Merge pull request #114 from router-for-me/plus

v6.7.9

config.example.yaml

@@ -90,6 +90,10 @@ nonstream-keepalive-interval: 0
 #   keepalive-seconds: 15   # Default: 0 (disabled). <= 0 disables keep-alives.
 #   bootstrap-retries: 1    # Default: 0 (disabled). Retries before first byte is sent.
 
+# When true, enable official Codex instructions injection for Codex API requests.
+# When false (default), CodexInstructionsForModel returns immediately without modification.
+codex-instructions-enabled: false
+
 # Gemini API keys
 # gemini-api-key:
 #   - api-key: "AIzaSy...01"
@@ -218,7 +222,7 @@ nonstream-keepalive-interval: 0
 
 # Global OAuth model name aliases (per channel)
 # These aliases rename model IDs for both model listing and request routing.
-# Supported channels: gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow.
+# Supported channels: gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow, kiro, github-copilot.
 # NOTE: Aliases do not apply to gemini-api-key, codex-api-key, claude-api-key, openai-compatibility, vertex-api-key, or ampcode.
 # You can repeat the same name with different aliases to expose multiple client model names.
 #oauth-model-alias:

internal/api/handlers/management/auth_files.go

@@ -24,6 +24,7 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/claude"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/codex"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/copilot"
 	geminiAuth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/gemini"
 	iflowauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/iflow"
 	kiroauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/kiro"
@@ -1707,7 +1708,7 @@ func (h *Handler) RequestQwenToken(c *gin.Context) {
 		// Create token storage
 		tokenStorage := qwenAuth.CreateTokenStorage(tokenData)
 
-		tokenStorage.Email = fmt.Sprintf("qwen-%d", time.Now().UnixMilli())
+		tokenStorage.Email = fmt.Sprintf("%d", time.Now().UnixMilli())
 		record := &coreauth.Auth{
 			ID:       fmt.Sprintf("qwen-%s.json", tokenStorage.Email),
 			Provider: "qwen",
@@ -1812,7 +1813,7 @@ func (h *Handler) RequestIFlowToken(c *gin.Context) {
 		tokenStorage := authSvc.CreateTokenStorage(tokenData)
 		identifier := strings.TrimSpace(tokenStorage.Email)
 		if identifier == "" {
-			identifier = fmt.Sprintf("iflow-%d", time.Now().UnixMilli())
+			identifier = fmt.Sprintf("%d", time.Now().UnixMilli())
 			tokenStorage.Email = identifier
 		}
 		record := &coreauth.Auth{
@@ -1843,6 +1844,89 @@ func (h *Handler) RequestIFlowToken(c *gin.Context) {
 	c.JSON(http.StatusOK, gin.H{"status": "ok", "url": authURL, "state": state})
 }
 
+func (h *Handler) RequestGitHubToken(c *gin.Context) {
+	ctx := context.Background()
+
+	fmt.Println("Initializing GitHub Copilot authentication...")
+
+	state := fmt.Sprintf("gh-%d", time.Now().UnixNano())
+
+	// Initialize Copilot auth service
+	// We need to import "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/copilot" first if not present
+	// Assuming copilot package is imported as "copilot"
+	deviceClient := copilot.NewDeviceFlowClient(h.cfg)
+
+	// Initiate device flow
+	deviceCode, err := deviceClient.RequestDeviceCode(ctx)
+	if err != nil {
+		log.Errorf("Failed to initiate device flow: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to initiate device flow"})
+		return
+	}
+
+	authURL := deviceCode.VerificationURI
+	userCode := deviceCode.UserCode
+
+	RegisterOAuthSession(state, "github")
+
+	go func() {
+		fmt.Printf("Please visit %s and enter code: %s\n", authURL, userCode)
+
+		tokenData, errPoll := deviceClient.PollForToken(ctx, deviceCode)
+		if errPoll != nil {
+			SetOAuthSessionError(state, "Authentication failed")
+			fmt.Printf("Authentication failed: %v\n", errPoll)
+			return
+		}
+
+		username, errUser := deviceClient.FetchUserInfo(ctx, tokenData.AccessToken)
+		if errUser != nil {
+			log.Warnf("Failed to fetch user info: %v", errUser)
+			username = "github-user"
+		}
+
+		tokenStorage := &copilot.CopilotTokenStorage{
+			AccessToken: tokenData.AccessToken,
+			TokenType:   tokenData.TokenType,
+			Scope:       tokenData.Scope,
+			Username:    username,
+			Type:        "github-copilot",
+		}
+
+		fileName := fmt.Sprintf("github-%s.json", username)
+		record := &coreauth.Auth{
+			ID:       fileName,
+			Provider: "github",
+			FileName: fileName,
+			Storage:  tokenStorage,
+			Metadata: map[string]any{
+				"email":    username,
+				"username": username,
+			},
+		}
+
+		savedPath, errSave := h.saveTokenRecord(ctx, record)
+		if errSave != nil {
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
+			SetOAuthSessionError(state, "Failed to save authentication tokens")
+			return
+		}
+
+		fmt.Printf("Authentication successful! Token saved to %s\n", savedPath)
+		fmt.Println("You can now use GitHub Copilot services through this CLI")
+		CompleteOAuthSession(state)
+		CompleteOAuthSessionsByProvider("github")
+	}()
+
+	c.JSON(200, gin.H{
+		"status":           "ok",
+		"url":              authURL,
+		"state":            state,
+		"user_code":        userCode,
+		"verification_uri": authURL,
+	})
+}
+
 func (h *Handler) RequestIFlowCookieToken(c *gin.Context) {
 	ctx := context.Background()
 
