context) *log.Entry {
+	if ctx == nil {
+		return log.NewEntry(log.StandardLogger())
+	}
+	requestID := logging.GetRequestID(ctx)
+	if requestID == "" {
+		return log.NewEntry(log.StandardLogger())
+	}
+	return log.WithField("request_id", requestID)
+}

internal/runtime/executor/openai_compat_executor.go

@@ -146,7 +146,7 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
 		return resp, err
 	}
@@ -239,7 +239,7 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		if errClose := httpResp.Body.Close(); errClose != nil {
 			log.Errorf("openai compat executor: close response body error: %v", errClose)
 		}

internal/runtime/executor/qwen_executor.go

@@ -133,7 +133,7 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
 		return resp, err
 	}
@@ -222,7 +222,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 	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))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		if errClose := httpResp.Body.Close(); errClose != nil {
 			log.Errorf("qwen executor: close response body error: %v", errClose)
 		}

internal/thinking/provider/iflow/apply.go

@@ -1,7 +1,7 @@
-// Package iflow implements thinking configuration for iFlow models (GLM, MiniMax).
+// Package iflow implements thinking configuration for iFlow models.
 //
 // iFlow models use boolean toggle semantics:
-//   - GLM models: chat_template_kwargs.enable_thinking (boolean)
+//   - Models using chat_template_kwargs.enable_thinking (boolean toggle)
 //   - MiniMax models: reasoning_split (boolean)
 //
 // Level values are converted to boolean: none=false, all others=true
@@ -20,6 +20,7 @@ import (
 // Applier implements thinking.ProviderApplier for iFlow models.
 //
 // iFlow-specific behavior:
+//   - enable_thinking toggle models: enable_thinking boolean
 //   - GLM models: enable_thinking boolean + clear_thinking=false
 //   - MiniMax models: reasoning_split boolean
 //   - Level to boolean: none=false, others=true
@@ -61,8 +62,8 @@ func (a *Applier) Apply(body []byte, config thinking.ThinkingConfig, modelInfo *
 		return body, nil
 	}
 
-	if isGLMModel(modelInfo.ID) {
-		return applyGLM(body, config), nil
+	if isEnableThinkingModel(modelInfo.ID) {
+		return applyEnableThinking(body, config, isGLMModel(modelInfo.ID)), nil
 	}
 
 	if isMiniMaxModel(modelInfo.ID) {
@@ -97,7 +98,8 @@ func configToBoolean(config thinking.ThinkingConfig) bool {
 	}
 }
 
-// applyGLM applies thinking configuration for GLM models.
+// applyEnableThinking applies thinking configuration for models that use
+// chat_template_kwargs.enable_thinking format.
 //
 // Output format when enabled:
 //
@@ -107,9 +109,8 @@ func configToBoolean(config thinking.ThinkingConfig) bool {
 //
 //	{"chat_template_kwargs": {"enable_thinking": false}}
 //
-// Note: clear_thinking is only set when thinking is enabled, to preserve
-// thinking output in the response.
-func applyGLM(body []byte, config thinking.ThinkingConfig) []byte {
+// Note: clear_thinking is only set for GLM models when thinking is enabled.
+func applyEnableThinking(body []byte, config thinking.ThinkingConfig, setClearThinking bool) []byte {
 	enableThinking := configToBoolean(config)
 
 	if len(body) == 0 || !gjson.ValidBytes(body) {
@@ -118,8 +119,11 @@ func applyGLM(body []byte, config thinking.ThinkingConfig) []byte {
 
 	result, _ := sjson.SetBytes(body, "chat_template_kwargs.enable_thinking", enableThinking)
 
+	// clear_thinking is a GLM-only knob, strip it for other models.
+	result, _ = sjson.DeleteBytes(result, "chat_template_kwargs.clear_thinking")
+
 	// clear_thinking only needed when thinking is enabled
-	if enableThinking {
+	if enableThinking && setClearThinking {
 		result, _ = sjson.SetBytes(result, "chat_template_kwargs.clear_thinking", false)
 	}
 
@@ -143,8 +147,21 @@ func applyMiniMax(body []byte, config thinking.ThinkingConfig) []byte {
 	return result
 }
 
+// isEnableThinkingModel determines if the model uses chat_template_kwargs.enable_thinking format.
+func isEnableThinkingModel(modelID string) bool {
+	if isGLMModel(modelID) {
+		return true
+	}
+	id := strings.ToLower(modelID)
+	switch id {
+	case "qwen3-max-preview", "deepseek-v3.2", "deepseek-v3.1":
+		return true
+	default:
+		return false
+	}
+}
+
 // isGLMModel determines if the model is a GLM series model.
-// GLM models use chat_template_kwargs.enable_thinking format.
 func isGLMModel(modelID string) bool {
 	return strings.HasPrefix(strings.ToLower(modelID), "glm")
 }

internal/translator/antigravity/gemini/antigravity_gemini_response.go

@@ -41,6 +41,7 @@ func ConvertAntigravityResponseToGemini(ctx context.Context, _ string, originalR
 			responseResult := gjson.GetBytes(rawJSON, "response")
 			if responseResult.Exists() {
 				chunk = []byte(responseResult.Raw)
+				chunk = restoreUsageMetadata(chunk)
 			}
 		} else {
 			chunkTemplate := "[]"
@@ -76,7 +77,8 @@ func ConvertAntigravityResponseToGemini(ctx context.Context, _ string, originalR
 func ConvertAntigravityResponseToGeminiNonStream(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) string {
 	responseResult := gjson.GetBytes(rawJSON, "response")
 	if responseResult.Exists() {
-		return responseResult.Raw
+		chunk := restoreUsageMetadata([]byte(responseResult.Raw))
+		return string(chunk)
 	}
 	return string(rawJSON)
 }
@@ -84,3 +86,15 @@ func ConvertAntigravityResponseToGeminiNonStream(_ context.Context, _ string, or
 func GeminiTokenCount(ctx context.Context, count int64) string {
 	return fmt.Sprintf(`{"totalTokens":%d,"promptTokensDetails":[{"modality":"TEXT","tokenCount":%d}]}`, count, count)
 }
+
+// restoreUsageMetadata renames cpaUsageMetadata back to usageMetadata.
+// The executor renames usageMetadata to cpaUsageMetadata in non-terminal chunks
+// to preserve usage data while hiding it from clients that don't expect it.
+// When returning standard Gemini API format, we must restore the original name.
+func restoreUsageMetadata(chunk []byte) []byte {
+	if cpaUsage := gjson.GetBytes(chunk, "cpaUsageMetadata"); cpaUsage.Exists() {
+		chunk, _ = sjson.SetRawBytes(chunk, "usageMetadata", []byte(cpaUsage.Raw))
+		chunk, _ = sjson.DeleteBytes(chunk, "cpaUsageMetadata")
+	}
+	return chunk
+}