@@ -1897,15 +1981,17 @@ func (h *Handler) RequestIFlowCookieToken(c *gin.Context) {
 	fileName := iflowauth.SanitizeIFlowFileName(email)
 	if fileName == "" {
 		fileName = fmt.Sprintf("iflow-%d", time.Now().UnixMilli())
+	} else {
+		fileName = fmt.Sprintf("iflow-%s", fileName)
 	}
 
 	tokenStorage.Email = email
 	timestamp := time.Now().Unix()
 
 	record := &coreauth.Auth{
-		ID:       fmt.Sprintf("iflow-%s-%d.json", fileName, timestamp),
+		ID:       fmt.Sprintf("%s-%d.json", fileName, timestamp),
 		Provider: "iflow",
-		FileName: fmt.Sprintf("iflow-%s-%d.json", fileName, timestamp),
+		FileName: fmt.Sprintf("%s-%d.json", fileName, timestamp),
 		Storage:  tokenStorage,
 		Metadata: map[string]any{
 			"email":        email,

internal/api/handlers/management/logs.go

@@ -13,7 +13,7 @@ import (
 	"time"
 
 	"github.com/gin-gonic/gin"
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/logging"
 )
 
 const (
@@ -360,16 +360,7 @@ func (h *Handler) logDirectory() string {
 	if h.logDir != "" {
 		return h.logDir
 	}
-	if base := util.WritablePath(); base != "" {
-		return filepath.Join(base, "logs")
-	}
-	if h.configFilePath != "" {
-		dir := filepath.Dir(h.configFilePath)
-		if dir != "" && dir != "." {
-			return filepath.Join(dir, "logs")
-		}
-	}
-	return "logs"
+	return logging.ResolveLogDirectory(h.cfg)
 }
 
 func (h *Handler) collectLogFiles(dir string) ([]string, error) {

internal/api/handlers/management/oauth_sessions.go

@@ -238,6 +238,8 @@ func NormalizeOAuthProvider(provider string) (string, error) {
 		return "qwen", nil
 	case "kiro":
 		return "kiro", nil
+	case "github":
+		return "github", nil
 	default:
 		return "", errUnsupportedOAuthFlow
 	}

internal/api/server.go

@@ -26,6 +26,7 @@ import (
 	"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/managementasset"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/usage"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	sdkaccess "github.com/router-for-me/CLIProxyAPI/v6/sdk/access"
@@ -254,15 +255,13 @@ func NewServer(cfg *config.Config, authManager *auth.Manager, accessManager *sdk
 	}
 	managementasset.SetCurrentConfig(cfg)
 	auth.SetQuotaCooldownDisabled(cfg.DisableCooling)
+	misc.SetCodexInstructionsEnabled(cfg.CodexInstructionsEnabled)
 	// Initialize management handler
 	s.mgmt = managementHandlers.NewHandler(cfg, configFilePath, authManager)
 	if optionState.localPassword != "" {
 		s.mgmt.SetLocalPassword(optionState.localPassword)
 	}
-	logDir := filepath.Join(s.currentPath, "logs")
-	if base := util.WritablePath(); base != "" {
-		logDir = filepath.Join(base, "logs")
-	}
+	logDir := logging.ResolveLogDirectory(cfg)
 	s.mgmt.SetLogDirectory(logDir)
 	s.localPassword = optionState.localPassword
 
@@ -641,6 +640,7 @@ func (s *Server) registerManagementRoutes() {
 		mgmt.GET("/iflow-auth-url", s.mgmt.RequestIFlowToken)
 		mgmt.POST("/iflow-auth-url", s.mgmt.RequestIFlowCookieToken)
 		mgmt.GET("/kiro-auth-url", s.mgmt.RequestKiroToken)
+		mgmt.GET("/github-auth-url", s.mgmt.RequestGitHubToken)
 		mgmt.POST("/oauth-callback", s.mgmt.PostOAuthCallback)
 		mgmt.GET("/get-auth-status", s.mgmt.GetAuthStatus)
 	}
@@ -933,6 +933,16 @@ func (s *Server) UpdateClients(cfg *config.Config) {
 			log.Debugf("disable_cooling toggled to %t", cfg.DisableCooling)
 		}
 	}
+
+	if oldCfg == nil || oldCfg.CodexInstructionsEnabled != cfg.CodexInstructionsEnabled {
+		misc.SetCodexInstructionsEnabled(cfg.CodexInstructionsEnabled)
+		if oldCfg != nil {
+			log.Debugf("codex_instructions_enabled updated from %t to %t", oldCfg.CodexInstructionsEnabled, cfg.CodexInstructionsEnabled)
+		} else {
+			log.Debugf("codex_instructions_enabled toggled to %t", cfg.CodexInstructionsEnabled)
+		}
+	}
+
 	if s.handlers != nil && s.handlers.AuthManager != nil {
 		s.handlers.AuthManager.SetRetryConfig(cfg.RequestRetry, time.Duration(cfg.MaxRetryInterval)*time.Second)
 	}

internal/config/config.go

@@ -71,6 +71,11 @@ type Config struct {
 	// WebsocketAuth enables or disables authentication for the WebSocket API.
 	WebsocketAuth bool `yaml:"ws-auth" json:"ws-auth"`
 
+	// CodexInstructionsEnabled controls whether official Codex instructions are injected.
+	// When false (default), CodexInstructionsForModel returns immediately without modification.
+	// When true, the original instruction injection logic is used.
+	CodexInstructionsEnabled bool `yaml:"codex-instructions-enabled" json:"codex-instructions-enabled"`
+
 	// GeminiKey defines Gemini API key configurations with optional routing overrides.
 	GeminiKey []GeminiKey `yaml:"gemini-api-key" json:"gemini-api-key"`
 
@@ -103,7 +108,7 @@ type Config struct {
 
 	// OAuthModelAlias defines global model name aliases for OAuth/file-backed auth channels.
 	// These aliases affect both model listing and model routing for supported channels:
-	// gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow.
+	// gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow, kiro, github-copilot.
 	//
 	// NOTE: This does not apply to existing per-credential model alias features under:
 	// gemini-api-key, codex-api-key, claude-api-key, openai-compatibility, vertex-api-key, and ampcode.

internal/config/oauth_model_alias_migration.go

@@ -122,6 +122,23 @@ func migrateFromOldField(configFile string, root *yaml.Node, rootMap *yaml.Node,
 		newAliases[channel] = converted
 	}
 