internal/translator/antigravity/gemini/antigravity_gemini_response_test.go

@@ -0,0 +1,95 @@
+package gemini
+
+import (
+	"context"
+	"testing"
+)
+
+func TestRestoreUsageMetadata(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    []byte
+		expected string
+	}{
+		{
+			name:     "cpaUsageMetadata renamed to usageMetadata",
+			input:    []byte(`{"modelVersion":"gemini-3-pro","cpaUsageMetadata":{"promptTokenCount":100,"candidatesTokenCount":200}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100,"candidatesTokenCount":200}}`,
+		},
+		{
+			name:     "no cpaUsageMetadata unchanged",
+			input:    []byte(`{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`,
+		},
+		{
+			name:     "empty input",
+			input:    []byte(`{}`),
+			expected: `{}`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := restoreUsageMetadata(tt.input)
+			if string(result) != tt.expected {
+				t.Errorf("restoreUsageMetadata() = %s, want %s", string(result), tt.expected)
+			}
+		})
+	}
+}
+
+func TestConvertAntigravityResponseToGeminiNonStream(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    []byte
+		expected string
+	}{
+		{
+			name:     "cpaUsageMetadata restored in response",
+			input:    []byte(`{"response":{"modelVersion":"gemini-3-pro","cpaUsageMetadata":{"promptTokenCount":100}}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`,
+		},
+		{
+			name:     "usageMetadata preserved",
+			input:    []byte(`{"response":{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := ConvertAntigravityResponseToGeminiNonStream(context.Background(), "", nil, nil, tt.input, nil)
+			if result != tt.expected {
+				t.Errorf("ConvertAntigravityResponseToGeminiNonStream() = %s, want %s", result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestConvertAntigravityResponseToGeminiStream(t *testing.T) {
+	ctx := context.WithValue(context.Background(), "alt", "")
+
+	tests := []struct {
+		name     string
+		input    []byte
+		expected string
+	}{
+		{
+			name:     "cpaUsageMetadata restored in streaming response",
+			input:    []byte(`data: {"response":{"modelVersion":"gemini-3-pro","cpaUsageMetadata":{"promptTokenCount":100}}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			results := ConvertAntigravityResponseToGemini(ctx, "", nil, nil, tt.input, nil)
+			if len(results) != 1 {
+				t.Fatalf("expected 1 result, got %d", len(results))
+			}
+			if results[0] != tt.expected {
+				t.Errorf("ConvertAntigravityResponseToGemini() = %s, want %s", results[0], tt.expected)
+			}
+		})
+	}
+}

internal/translator/openai/claude/openai_claude_request_test.go

@@ -181,11 +181,11 @@ func TestConvertClaudeRequestToOpenAI_ThinkingToReasoningContent(t *testing.T) {
 			result := ConvertClaudeRequestToOpenAI("test-model", []byte(tt.inputJSON), false)
 			resultJSON := gjson.ParseBytes(result)
 
-			// Find the relevant message (skip system message at index 0)
+			// Find the relevant message
 			messages := resultJSON.Get("messages").Array()
-			if len(messages) < 2 {
+			if len(messages) < 1 {
 				if tt.wantHasReasoningContent || tt.wantHasContent {
-					t.Fatalf("Expected at least 2 messages (system + user/assistant), got %d", len(messages))
+					t.Fatalf("Expected at least 1 message, got %d", len(messages))
 				}
 				return
 			}
@@ -272,15 +272,15 @@ func TestConvertClaudeRequestToOpenAI_ThinkingOnlyMessagePreserved(t *testing.T)
 
 	messages := resultJSON.Get("messages").Array()
 
-	// Should have: system (auto-added) + user + assistant (thinking-only) + user = 4 messages
-	if len(messages) != 4 {
-		t.Fatalf("Expected 4 messages, got %d. Messages: %v", len(messages), resultJSON.Get("messages").Raw)
+	// Should have: user + assistant (thinking-only) + user = 3 messages
+	if len(messages) != 3 {
+		t.Fatalf("Expected 3 messages, got %d. Messages: %v", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	// Check the assistant message (index 2) has reasoning_content
-	assistantMsg := messages[2]
+	// Check the assistant message (index 1) has reasoning_content
+	assistantMsg := messages[1]
 	if assistantMsg.Get("role").String() != "assistant" {
-		t.Errorf("Expected message[2] to be assistant, got %s", assistantMsg.Get("role").String())
+		t.Errorf("Expected message[1] to be assistant, got %s", assistantMsg.Get("role").String())
 	}
 
 	if !assistantMsg.Get("reasoning_content").Exists() {
@@ -292,6 +292,104 @@ func TestConvertClaudeRequestToOpenAI_ThinkingOnlyMessagePreserved(t *testing.T)
 	}
 }
 
+func TestConvertClaudeRequestToOpenAI_SystemMessageScenarios(t *testing.T) {
+	tests := []struct {
+		name        string
+		inputJSON   string
+		wantHasSys  bool
+		wantSysText string
+	}{
+		{
+			name: "No system field",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys: false,
+		},
+		{
+			name: "Empty string system field",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"system": "",
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys: false,
+		},
+		{
+			name: "String system field",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"system": "Be helpful",
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys:  true,
+			wantSysText: "Be helpful",
+		},
+		{
+			name: "Array system field with text",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"system": [{"type": "text", "text": "Array system"}],
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys:  true,
+			wantSysText: "Array system",
+		},
+		{
+			name: "Array system field with multiple text blocks",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"system": [
+					{"type": "text", "text": "Block 1"},
+					{"type": "text", "text": "Block 2"}
+				],
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys:  true,
+			wantSysText: "Block 2", // We will update the test logic to check all blocks or specifically the second one
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := ConvertClaudeRequestToOpenAI("test-model", []byte(tt.inputJSON), false)
+			resultJSON := gjson.ParseBytes(result)
+			messages := resultJSON.Get("messages").Array()
+
+			hasSys := false
+			var sysMsg gjson.Result
+			if len(messages) > 0 && messages[0].Get("role").String() == "system" {
+				hasSys = true
+				sysMsg = messages[0]
+			}
+
+			if hasSys != tt.wantHasSys {
+				t.Errorf("got hasSystem = %v, want %v", hasSys, tt.wantHasSys)
+			}
+
+			if tt.wantHasSys {
+				// Check content - it could be string or array in OpenAI
+				content := sysMsg.Get("content")
+				var gotText string
+				if content.IsArray() {
+					arr := content.Array()
+					if len(arr) > 0 {
+						// Get the last element's text for validation
+						gotText = arr[len(arr)-1].Get("text").String()
+					}
+				} else {
+					gotText = content.String()
+				}
+
+				if tt.wantSysText != "" && gotText != tt.wantSysText {
+					t.Errorf("got system text = %q, want %q", gotText, tt.wantSysText)
+				}
+			}
+		})
+	}
+}
+
 func TestConvertClaudeRequestToOpenAI_ToolResultOrderAndContent(t *testing.T) {
 	inputJSON := `{
 		"model": "claude-3-opus",
@@ -318,39 +416,35 @@ func TestConvertClaudeRequestToOpenAI_ToolResultOrderAndContent(t *testing.T) {
 	messages := resultJSON.Get("messages").Array()
 
 	// OpenAI requires: tool messages MUST immediately follow assistant(tool_calls).
-	// Correct order: system + assistant(tool_calls) + tool(result) + user(before+after)
-	if len(messages) != 4 {
-		t.Fatalf("Expected 4 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
+	// Correct order: assistant(tool_calls) + tool(result) + user(before+after)
+	if len(messages) != 3 {
+		t.Fatalf("Expected 3 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	if messages[0].Get("role").String() != "system" {
-		t.Fatalf("Expected messages[0] to be system, got %s", messages[0].Get("role").String())
-	}
-
-	if messages[1].Get("role").String() != "assistant" || !messages[1].Get("tool_calls").Exists() {
-		t.Fatalf("Expected messages[1] to be assistant tool_calls, got %s: %s", messages[1].Get("role").String(), messages[1].Raw)
+	if messages[0].Get("role").String() != "assistant" || !messages[0].Get("tool_calls").Exists() {
+		t.Fatalf("Expected messages[0] to be assistant tool_calls, got %s: %s", messages[0].Get("role").String(), messages[0].Raw)
 	}
 
 	// tool message MUST immediately follow assistant(tool_calls) per OpenAI spec
-	if messages[2].Get("role").String() != "tool" {
-		t.Fatalf("Expected messages[2] to be tool (must follow tool_calls), got %s", messages[2].Get("role").String())
+	if messages[1].Get("role").String() != "tool" {
+		t.Fatalf("Expected messages[1] to be tool (must follow tool_calls), got %s", messages[1].Get("role").String())
 	}
-	if got := messages[2].Get("tool_call_id").String(); got != "call_1" {
+	if got := messages[1].Get("tool_call_id").String(); got != "call_1" {
 		t.Fatalf("Expected tool_call_id %q, got %q", "call_1", got)
 	}
-	if got := messages[2].Get("content").String(); got != "tool ok" {
+	if got := messages[1].Get("content").String(); got != "tool ok" {
 		t.Fatalf("Expected tool content %q, got %q", "tool ok", got)
 	}
 