+	// For antigravity channel, supplement missing default aliases
+	if antigravityEntries, exists := newAliases["antigravity"]; exists {
+		// Build a set of already configured model names (upstream names)
+		configuredModels := make(map[string]bool, len(antigravityEntries))
+		for _, entry := range antigravityEntries {
+			configuredModels[entry.Name] = true
+		}
+
+		// Add missing default aliases
+		for _, defaultAlias := range defaultAntigravityAliases() {
+			if !configuredModels[defaultAlias.Name] {
+				antigravityEntries = append(antigravityEntries, defaultAlias)
+			}
+		}
+		newAliases["antigravity"] = antigravityEntries
+	}
+
 	// Build new node
 	newNode := buildOAuthModelAliasNode(newAliases)
 

internal/config/oauth_model_alias_migration_test.go

@@ -114,6 +114,23 @@ func TestMigrateOAuthModelAlias_ConvertsAntigravityModels(t *testing.T) {
 	if !strings.Contains(content, "gemini-3-pro-high") {
 		t.Fatal("expected gemini-3-pro-preview to be converted to gemini-3-pro-high")
 	}
+
+	// Verify missing default aliases were supplemented
+	if !strings.Contains(content, "gemini-3-pro-image") {
+		t.Fatal("expected missing default alias gemini-3-pro-image to be added")
+	}
+	if !strings.Contains(content, "gemini-3-flash") {
+		t.Fatal("expected missing default alias gemini-3-flash to be added")
+	}
+	if !strings.Contains(content, "claude-sonnet-4-5") {
+		t.Fatal("expected missing default alias claude-sonnet-4-5 to be added")
+	}
+	if !strings.Contains(content, "claude-sonnet-4-5-thinking") {
+		t.Fatal("expected missing default alias claude-sonnet-4-5-thinking to be added")
+	}
+	if !strings.Contains(content, "claude-opus-4-5-thinking") {
+		t.Fatal("expected missing default alias claude-opus-4-5-thinking to be added")
+	}
 }
 
 func TestMigrateOAuthModelAlias_AddsDefaultIfNeitherExists(t *testing.T) {

internal/logging/global_logger.go

@@ -30,7 +30,7 @@ var (
 type LogFormatter struct{}
 
 // logFieldOrder defines the display order for common log fields.
-var logFieldOrder = []string{"provider", "model", "mode", "budget", "level", "original_value", "min", "max", "clamped_to", "error"}
+var logFieldOrder = []string{"provider", "model", "mode", "budget", "level", "original_mode", "original_value", "min", "max", "clamped_to", "error"}
 
 // Format renders a single log entry with custom formatting.
 func (m *LogFormatter) Format(entry *log.Entry) ([]byte, error) {
@@ -122,6 +122,24 @@ func isDirWritable(dir string) bool {
 	return true
 }
 
+// ResolveLogDirectory determines the directory used for application logs.
+func ResolveLogDirectory(cfg *config.Config) string {
+	logDir := "logs"
+	if base := util.WritablePath(); base != "" {
+		return filepath.Join(base, "logs")
+	}
+	if cfg == nil {
+		return logDir
+	}
+	if !isDirWritable(logDir) {
+		authDir := strings.TrimSpace(cfg.AuthDir)
+		if authDir != "" {
+			logDir = filepath.Join(authDir, "logs")
+		}
+	}
+	return logDir
+}
+
 // ConfigureLogOutput switches the global log destination between rotating files and stdout.
 // When logsMaxTotalSizeMB > 0, a background cleaner removes the oldest log files in the logs directory
 // until the total size is within the limit.
@@ -131,12 +149,7 @@ func ConfigureLogOutput(cfg *config.Config) error {
 	writerMu.Lock()
 	defer writerMu.Unlock()
 
-	logDir := "logs"
-	if base := util.WritablePath(); base != "" {
-		logDir = filepath.Join(base, "logs")
-	} else if !isDirWritable(logDir) {
-		logDir = filepath.Join(cfg.AuthDir, "logs")
-	}
+	logDir := ResolveLogDirectory(cfg)
 
 	protectedPath := ""
 	if cfg.LoggingToFile {

internal/misc/codex_instructions.go

@@ -7,11 +7,27 @@ import (
 	"embed"
 	_ "embed"
 	"strings"
+	"sync/atomic"
 
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
 )
 
+// codexInstructionsEnabled controls whether CodexInstructionsForModel returns official instructions.
+// When false (default), CodexInstructionsForModel returns (true, "") immediately.
+// Set via SetCodexInstructionsEnabled from config.
+var codexInstructionsEnabled atomic.Bool
+
+// SetCodexInstructionsEnabled sets whether codex instructions processing is enabled.
+func SetCodexInstructionsEnabled(enabled bool) {
+	codexInstructionsEnabled.Store(enabled)
+}
+
+// GetCodexInstructionsEnabled returns whether codex instructions processing is enabled.
+func GetCodexInstructionsEnabled() bool {
+	return codexInstructionsEnabled.Load()
+}
+
 //go:embed codex_instructions
 var codexInstructionsDir embed.FS
 