 	// User message comes after tool message
-	if messages[3].Get("role").String() != "user" {
-		t.Fatalf("Expected messages[3] to be user, got %s", messages[3].Get("role").String())
+	if messages[2].Get("role").String() != "user" {
+		t.Fatalf("Expected messages[2] to be user, got %s", messages[2].Get("role").String())
 	}
 	// User message should contain both "before" and "after" text
-	if got := messages[3].Get("content.0.text").String(); got != "before" {
+	if got := messages[2].Get("content.0.text").String(); got != "before" {
 		t.Fatalf("Expected user text[0] %q, got %q", "before", got)
 	}
-	if got := messages[3].Get("content.1.text").String(); got != "after" {
+	if got := messages[2].Get("content.1.text").String(); got != "after" {
 		t.Fatalf("Expected user text[1] %q, got %q", "after", got)
 	}
 }
@@ -378,16 +472,16 @@ func TestConvertClaudeRequestToOpenAI_ToolResultObjectContent(t *testing.T) {
 	resultJSON := gjson.ParseBytes(result)
 	messages := resultJSON.Get("messages").Array()
 
-	// system + assistant(tool_calls) + tool(result)
-	if len(messages) != 3 {
-		t.Fatalf("Expected 3 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
+	// assistant(tool_calls) + tool(result)
+	if len(messages) != 2 {
+		t.Fatalf("Expected 2 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	if messages[2].Get("role").String() != "tool" {
-		t.Fatalf("Expected messages[2] to be tool, got %s", messages[2].Get("role").String())
+	if messages[1].Get("role").String() != "tool" {
+		t.Fatalf("Expected messages[1] to be tool, got %s", messages[1].Get("role").String())
 	}
 
-	toolContent := messages[2].Get("content").String()
+	toolContent := messages[1].Get("content").String()
 	parsed := gjson.Parse(toolContent)
 	if parsed.Get("foo").String() != "bar" {
 		t.Fatalf("Expected tool content JSON foo=bar, got %q", toolContent)
@@ -414,18 +508,14 @@ func TestConvertClaudeRequestToOpenAI_AssistantTextToolUseTextOrder(t *testing.T
 	messages := resultJSON.Get("messages").Array()
 
 	// New behavior: content + tool_calls unified in single assistant message
-	// Expect: system + assistant(content[pre,post] + tool_calls)
-	if len(messages) != 2 {
-		t.Fatalf("Expected 2 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
+	// Expect: assistant(content[pre,post] + tool_calls)
+	if len(messages) != 1 {
+		t.Fatalf("Expected 1 message, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	if messages[0].Get("role").String() != "system" {
-		t.Fatalf("Expected messages[0] to be system, got %s", messages[0].Get("role").String())
-	}
-
-	assistantMsg := messages[1]
+	assistantMsg := messages[0]
 	if assistantMsg.Get("role").String() != "assistant" {
-		t.Fatalf("Expected messages[1] to be assistant, got %s", assistantMsg.Get("role").String())
+		t.Fatalf("Expected messages[0] to be assistant, got %s", assistantMsg.Get("role").String())
 	}
 
 	// Should have both content and tool_calls in same message
@@ -470,14 +560,14 @@ func TestConvertClaudeRequestToOpenAI_AssistantThinkingToolUseThinkingSplit(t *t
 	messages := resultJSON.Get("messages").Array()
 
 	// New behavior: all content, thinking, and tool_calls unified in single assistant message
-	// Expect: system + assistant(content[pre,post] + tool_calls + reasoning_content[t1+t2])
-	if len(messages) != 2 {
-		t.Fatalf("Expected 2 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
+	// Expect: assistant(content[pre,post] + tool_calls + reasoning_content[t1+t2])
+	if len(messages) != 1 {
+		t.Fatalf("Expected 1 message, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	assistantMsg := messages[1]
+	assistantMsg := messages[0]
 	if assistantMsg.Get("role").String() != "assistant" {
-		t.Fatalf("Expected messages[1] to be assistant, got %s", assistantMsg.Get("role").String())
+		t.Fatalf("Expected messages[0] to be assistant, got %s", assistantMsg.Get("role").String())
 	}
 
 	// Should have content with both pre and post

internal/util/gemini_schema.go

@@ -175,7 +175,7 @@ func convertConstToEnum(jsonStr string) string {
 	return jsonStr
 }
 
-// convertEnumValuesToStrings ensures all enum values are strings.
+// convertEnumValuesToStrings ensures all enum values are strings and the schema type is set to string.
 // Gemini API requires enum values to be of type string, not numbers or booleans.
 func convertEnumValuesToStrings(jsonStr string) string {
 	for _, p := range findPaths(jsonStr, "enum") {
@@ -185,19 +185,15 @@ func convertEnumValuesToStrings(jsonStr string) string {
 		}
 
 		var stringVals []string
-		needsConversion := false
 		for _, item := range arr.Array() {
-			// Check if any value is not a string
-			if item.Type != gjson.String {
-				needsConversion = true
-			}
 			stringVals = append(stringVals, item.String())
 		}
 
-		// Only update if we found non-string values
-		if needsConversion {
-			jsonStr, _ = sjson.Set(jsonStr, p, stringVals)
-		}
+		// Always update enum values to strings and set type to "string"
+		// This ensures compatibility with Antigravity Gemini which only allows enum for STRING type
+		jsonStr, _ = sjson.Set(jsonStr, p, stringVals)
+		parentPath := trimSuffix(p, ".enum")
+		jsonStr, _ = sjson.Set(jsonStr, joinPath(parentPath, "type"), "string")
 	}
 	return jsonStr
 }

internal/watcher/synthesizer/file.go

@@ -86,12 +86,19 @@ func (s *FileSynthesizer) Synthesize(ctx *SynthesisContext) ([]*coreauth.Auth, e
 			}
 		}
 
+		disabled, _ := metadata["disabled"].(bool)
+		status := coreauth.StatusActive
+		if disabled {
+			status = coreauth.StatusDisabled
+		}
+
 		a := &coreauth.Auth{
 			ID:       id,
 			Provider: provider,
 			Label:    label,
 			Prefix:   prefix,
-			Status:   coreauth.StatusActive,
+			Status:   status,
+			Disabled: disabled,
 			Attributes: map[string]string{
 				"source": full,
 				"path":   full,
@@ -167,6 +174,16 @@ func SynthesizeGeminiVirtualAuths(primary *coreauth.Auth, metadata map[string]an
 			"virtual_parent_id": primary.ID,
 			"type":              metadata["type"],
 		}
+		if v, ok := metadata["disable_cooling"]; ok {
+			metadataCopy["disable_cooling"] = v
+		} else if v, ok := metadata["disable-cooling"]; ok {
+			metadataCopy["disable_cooling"] = v
+		}
+		if v, ok := metadata["request_retry"]; ok {
+			metadataCopy["request_retry"] = v
+		} else if v, ok := metadata["request-retry"]; ok {
+			metadataCopy["request_retry"] = v
+		}
 		proxy := strings.TrimSpace(primary.ProxyURL)
 		if proxy != "" {
 			metadataCopy["proxy_url"] = proxy

internal/watcher/synthesizer/file_test.go

@@ -69,10 +69,12 @@ func TestFileSynthesizer_Synthesize_ValidAuthFile(t *testing.T) {
 
 	// Create a valid auth file
 	authData := map[string]any{
-		"type":      "claude",
-		"email":     "test@example.com",
-		"proxy_url": "http://proxy.local",
-		"prefix":    "test-prefix",
+		"type":            "claude",
+		"email":           "test@example.com",
+		"proxy_url":       "http://proxy.local",
+		"prefix":          "test-prefix",
+		"disable_cooling": true,
+		"request_retry":   2,
 	}
 	data, _ := json.Marshal(authData)
 	err := os.WriteFile(filepath.Join(tempDir, "claude-auth.json"), data, 0644)
@@ -108,6 +110,12 @@ func TestFileSynthesizer_Synthesize_ValidAuthFile(t *testing.T) {
 	if auths[0].ProxyURL != "http://proxy.local" {
 		t.Errorf("expected proxy_url http://proxy.local, got %s", auths[0].ProxyURL)
 	}
+	if v, ok := auths[0].Metadata["disable_cooling"].(bool); !ok || !v {
+		t.Errorf("expected disable_cooling true, got %v", auths[0].Metadata["disable_cooling"])
+	}
+	if v, ok := auths[0].Metadata["request_retry"].(float64); !ok || int(v) != 2 {
+		t.Errorf("expected request_retry 2, got %v", auths[0].Metadata["request_retry"])
+	}
 	if auths[0].Status != coreauth.StatusActive {
 		t.Errorf("expected status active, got %s", auths[0].Status)
 	}
@@ -336,9 +344,11 @@ func TestSynthesizeGeminiVirtualAuths_MultiProject(t *testing.T) {
 		},
 	}
 	metadata := map[string]any{
-		"project_id": "project-a, project-b, project-c",
-		"email":      "test@example.com",
-		"type":       "gemini",
+		"project_id":      "project-a, project-b, project-c",
+		"email":           "test@example.com",
+		"type":            "gemini",
+		"request_retry":   2,
+		"disable_cooling": true,
 	}
 
 	virtuals := SynthesizeGeminiVirtualAuths(primary, metadata, now)
@@ -376,6 +386,12 @@ func TestSynthesizeGeminiVirtualAuths_MultiProject(t *testing.T) {
 		if v.ProxyURL != "http://proxy.local" {
 			t.Errorf("expected proxy_url http://proxy.local, got %s", v.ProxyURL)
 		}
+		if vv, ok := v.Metadata["disable_cooling"].(bool); !ok || !vv {
+			t.Errorf("expected disable_cooling true, got %v", v.Metadata["disable_cooling"])
+		}
+		if vv, ok := v.Metadata["request_retry"].(int); !ok || vv != 2 {
+			t.Errorf("expected request_retry 2, got %v", v.Metadata["request_retry"])
+		}
 		if v.Attributes["runtime_only"] != "true" {
 			t.Error("expected runtime_only=true")
 		}

internal/wsrelay/http.go

@@ -124,32 +124,47 @@ func (m *Manager) Stream(ctx context.Context, provider string, req *HTTPRequest)
 	out := make(chan StreamEvent)
 	go func() {
 		defer close(out)
+		send := func(ev StreamEvent) bool {
+			if ctx == nil {
+				out <- ev
+				return true
+			}
+			select {
+			case <-ctx.Done():
+				return false
+			case out <- ev:
+				return true
+			}
+		}
 		for {
 			select {
 			case <-ctx.Done():
-				out <- StreamEvent{Err: ctx.Err()}
 				return
 			case msg, ok := <-respCh:
 				if !ok {
-					out <- StreamEvent{Err: errors.New("wsrelay: stream closed")}
+					_ = send(StreamEvent{Err: errors.New("wsrelay: stream closed")})
 					return
 				}
 				switch msg.Type {
 				case MessageTypeStreamStart:
 					resp := decodeResponse(msg.Payload)
-					out <- StreamEvent{Type: MessageTypeStreamStart, Status: resp.Status, Headers: resp.Headers}
+					if okSend := send(StreamEvent{Type: MessageTypeStreamStart, Status: resp.Status, Headers: resp.Headers}); !okSend {
+						return
+					}
 				case MessageTypeStreamChunk:
 					chunk := decodeChunk(msg.Payload)
-					out <- StreamEvent{Type: MessageTypeStreamChunk, Payload: chunk}
+					if okSend := send(StreamEvent{Type: MessageTypeStreamChunk, Payload: chunk}); !okSend {
+						return
+					}
 				case MessageTypeStreamEnd:
-					out <- StreamEvent{Type: MessageTypeStreamEnd}
+					_ = send(StreamEvent{Type: MessageTypeStreamEnd})
 					return
 				case MessageTypeError:
-					out <- StreamEvent{Type: MessageTypeError, Err: decodeError(msg.Payload)}
+					_ = send(StreamEvent{Type: MessageTypeError, Err: decodeError(msg.Payload)})
 					return
 				case MessageTypeHTTPResp:
 					resp := decodeResponse(msg.Payload)
-					out <- StreamEvent{Type: MessageTypeHTTPResp, Status: resp.Status, Headers: resp.Headers, Payload: resp.Body}
+					_ = send(StreamEvent{Type: MessageTypeHTTPResp, Status: resp.Status, Headers: resp.Headers, Payload: resp.Body})
 					return
 				default:
 				}

sdk/api/handlers/gemini/gemini-cli_handlers.go

@@ -124,6 +124,7 @@ func (h *GeminiCLIAPIHandler) CLIHandler(c *gin.Context) {
 			log.Errorf("Failed to read response body: %v", err)
 			return
 		}
+		c.Set("API_RESPONSE_TIMESTAMP", time.Now())
 		_, _ = c.Writer.Write(output)
 		c.Set("API_RESPONSE", output)
 	}

sdk/api/handlers/handlers.go

@@ -361,6 +361,11 @@ func appendAPIResponse(c *gin.Context, data []byte) {
 		return
 	}
 
+	// Capture timestamp on first API response
+	if _, exists := c.Get("API_RESPONSE_TIMESTAMP"); !exists {
+		c.Set("API_RESPONSE_TIMESTAMP", time.Now())
+	}
+
 	if existing, exists := c.Get("API_RESPONSE"); exists {
 		if existingBytes, ok := existing.([]byte); ok && len(existingBytes) > 0 {
 			combined := make([]byte, 0, len(existingBytes)+len(data)+1)
@@ -506,6 +511,32 @@ func (h *BaseAPIHandler) ExecuteStreamWithAuthManager(ctx context.Context, handl
 		bootstrapRetries := 0
 		maxBootstrapRetries := StreamingBootstrapRetries(h.Cfg)
 
+		sendErr := func(msg *interfaces.ErrorMessage) bool {
+			if ctx == nil {
+				errChan <- msg
+				return true
+			}
+			select {
+			case <-ctx.Done():
+				return false
+			case errChan <- msg:
+				return true
+			}
+		}
+
+		sendData := func(chunk []byte) bool {
+			if ctx == nil {
+				dataChan <- chunk
+				return true
+			}
+			select {
+			case <-ctx.Done():
+				return false
+			case dataChan <- chunk:
+				return true
+			}
+		}
+
 		bootstrapEligible := func(err error) bool {
 			status := statusFromError(err)
 			if status == 0 {
@@ -565,12 +596,14 @@ func (h *BaseAPIHandler) ExecuteStreamWithAuthManager(ctx context.Context, handl
 							addon = hdr.Clone()
 						}
 					}
-					errChan <- &interfaces.ErrorMessage{StatusCode: status, Error: streamErr, Addon: addon}
+					_ = sendErr(&interfaces.ErrorMessage{StatusCode: status, Error: streamErr, Addon: addon})
 					return
 				}
 				if len(chunk.Payload) > 0 {
 					sentPayload = true
-					dataChan <- cloneBytes(chunk.Payload)
+					if okSendData := sendData(cloneBytes(chunk.Payload)); !okSendData {
+						return
+					}
 				}
 			}
 		}

sdk/api/handlers/handlers_stream_bootstrap_test.go

@@ -70,6 +70,58 @@ func (e *failOnceStreamExecutor) Calls() int {
 	return e.calls
 }
 