@@ -124,6 +140,9 @@ func codexInstructionsForCodex(modelName, systemInstructions string) (bool, stri
 }
 
 func CodexInstructionsForModel(modelName, systemInstructions, userAgent string) (bool, string) {
+	if !GetCodexInstructionsEnabled() {
+		return true, ""
+	}
 	if IsOpenCodeUserAgent(userAgent) {
 		return codexInstructionsForOpenCode(systemInstructions)
 	}

internal/registry/model_definitions.go

@@ -847,6 +847,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "OpenAI GPT-5 via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 32768,
+			SupportedEndpoints:  []string{"/chat/completions", "/responses"},
 		},
 		{
 			ID:                  "gpt-5-mini",
@@ -858,6 +859,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "OpenAI GPT-5 Mini via GitHub Copilot",
 			ContextLength:       128000,
 			MaxCompletionTokens: 16384,
+			SupportedEndpoints:  []string{"/chat/completions", "/responses"},
 		},
 		{
 			ID:                  "gpt-5-codex",
@@ -869,6 +871,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "OpenAI GPT-5 Codex via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 32768,
+			SupportedEndpoints:  []string{"/responses"},
 		},
 		{
 			ID:                  "gpt-5.1",
@@ -880,6 +883,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "OpenAI GPT-5.1 via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 32768,
+			SupportedEndpoints:  []string{"/chat/completions", "/responses"},
 		},
 		{
 			ID:                  "gpt-5.1-codex",
@@ -891,6 +895,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "OpenAI GPT-5.1 Codex via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 32768,
+			SupportedEndpoints:  []string{"/responses"},
 		},
 		{
 			ID:                  "gpt-5.1-codex-mini",
@@ -902,6 +907,19 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "OpenAI GPT-5.1 Codex Mini via GitHub Copilot",
 			ContextLength:       128000,
 			MaxCompletionTokens: 16384,
+			SupportedEndpoints:  []string{"/responses"},
+		},
+		{
+			ID:                  "gpt-5.1-codex-max",
+			Object:              "model",
+			Created:             now,
+			OwnedBy:             "github-copilot",
+			Type:                "github-copilot",
+			DisplayName:         "GPT-5.1 Codex Max",
+			Description:         "OpenAI GPT-5.1 Codex Max via GitHub Copilot",
+			ContextLength:       200000,
+			MaxCompletionTokens: 32768,
+			SupportedEndpoints:  []string{"/responses"},
 		},
 		{
 			ID:                  "gpt-5.2",
@@ -913,6 +931,19 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "OpenAI GPT-5.2 via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 32768,
+			SupportedEndpoints:  []string{"/chat/completions", "/responses"},
+		},
+		{
+			ID:                  "gpt-5.2-codex",
+			Object:              "model",
+			Created:             now,
+			OwnedBy:             "github-copilot",
+			Type:                "github-copilot",
+			DisplayName:         "GPT-5.2 Codex",
+			Description:         "OpenAI GPT-5.2 Codex via GitHub Copilot",
+			ContextLength:       200000,
+			MaxCompletionTokens: 32768,
+			SupportedEndpoints:  []string{"/responses"},
 		},
 		{
 			ID:                  "claude-haiku-4.5",
@@ -924,6 +955,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "Anthropic Claude Haiku 4.5 via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
+			SupportedEndpoints:  []string{"/chat/completions"},
 		},
 		{
 			ID:                  "claude-opus-4.1",
@@ -935,6 +967,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "Anthropic Claude Opus 4.1 via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 32000,
+			SupportedEndpoints:  []string{"/chat/completions"},
 		},
 		{
 			ID:                  "claude-opus-4.5",
@@ -946,6 +979,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "Anthropic Claude Opus 4.5 via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
+			SupportedEndpoints:  []string{"/chat/completions"},
 		},
 		{
 			ID:                  "claude-sonnet-4",
@@ -957,6 +991,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "Anthropic Claude Sonnet 4 via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
+			SupportedEndpoints:  []string{"/chat/completions"},
 		},
 		{
 			ID:                  "claude-sonnet-4.5",
@@ -968,6 +1003,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "Anthropic Claude Sonnet 4.5 via GitHub Copilot",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
+			SupportedEndpoints:  []string{"/chat/completions"},
 		},
 		{
 			ID:                  "gemini-2.5-pro",
@@ -981,13 +1017,24 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			MaxCompletionTokens: 65536,
 		},
 		{
-			ID:                  "gemini-3-pro",
+			ID:                  "gemini-3-pro-preview",
 			Object:              "model",
 			Created:             now,
 			OwnedBy:             "github-copilot",
 			Type:                "github-copilot",
-			DisplayName:         "Gemini 3 Pro",
-			Description:         "Google Gemini 3 Pro via GitHub Copilot",
+			DisplayName:         "Gemini 3 Pro (Preview)",
+			Description:         "Google Gemini 3 Pro Preview via GitHub Copilot",
+			ContextLength:       1048576,
+			MaxCompletionTokens: 65536,
+		},
+		{
+			ID:                  "gemini-3-flash-preview",
+			Object:              "model",
+			Created:             now,
+			OwnedBy:             "github-copilot",
+			Type:                "github-copilot",
+			DisplayName:         "Gemini 3 Flash (Preview)",
+			Description:         "Google Gemini 3 Flash Preview via GitHub Copilot",
 			ContextLength:       1048576,
 			MaxCompletionTokens: 65536,
 		},
@@ -1003,15 +1050,16 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			MaxCompletionTokens: 16384,
 		},
 		{
-			ID:                  "raptor-mini",
+			ID:                  "oswe-vscode-prime",
 			Object:              "model",
 			Created:             now,
 			OwnedBy:             "github-copilot",
 			Type:                "github-copilot",
-			DisplayName:         "Raptor Mini",
-			Description:         "Raptor Mini via GitHub Copilot",
+			DisplayName:         "Raptor mini (Preview)",
+			Description:         "Raptor mini via GitHub Copilot",
 			ContextLength:       128000,
 			MaxCompletionTokens: 16384,
+			SupportedEndpoints:  []string{"/chat/completions", "/responses"},
 		},
 	}
 }
@@ -1020,6 +1068,18 @@ func GetGitHubCopilotModels() []*ModelInfo {
 func GetKiroModels() []*ModelInfo {
 	return []*ModelInfo{
 		// --- Base Models ---
+		{
+			ID:                  "kiro-auto",
+			Object:              "model",
+			Created:             1732752000,
+			OwnedBy:             "aws",
+			Type:                "kiro",
+			DisplayName:         "Kiro Auto",
+			Description:         "Automatic model selection by Kiro",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 32000, ZeroAllowed: true, DynamicAllowed: true},
+		},
 		{
 			ID:                  "kiro-claude-opus-4-5",
 			Object:              "model",