+type payloadThenErrorStreamExecutor struct {
+	mu    sync.Mutex
+	calls int
+}
+
+func (e *payloadThenErrorStreamExecutor) Identifier() string { return "codex" }
+
+func (e *payloadThenErrorStreamExecutor) Execute(context.Context, *coreauth.Auth, coreexecutor.Request, coreexecutor.Options) (coreexecutor.Response, error) {
+	return coreexecutor.Response{}, &coreauth.Error{Code: "not_implemented", Message: "Execute not implemented"}
+}
+
+func (e *payloadThenErrorStreamExecutor) ExecuteStream(context.Context, *coreauth.Auth, coreexecutor.Request, coreexecutor.Options) (<-chan coreexecutor.StreamChunk, error) {
+	e.mu.Lock()
+	e.calls++
+	e.mu.Unlock()
+
+	ch := make(chan coreexecutor.StreamChunk, 2)
+	ch <- coreexecutor.StreamChunk{Payload: []byte("partial")}
+	ch <- coreexecutor.StreamChunk{
+		Err: &coreauth.Error{
+			Code:       "upstream_closed",
+			Message:    "upstream closed",
+			Retryable:  false,
+			HTTPStatus: http.StatusBadGateway,
+		},
+	}
+	close(ch)
+	return ch, nil
+}
+
+func (e *payloadThenErrorStreamExecutor) Refresh(ctx context.Context, auth *coreauth.Auth) (*coreauth.Auth, error) {
+	return auth, nil
+}
+
+func (e *payloadThenErrorStreamExecutor) CountTokens(context.Context, *coreauth.Auth, coreexecutor.Request, coreexecutor.Options) (coreexecutor.Response, error) {
+	return coreexecutor.Response{}, &coreauth.Error{Code: "not_implemented", Message: "CountTokens not implemented"}
+}
+
+func (e *payloadThenErrorStreamExecutor) HttpRequest(ctx context.Context, auth *coreauth.Auth, req *http.Request) (*http.Response, error) {
+	return nil, &coreauth.Error{
+		Code:       "not_implemented",
+		Message:    "HttpRequest not implemented",
+		HTTPStatus: http.StatusNotImplemented,
+	}
+}
+
+func (e *payloadThenErrorStreamExecutor) Calls() int {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	return e.calls
+}
+
 func TestExecuteStreamWithAuthManager_RetriesBeforeFirstByte(t *testing.T) {
 	executor := &failOnceStreamExecutor{}
 	manager := coreauth.NewManager(nil, nil, nil)
@@ -130,3 +182,73 @@ func TestExecuteStreamWithAuthManager_RetriesBeforeFirstByte(t *testing.T) {
 		t.Fatalf("expected 2 stream attempts, got %d", executor.Calls())
 	}
 }
+
+func TestExecuteStreamWithAuthManager_DoesNotRetryAfterFirstByte(t *testing.T) {
+	executor := &payloadThenErrorStreamExecutor{}
+	manager := coreauth.NewManager(nil, nil, nil)
+	manager.RegisterExecutor(executor)
+
+	auth1 := &coreauth.Auth{
+		ID:       "auth1",
+		Provider: "codex",
+		Status:   coreauth.StatusActive,
+		Metadata: map[string]any{"email": "test1@example.com"},
+	}
+	if _, err := manager.Register(context.Background(), auth1); err != nil {
+		t.Fatalf("manager.Register(auth1): %v", err)
+	}
+
+	auth2 := &coreauth.Auth{
+		ID:       "auth2",
+		Provider: "codex",
+		Status:   coreauth.StatusActive,
+		Metadata: map[string]any{"email": "test2@example.com"},
+	}
+	if _, err := manager.Register(context.Background(), auth2); err != nil {
+		t.Fatalf("manager.Register(auth2): %v", err)
+	}
+
+	registry.GetGlobalRegistry().RegisterClient(auth1.ID, auth1.Provider, []*registry.ModelInfo{{ID: "test-model"}})
+	registry.GetGlobalRegistry().RegisterClient(auth2.ID, auth2.Provider, []*registry.ModelInfo{{ID: "test-model"}})
+	t.Cleanup(func() {
+		registry.GetGlobalRegistry().UnregisterClient(auth1.ID)
+		registry.GetGlobalRegistry().UnregisterClient(auth2.ID)
+	})
+
+	handler := NewBaseAPIHandlers(&sdkconfig.SDKConfig{
+		Streaming: sdkconfig.StreamingConfig{
+			BootstrapRetries: 1,
+		},
+	}, manager)
+	dataChan, errChan := handler.ExecuteStreamWithAuthManager(context.Background(), "openai", "test-model", []byte(`{"model":"test-model"}`), "")
+	if dataChan == nil || errChan == nil {
+		t.Fatalf("expected non-nil channels")
+	}
+
+	var got []byte
+	for chunk := range dataChan {
+		got = append(got, chunk...)
+	}
+
+	var gotErr error
+	var gotStatus int
+	for msg := range errChan {
+		if msg != nil && msg.Error != nil {
+			gotErr = msg.Error
+			gotStatus = msg.StatusCode
+		}
+	}
+
+	if string(got) != "partial" {
+		t.Fatalf("expected payload partial, got %q", string(got))
+	}
+	if gotErr == nil {
+		t.Fatalf("expected terminal error, got nil")
+	}
+	if gotStatus != http.StatusBadGateway {
+		t.Fatalf("expected status %d, got %d", http.StatusBadGateway, gotStatus)
+	}
+	if executor.Calls() != 1 {
+		t.Fatalf("expected 1 stream attempt, got %d", executor.Calls())
+	}
+}

sdk/auth/filestore.go

@@ -68,14 +68,13 @@ func (s *FileTokenStore) Save(ctx context.Context, auth *cliproxyauth.Auth) (str
 			return "", err
 		}
 	case auth.Metadata != nil:
+		auth.Metadata["disabled"] = auth.Disabled
 		raw, errMarshal := json.Marshal(auth.Metadata)
 		if errMarshal != nil {
 			return "", fmt.Errorf("auth filestore: marshal metadata failed: %w", errMarshal)
 		}
 		if existing, errRead := os.ReadFile(path); errRead == nil {
-			// Use metadataEqualIgnoringTimestamps to skip writes when only timestamp fields change.
-			// This prevents the token refresh loop caused by timestamp/expired/expires_in changes.
-			if metadataEqualIgnoringTimestamps(existing, raw, auth.Provider) {
+			if jsonEqual(existing, raw) {
 				return path, nil
 			}
 			file, errOpen := os.OpenFile(path, os.O_WRONLY|os.O_TRUNC, 0o600)
@@ -216,12 +215,18 @@ func (s *FileTokenStore) readAuthFile(path, baseDir string) (*cliproxyauth.Auth,
 		return nil, fmt.Errorf("stat file: %w", err)
 	}
 	id := s.idFor(path, baseDir)
+	disabled, _ := metadata["disabled"].(bool)
+	status := cliproxyauth.StatusActive
+	if disabled {
+		status = cliproxyauth.StatusDisabled
+	}
 	auth := &cliproxyauth.Auth{
 		ID:               id,
 		Provider:         provider,
 		FileName:         id,
 		Label:            s.labelFor(metadata),
-		Status:           cliproxyauth.StatusActive,
+		Status:           status,
+		Disabled:         disabled,
 		Attributes:       map[string]string{"path": path},
 		Metadata:         metadata,
 		CreatedAt:        info.ModTime(),
@@ -299,8 +304,7 @@ func (s *FileTokenStore) baseDirSnapshot() string {
 	return s.baseDir
 }
 
-// DEPRECATED: Use metadataEqualIgnoringTimestamps for comparing auth metadata.
-// This function is kept for backward compatibility but can cause refresh loops.
+// jsonEqual compares two JSON blobs by parsing them into Go objects and deep comparing.
 func jsonEqual(a, b []byte) bool {
 	var objA any
 	var objB any
@@ -313,41 +317,6 @@ func jsonEqual(a, b []byte) bool {
 	return deepEqualJSON(objA, objB)
 }
 