internal/registry/model_registry.go

@@ -47,6 +47,8 @@ type ModelInfo struct {
 	MaxCompletionTokens int `json:"max_completion_tokens,omitempty"`
 	// SupportedParameters lists supported parameters
 	SupportedParameters []string `json:"supported_parameters,omitempty"`
+	// SupportedEndpoints lists supported API endpoints (e.g., "/chat/completions", "/responses").
+	SupportedEndpoints []string `json:"supported_endpoints,omitempty"`
 
 	// Thinking holds provider-specific reasoning/thinking budget capabilities.
 	// This is optional and currently used for Gemini thinking budget normalization.
@@ -476,6 +478,9 @@ func cloneModelInfo(model *ModelInfo) *ModelInfo {
 	if len(model.SupportedParameters) > 0 {
 		copyModel.SupportedParameters = append([]string(nil), model.SupportedParameters...)
 	}
+	if len(model.SupportedEndpoints) > 0 {
+		copyModel.SupportedEndpoints = append([]string(nil), model.SupportedEndpoints...)
+	}
 	return &copyModel
 }
 
@@ -988,6 +993,9 @@ func (r *ModelRegistry) convertModelToMap(model *ModelInfo, handlerType string)
 		if len(model.SupportedParameters) > 0 {
 			result["supported_parameters"] = model.SupportedParameters
 		}
+		if len(model.SupportedEndpoints) > 0 {
+			result["supported_endpoints"] = model.SupportedEndpoints
+		}
 		return result
 
 	case "claude", "kiro", "antigravity":

internal/runtime/executor/aistudio_executor.go

@@ -393,7 +393,7 @@ func (e *AIStudioExecutor) translateRequest(req cliproxyexecutor.Request, opts c
 	}
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, stream)
 	payload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), stream)
-	payload, err := thinking.ApplyThinking(payload, req.Model, "gemini")
+	payload, err := thinking.ApplyThinking(payload, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, translatedPayload{}, err
 	}

internal/runtime/executor/antigravity_executor.go

@@ -137,7 +137,7 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
 	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-	translated, err = thinking.ApplyThinking(translated, req.Model, "antigravity")
+	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -256,7 +256,7 @@ func (e *AntigravityExecutor) executeClaudeNonStream(ctx context.Context, auth *
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
 	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 
-	translated, err = thinking.ApplyThinking(translated, req.Model, "antigravity")
+	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -517,8 +517,8 @@ func (e *AntigravityExecutor) convertStreamToNonStream(stream []byte) []byte {
 		}
 		if usageResult := responseNode.Get("usageMetadata"); usageResult.Exists() {
 			usageRaw = usageResult.Raw
-		} else if usageResult := root.Get("usageMetadata"); usageResult.Exists() {
-			usageRaw = usageResult.Raw
+		} else if usageMetadataResult := root.Get("usageMetadata"); usageMetadataResult.Exists() {
+			usageRaw = usageMetadataResult.Raw
 		}
 
 		if partsResult := responseNode.Get("candidates.0.content.parts"); partsResult.IsArray() {
@@ -622,7 +622,7 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
 	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 
-	translated, err = thinking.ApplyThinking(translated, req.Model, "antigravity")
+	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, err
 	}
@@ -642,7 +642,6 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 			err = errReq
 			return nil, err
 		}
-
 		httpResp, errDo := httpClient.Do(httpReq)
 		if errDo != nil {
 			recordAPIResponseError(ctx, e.cfg, errDo)
@@ -803,7 +802,7 @@ func (e *AntigravityExecutor) CountTokens(ctx context.Context, auth *cliproxyaut
 	// Prepare payload once (doesn't depend on baseURL)
 	payload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-	payload, err := thinking.ApplyThinking(payload, req.Model, "antigravity")
+	payload, err := thinking.ApplyThinking(payload, req.Model, from.String(), to.String())
 	if err != nil {
 		return cliproxyexecutor.Response{}, err
 	}
@@ -1004,10 +1003,10 @@ func FetchAntigravityModels(ctx context.Context, auth *cliproxyauth.Auth, cfg *c
 			case "chat_20706", "chat_23310", "gemini-2.5-flash-thinking", "gemini-3-pro-low", "gemini-2.5-pro":
 				continue
 			}
-			cfg := modelConfig[modelID]
+			modelCfg := modelConfig[modelID]
 			modelName := modelID
-			if cfg != nil && cfg.Name != "" {
-				modelName = cfg.Name
+			if modelCfg != nil && modelCfg.Name != "" {
+				modelName = modelCfg.Name
 			}
 			modelInfo := &registry.ModelInfo{
 				ID:          modelID,
@@ -1021,12 +1020,12 @@ func FetchAntigravityModels(ctx context.Context, auth *cliproxyauth.Auth, cfg *c
 				Type:        antigravityAuthType,
 			}
 			// Look up Thinking support from static config using upstream model name.
-			if cfg != nil {
-				if cfg.Thinking != nil {
-					modelInfo.Thinking = cfg.Thinking
+			if modelCfg != nil {
+				if modelCfg.Thinking != nil {
+					modelInfo.Thinking = modelCfg.Thinking
 				}
-				if cfg.MaxCompletionTokens > 0 {
-					modelInfo.MaxCompletionTokens = cfg.MaxCompletionTokens
+				if modelCfg.MaxCompletionTokens > 0 {
+					modelInfo.MaxCompletionTokens = modelCfg.MaxCompletionTokens
 				}
 			}
 			models = append(models, modelInfo)
@@ -1221,7 +1220,7 @@ func (e *AntigravityExecutor) buildRequest(ctx context.Context, auth *cliproxyau
 		payload = []byte(strJSON)
 	}
 