-// metadataEqualIgnoringTimestamps compares two metadata JSON blobs,
-// ignoring fields that change on every refresh but don't affect functionality.
-// This prevents unnecessary file writes that would trigger watcher events and
-// create refresh loops.
-// The provider parameter controls whether access_token is ignored: providers like
-// Google OAuth (gemini, gemini-cli) can re-fetch tokens when needed, while others
-// like iFlow require the refreshed token to be persisted.
-func metadataEqualIgnoringTimestamps(a, b []byte, provider string) bool {
-	var objA, objB map[string]any
-	if err := json.Unmarshal(a, &objA); err != nil {
-		return false
-	}
-	if err := json.Unmarshal(b, &objB); err != nil {
-		return false
-	}
-
-	// Fields to ignore: these change on every refresh but don't affect authentication logic.
-	// - timestamp, expired, expires_in, last_refresh: time-related fields that change on refresh
-	ignoredFields := []string{"timestamp", "expired", "expires_in", "last_refresh"}
-
-	// For providers that can re-fetch tokens when needed (e.g., Google OAuth),
-	// we ignore access_token to avoid unnecessary file writes.
-	switch provider {
-	case "gemini", "gemini-cli", "antigravity":
-		ignoredFields = append(ignoredFields, "access_token")
-	}
-
-	for _, field := range ignoredFields {
-		delete(objA, field)
-		delete(objB, field)
-	}
-
-	return deepEqualJSON(objA, objB)
-}
-
 func deepEqualJSON(a, b any) bool {
 	switch valA := a.(type) {
 	case map[string]any:

sdk/cliproxy/auth/conductor.go

@@ -61,6 +61,15 @@ func SetQuotaCooldownDisabled(disable bool) {
 	quotaCooldownDisabled.Store(disable)
 }
 
+func quotaCooldownDisabledForAuth(auth *Auth) bool {
+	if auth != nil {
+		if override, ok := auth.DisableCoolingOverride(); ok {
+			return override
+		}
+	}
+	return quotaCooldownDisabled.Load()
+}
+
 // Result captures execution outcome used to adjust auth state.
 type Result struct {
 	// AuthID references the auth that produced this result.
@@ -468,20 +477,16 @@ func (m *Manager) Execute(ctx context.Context, providers []string, req cliproxye
 		return cliproxyexecutor.Response{}, &Error{Code: "provider_not_found", Message: "no provider supplied"}
 	}
 
-	retryTimes, maxWait := m.retrySettings()
-	attempts := retryTimes + 1
-	if attempts < 1 {
-		attempts = 1
-	}
+	_, maxWait := m.retrySettings()
 
 	var lastErr error
-	for attempt := 0; attempt < attempts; attempt++ {
+	for attempt := 0; ; attempt++ {
 		resp, errExec := m.executeMixedOnce(ctx, normalized, req, opts)
 		if errExec == nil {
 			return resp, nil
 		}
 		lastErr = errExec
-		wait, shouldRetry := m.shouldRetryAfterError(errExec, attempt, attempts, normalized, req.Model, maxWait)
+		wait, shouldRetry := m.shouldRetryAfterError(errExec, attempt, normalized, req.Model, maxWait)
 		if !shouldRetry {
 			break
 		}
@@ -503,20 +508,16 @@ func (m *Manager) ExecuteCount(ctx context.Context, providers []string, req clip
 		return cliproxyexecutor.Response{}, &Error{Code: "provider_not_found", Message: "no provider supplied"}
 	}
 
-	retryTimes, maxWait := m.retrySettings()
-	attempts := retryTimes + 1
-	if attempts < 1 {
-		attempts = 1
-	}
+	_, maxWait := m.retrySettings()
 
 	var lastErr error
-	for attempt := 0; attempt < attempts; attempt++ {
+	for attempt := 0; ; attempt++ {
 		resp, errExec := m.executeCountMixedOnce(ctx, normalized, req, opts)
 		if errExec == nil {
 			return resp, nil
 		}
 		lastErr = errExec
-		wait, shouldRetry := m.shouldRetryAfterError(errExec, attempt, attempts, normalized, req.Model, maxWait)
+		wait, shouldRetry := m.shouldRetryAfterError(errExec, attempt, normalized, req.Model, maxWait)
 		if !shouldRetry {
 			break
 		}
@@ -538,20 +539,16 @@ func (m *Manager) ExecuteStream(ctx context.Context, providers []string, req cli
 		return nil, &Error{Code: "provider_not_found", Message: "no provider supplied"}
 	}
 
-	retryTimes, maxWait := m.retrySettings()
-	attempts := retryTimes + 1
-	if attempts < 1 {
-		attempts = 1
-	}
+	_, maxWait := m.retrySettings()
 
 	var lastErr error
-	for attempt := 0; attempt < attempts; attempt++ {
+	for attempt := 0; ; attempt++ {
 		chunks, errStream := m.executeStreamMixedOnce(ctx, normalized, req, opts)
 		if errStream == nil {
 			return chunks, nil
 		}
 		lastErr = errStream
-		wait, shouldRetry := m.shouldRetryAfterError(errStream, attempt, attempts, normalized, req.Model, maxWait)
+		wait, shouldRetry := m.shouldRetryAfterError(errStream, attempt, normalized, req.Model, maxWait)
 		if !shouldRetry {
 			break
 		}
@@ -721,6 +718,7 @@ func (m *Manager) executeStreamMixedOnce(ctx context.Context, providers []string
 		go func(streamCtx context.Context, streamAuth *Auth, streamProvider string, streamChunks <-chan cliproxyexecutor.StreamChunk) {
 			defer close(out)
 			var failed bool
+			forward := true
 			for chunk := range streamChunks {
 				if chunk.Err != nil && !failed {
 					failed = true
@@ -731,7 +729,18 @@ func (m *Manager) executeStreamMixedOnce(ctx context.Context, providers []string
 					}
 					m.MarkResult(streamCtx, Result{AuthID: streamAuth.ID, Provider: streamProvider, Model: routeModel, Success: false, Error: rerr})
 				}
-				out <- chunk
+				if !forward {
+					continue
+				}
+				if streamCtx == nil {
+					out <- chunk
+					continue
+				}
+				select {
+				case <-streamCtx.Done():
+					forward = false
+				case out <- chunk:
+				}
 			}
 			if !failed {
 				m.MarkResult(streamCtx, Result{AuthID: streamAuth.ID, Provider: streamProvider, Model: routeModel, Success: true})
@@ -1034,11 +1043,15 @@ func (m *Manager) retrySettings() (int, time.Duration) {
 	return int(m.requestRetry.Load()), time.Duration(m.maxRetryInterval.Load())
 }
 
-func (m *Manager) closestCooldownWait(providers []string, model string) (time.Duration, bool) {
+func (m *Manager) closestCooldownWait(providers []string, model string, attempt int) (time.Duration, bool) {
 	if m == nil || len(providers) == 0 {
 		return 0, false
 	}
 	now := time.Now()
+	defaultRetry := int(m.requestRetry.Load())
+	if defaultRetry < 0 {
+		defaultRetry = 0
+	}
 	providerSet := make(map[string]struct{}, len(providers))
 	for i := range providers {
 		key := strings.TrimSpace(strings.ToLower(providers[i]))
@@ -1061,6 +1074,16 @@ func (m *Manager) closestCooldownWait(providers []string, model string) (time.Du
 		if _, ok := providerSet[providerKey]; !ok {
 			continue
 		}
+		effectiveRetry := defaultRetry
+		if override, ok := auth.RequestRetryOverride(); ok {
+			effectiveRetry = override
+		}
+		if effectiveRetry < 0 {
+			effectiveRetry = 0
+		}
+		if attempt >= effectiveRetry {
+			continue
+		}
 		blocked, reason, next := isAuthBlockedForModel(auth, model, now)
 		if !blocked || next.IsZero() || reason == blockReasonDisabled {
 			continue
@@ -1077,8 +1100,8 @@ func (m *Manager) closestCooldownWait(providers []string, model string) (time.Du
 	return minWait, found
 }
 
-func (m *Manager) shouldRetryAfterError(err error, attempt, maxAttempts int, providers []string, model string, maxWait time.Duration) (time.Duration, bool) {
-	if err == nil || attempt >= maxAttempts-1 {
+func (m *Manager) shouldRetryAfterError(err error, attempt int, providers []string, model string, maxWait time.Duration) (time.Duration, bool) {
+	if err == nil {
 		return 0, false
 	}
 	if maxWait <= 0 {
@@ -1087,7 +1110,7 @@ func (m *Manager) shouldRetryAfterError(err error, attempt, maxAttempts int, pro
 	if status := statusCodeFromError(err); status == http.StatusOK {
 		return 0, false
 	}
-	wait, found := m.closestCooldownWait(providers, model)
+	wait, found := m.closestCooldownWait(providers, model, attempt)
 	if !found || wait > maxWait {
 		return 0, false
 	}
@@ -1176,7 +1199,7 @@ func (m *Manager) MarkResult(ctx context.Context, result Result) {
 					if result.RetryAfter != nil {
 						next = now.Add(*result.RetryAfter)
 					} else {
-						cooldown, nextLevel := nextQuotaCooldown(backoffLevel)
+						cooldown, nextLevel := nextQuotaCooldown(backoffLevel, quotaCooldownDisabledForAuth(auth))
 						if cooldown > 0 {
 							next = now.Add(cooldown)
 						}
@@ -1193,7 +1216,7 @@ func (m *Manager) MarkResult(ctx context.Context, result Result) {
 					shouldSuspendModel = true
 					setModelQuota = true
 				case 408, 500, 502, 503, 504:
-					if quotaCooldownDisabled.Load() {
+					if quotaCooldownDisabledForAuth(auth) {
 						state.NextRetryAfter = time.Time{}
 					} else {
 						next := now.Add(1 * time.Minute)
@@ -1439,7 +1462,7 @@ func applyAuthFailureState(auth *Auth, resultErr *Error, retryAfter *time.Durati
 		if retryAfter != nil {
 			next = now.Add(*retryAfter)
 		} else {
-			cooldown, nextLevel := nextQuotaCooldown(auth.Quota.BackoffLevel)
+			cooldown, nextLevel := nextQuotaCooldown(auth.Quota.BackoffLevel, quotaCooldownDisabledForAuth(auth))
 			if cooldown > 0 {
 				next = now.Add(cooldown)
 			}
@@ -1449,7 +1472,7 @@ func applyAuthFailureState(auth *Auth, resultErr *Error, retryAfter *time.Durati
 		auth.NextRetryAfter = next
 	case 408, 500, 502, 503, 504:
 		auth.StatusMessage = "transient upstream error"
-		if quotaCooldownDisabled.Load() {
+		if quotaCooldownDisabledForAuth(auth) {
 			auth.NextRetryAfter = time.Time{}
 		} else {
 			auth.NextRetryAfter = now.Add(1 * time.Minute)
@@ -1462,11 +1485,11 @@ func applyAuthFailureState(auth *Auth, resultErr *Error, retryAfter *time.Durati
 }
 
 // nextQuotaCooldown returns the next cooldown duration and updated backoff level for repeated quota errors.
-func nextQuotaCooldown(prevLevel int) (time.Duration, int) {
+func nextQuotaCooldown(prevLevel int, disableCooling bool) (time.Duration, int) {
 	if prevLevel < 0 {
 		prevLevel = 0
 	}
-	if quotaCooldownDisabled.Load() {
+	if disableCooling {
 		return 0, prevLevel
 	}
 	cooldown := quotaBackoffBase * time.Duration(1<<prevLevel)
@@ -1642,6 +1665,9 @@ func (m *Manager) persist(ctx context.Context, auth *Auth) error {
 	if m.store == nil || auth == nil {
 		return nil
 	}
+	if shouldSkipPersist(ctx) {
+		return nil
+	}
 	if auth.Attributes != nil {
 		if v := strings.ToLower(strings.TrimSpace(auth.Attributes["runtime_only"])); v == "true" {
 			return nil

sdk/cliproxy/auth/conductor_overrides_test.go

@@ -0,0 +1,97 @@
+package auth
+
+import (
+	"context"
+	"testing"
+	"time"
+)
+
+func TestManager_ShouldRetryAfterError_RespectsAuthRequestRetryOverride(t *testing.T) {
+	m := NewManager(nil, nil, nil)
+	m.SetRetryConfig(3, 30*time.Second)
+
+	model := "test-model"
+	next := time.Now().Add(5 * time.Second)
+
+	auth := &Auth{
+		ID:       "auth-1",
+		Provider: "claude",
+		Metadata: map[string]any{
+			"request_retry": float64(0),
+		},
+		ModelStates: map[string]*ModelState{
+			model: {
+				Unavailable:    true,
+				Status:         StatusError,
+				NextRetryAfter: next,
+			},
+		},
+	}
+	if _, errRegister := m.Register(context.Background(), auth); errRegister != nil {
+		t.Fatalf("register auth: %v", errRegister)
+	}
+
+	_, maxWait := m.retrySettings()
+	wait, shouldRetry := m.shouldRetryAfterError(&Error{HTTPStatus: 500, Message: "boom"}, 0, []string{"claude"}, model, maxWait)
+	if shouldRetry {
+		t.Fatalf("expected shouldRetry=false for request_retry=0, got true (wait=%v)", wait)
+	}
+
+	auth.Metadata["request_retry"] = float64(1)
+	if _, errUpdate := m.Update(context.Background(), auth); errUpdate != nil {
+		t.Fatalf("update auth: %v", errUpdate)
+	}
+
+	wait, shouldRetry = m.shouldRetryAfterError(&Error{HTTPStatus: 500, Message: "boom"}, 0, []string{"claude"}, model, maxWait)
+	if !shouldRetry {
+		t.Fatalf("expected shouldRetry=true for request_retry=1, got false")
+	}
+	if wait <= 0 {
+		t.Fatalf("expected wait > 0, got %v", wait)
+	}
+
+	_, shouldRetry = m.shouldRetryAfterError(&Error{HTTPStatus: 500, Message: "boom"}, 1, []string{"claude"}, model, maxWait)
+	if shouldRetry {
+		t.Fatalf("expected shouldRetry=false on attempt=1 for request_retry=1, got true")
+	}
+}
+
+func TestManager_MarkResult_RespectsAuthDisableCoolingOverride(t *testing.T) {
+	prev := quotaCooldownDisabled.Load()
+	quotaCooldownDisabled.Store(false)
+	t.Cleanup(func() { quotaCooldownDisabled.Store(prev) })
+
+	m := NewManager(nil, nil, nil)
+
+	auth := &Auth{
+		ID:       "auth-1",
+		Provider: "claude",
+		Metadata: map[string]any{
+			"disable_cooling": true,
+		},
+	}
+	if _, errRegister := m.Register(context.Background(), auth); errRegister != nil {
+		t.Fatalf("register auth: %v", errRegister)
+	}
+
+	model := "test-model"
+	m.MarkResult(context.Background(), Result{
+		AuthID:   "auth-1",
+		Provider: "claude",
+		Model:    model,
+		Success:  false,
+		Error:    &Error{HTTPStatus: 500, Message: "boom"},
+	})
+
+	updated, ok := m.GetByID("auth-1")
+	if !ok || updated == nil {
+		t.Fatalf("expected auth to be present")
+	}
+	state := updated.ModelStates[model]
+	if state == nil {
+		t.Fatalf("expected model state to be present")
+	}
+	if !state.NextRetryAfter.IsZero() {
+		t.Fatalf("expected NextRetryAfter to be zero when disable_cooling=true, got %v", state.NextRetryAfter)
+	}
+}

sdk/cliproxy/auth/persist_policy.go

@@ -0,0 +1,24 @@
+package auth
+
+import "context"
+
+type skipPersistContextKey struct{}
+
+// WithSkipPersist returns a derived context that disables persistence for Manager Update/Register calls.
+// It is intended for code paths that are reacting to file watcher events, where the file on disk is
+// already the source of truth and persisting again would create a write-back loop.
+func WithSkipPersist(ctx context.Context) context.Context {
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	return context.WithValue(ctx, skipPersistContextKey{}, true)
+}
+
+func shouldSkipPersist(ctx context.Context) bool {
+	if ctx == nil {
+		return false
+	}
+	v := ctx.Value(skipPersistContextKey{})
+	enabled, ok := v.(bool)
+	return ok && enabled
+}

sdk/cliproxy/auth/persist_policy_test.go

@@ -0,0 +1,62 @@
+package auth
+
+import (
+	"context"
+	"sync/atomic"
+	"testing"
+)
+
+type countingStore struct {
+	saveCount atomic.Int32
+}
+
+func (s *countingStore) List(context.Context) ([]*Auth, error) { return nil, nil }
+
+func (s *countingStore) Save(context.Context, *Auth) (string, error) {
+	s.saveCount.Add(1)
+	return "", nil
+}
+
+func (s *countingStore) Delete(context.Context, string) error { return nil }
+
+func TestWithSkipPersist_DisablesUpdatePersistence(t *testing.T) {
+	store := &countingStore{}
+	mgr := NewManager(store, nil, nil)
+	auth := &Auth{
+		ID:       "auth-1",
+		Provider: "antigravity",
+		Metadata: map[string]any{"type": "antigravity"},
+	}
+
+	if _, err := mgr.Update(context.Background(), auth); err != nil {
+		t.Fatalf("Update returned error: %v", err)
+	}
+	if got := store.saveCount.Load(); got != 1 {
+		t.Fatalf("expected 1 Save call, got %d", got)
+	}
+
+	ctxSkip := WithSkipPersist(context.Background())
+	if _, err := mgr.Update(ctxSkip, auth); err != nil {
+		t.Fatalf("Update(skipPersist) returned error: %v", err)
+	}
+	if got := store.saveCount.Load(); got != 1 {
+		t.Fatalf("expected Save call count to remain 1, got %d", got)
+	}
+}
+
+func TestWithSkipPersist_DisablesRegisterPersistence(t *testing.T) {
+	store := &countingStore{}
+	mgr := NewManager(store, nil, nil)
+	auth := &Auth{
+		ID:       "auth-1",
+		Provider: "antigravity",
+		Metadata: map[string]any{"type": "antigravity"},
+	}
+
+	if _, err := mgr.Register(WithSkipPersist(context.Background()), auth); err != nil {
+		t.Fatalf("Register(skipPersist) returned error: %v", err)
+	}
+	if got := store.saveCount.Load(); got != 0 {
+		t.Fatalf("expected 0 Save calls, got %d", got)
+	}
+}

sdk/cliproxy/auth/types.go

@@ -194,6 +194,108 @@ func (a *Auth) ProxyInfo() string {
 	return "via proxy"
 }
 
+// DisableCoolingOverride returns the auth-file scoped disable_cooling override when present.
+// The value is read from metadata key "disable_cooling" (or legacy "disable-cooling").
+func (a *Auth) DisableCoolingOverride() (bool, bool) {
+	if a == nil || a.Metadata == nil {
+		return false, false
+	}
+	if val, ok := a.Metadata["disable_cooling"]; ok {
+		if parsed, okParse := parseBoolAny(val); okParse {
+			return parsed, true
+		}
+	}
+	if val, ok := a.Metadata["disable-cooling"]; ok {
+		if parsed, okParse := parseBoolAny(val); okParse {
+			return parsed, true
+		}
+	}
+	return false, false
+}
+
+// RequestRetryOverride returns the auth-file scoped request_retry override when present.
+// The value is read from metadata key "request_retry" (or legacy "request-retry").
+func (a *Auth) RequestRetryOverride() (int, bool) {
+	if a == nil || a.Metadata == nil {
+		return 0, false
+	}
+	if val, ok := a.Metadata["request_retry"]; ok {
+		if parsed, okParse := parseIntAny(val); okParse {
+			if parsed < 0 {
+				parsed = 0
+			}
+			return parsed, true
+		}
+	}
+	if val, ok := a.Metadata["request-retry"]; ok {
+		if parsed, okParse := parseIntAny(val); okParse {
+			if parsed < 0 {
+				parsed = 0
+			}
+			return parsed, true
+		}
+	}
+	return 0, false
+}
+
+func parseBoolAny(val any) (bool, bool) {
+	switch typed := val.(type) {
+	case bool:
+		return typed, true
+	case string:
+		trimmed := strings.TrimSpace(typed)
+		if trimmed == "" {
+			return false, false
+		}
+		parsed, err := strconv.ParseBool(trimmed)
+		if err != nil {
+			return false, false
+		}
+		return parsed, true
+	case float64:
+		return typed != 0, true
+	case json.Number:
+		parsed, err := typed.Int64()
+		if err != nil {
+			return false, false
+		}
+		return parsed != 0, true
+	default:
+		return false, false
+	}
+}
+
+func parseIntAny(val any) (int, bool) {
+	switch typed := val.(type) {
+	case int:
+		return typed, true
+	case int32:
+		return int(typed), true
+	case int64:
+		return int(typed), true
+	case float64:
+		return int(typed), true
+	case json.Number:
+		parsed, err := typed.Int64()
+		if err != nil {
+			return 0, false
+		}
+		return int(parsed), true
+	case string:
+		trimmed := strings.TrimSpace(typed)
+		if trimmed == "" {
+			return 0, false
+		}
+		parsed, err := strconv.Atoi(trimmed)
+		if err != nil {
+			return 0, false
+		}
+		return parsed, true
+	default:
+		return 0, false
+	}
+}
+
 func (a *Auth) AccountInfo() (string, string) {
 	if a == nil {
 		return "", ""

sdk/cliproxy/service.go

@@ -124,6 +124,7 @@ func (s *Service) ensureAuthUpdateQueue(ctx context.Context) {
 }
 
 func (s *Service) consumeAuthUpdates(ctx context.Context) {
+	ctx = coreauth.WithSkipPersist(ctx)
 	for {
 		select {
 		case <-ctx.Done():
@@ -680,6 +681,10 @@ func (s *Service) registerModelsForAuth(a *coreauth.Auth) {
 	if a == nil || a.ID == "" {
 		return
 	}
+	if a.Disabled {
+		GlobalModelRegistry().UnregisterClient(a.ID)
+		return
+	}
 	authKind := strings.ToLower(strings.TrimSpace(a.Attributes["auth_kind"]))
 	if authKind == "" {
 		if kind, _ := a.AccountInfo(); strings.EqualFold(kind, "api_key") {

test/thinking_conversion_test.go

@@ -2,6 +2,7 @@ package test
 
 import (
 	"fmt"
+	"strings"
 	"testing"
 	"time"
 
@@ -2778,12 +2779,18 @@ func runThinkingTests(t *testing.T, cases []thinkingTestCase) {
 
 			// Verify clear_thinking for iFlow GLM models when enable_thinking=true
 			if tc.to == "iflow" && tc.expectField == "chat_template_kwargs.enable_thinking" && tc.expectValue == "true" {
+				baseModel := thinking.ParseSuffix(tc.model).ModelName
+				isGLM := strings.HasPrefix(strings.ToLower(baseModel), "glm")
 				ctVal := gjson.GetBytes(body, "chat_template_kwargs.clear_thinking")
-				if !ctVal.Exists() {
-					t.Fatalf("expected clear_thinking field not found for GLM model, body=%s", string(body))
-				}
-				if ctVal.Bool() != false {
-					t.Fatalf("clear_thinking: expected false, got %v, body=%s", ctVal.Bool(), string(body))
+				if isGLM {
+					if !ctVal.Exists() {
+						t.Fatalf("expected clear_thinking field not found for GLM model, body=%s", string(body))
+					}
+					if ctVal.Bool() != false {
+						t.Fatalf("clear_thinking: expected false, got %v, body=%s", ctVal.Bool(), string(body))
+					}
+				} else if ctVal.Exists() {
+					t.Fatalf("expected no clear_thinking field for non-GLM enable_thinking model, body=%s", string(body))
 				}
 			}
 		})