-	if strings.Contains(modelName, "claude") || strings.Contains(modelName, "gemini-3-pro-preview") {
+	if strings.Contains(modelName, "claude") || strings.Contains(modelName, "gemini-3-pro-high") {
 		systemInstructionPartsResult := gjson.GetBytes(payload, "request.systemInstruction.parts")
 		payload, _ = sjson.SetBytes(payload, "request.systemInstruction.role", "user")
 		payload, _ = sjson.SetBytes(payload, "request.systemInstruction.parts.0.text", systemInstruction)

internal/runtime/executor/claude_executor.go

@@ -106,7 +106,7 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), stream)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "claude")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -239,7 +239,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "claude")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, err
 	}

internal/runtime/executor/codex_executor.go

@@ -96,7 +96,7 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	body = sdktranslator.TranslateRequest(from, to, baseModel, body, false)
 	body = misc.StripCodexUserAgent(body)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "codex")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -106,6 +106,10 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	body, _ = sjson.SetBytes(body, "stream", true)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
 	body, _ = sjson.DeleteBytes(body, "prompt_cache_retention")
+	body, _ = sjson.DeleteBytes(body, "safety_identifier")
+	if !gjson.GetBytes(body, "instructions").Exists() {
+		body, _ = sjson.SetBytes(body, "instructions", "")
+	}
 
 	url := strings.TrimSuffix(baseURL, "/") + "/responses"
 	httpReq, err := e.cacheHelper(ctx, from, url, req, body)
@@ -204,7 +208,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	body = sdktranslator.TranslateRequest(from, to, baseModel, body, true)
 	body = misc.StripCodexUserAgent(body)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "codex")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, err
 	}
@@ -212,7 +216,11 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
 	body, _ = sjson.DeleteBytes(body, "prompt_cache_retention")
+	body, _ = sjson.DeleteBytes(body, "safety_identifier")
 	body, _ = sjson.SetBytes(body, "model", baseModel)
+	if !gjson.GetBytes(body, "instructions").Exists() {
+		body, _ = sjson.SetBytes(body, "instructions", "")
+	}
 
 	url := strings.TrimSuffix(baseURL, "/") + "/responses"
 	httpReq, err := e.cacheHelper(ctx, from, url, req, body)
@@ -308,7 +316,7 @@ func (e *CodexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth
 	body = sdktranslator.TranslateRequest(from, to, baseModel, body, false)
 	body = misc.StripCodexUserAgent(body)
 
-	body, err := thinking.ApplyThinking(body, req.Model, "codex")
+	body, err := thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return cliproxyexecutor.Response{}, err
 	}
@@ -316,7 +324,11 @@ func (e *CodexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
 	body, _ = sjson.DeleteBytes(body, "prompt_cache_retention")
+	body, _ = sjson.DeleteBytes(body, "safety_identifier")
 	body, _ = sjson.SetBytes(body, "stream", false)
+	if !gjson.GetBytes(body, "instructions").Exists() {
+		body, _ = sjson.SetBytes(body, "instructions", "")
+	}
 
 	enc, err := tokenizerForCodexModel(baseModel)
 	if err != nil {

internal/runtime/executor/gemini_cli_executor.go

@@ -123,7 +123,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
 	basePayload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-	basePayload, err = thinking.ApplyThinking(basePayload, req.Model, "gemini-cli")
+	basePayload, err = thinking.ApplyThinking(basePayload, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -272,7 +272,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
 	basePayload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 
-	basePayload, err = thinking.ApplyThinking(basePayload, req.Model, "gemini-cli")
+	basePayload, err = thinking.ApplyThinking(basePayload, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, err
 	}
@@ -479,7 +479,7 @@ func (e *GeminiCLIExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.
 	for range models {
 		payload := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-		payload, err = thinking.ApplyThinking(payload, req.Model, "gemini-cli")
+		payload, err = thinking.ApplyThinking(payload, req.Model, from.String(), to.String())
 		if err != nil {
 			return cliproxyexecutor.Response{}, err
 		}

internal/runtime/executor/gemini_executor.go

@@ -120,7 +120,7 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "gemini")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -222,7 +222,7 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "gemini")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, err
 	}
@@ -338,7 +338,7 @@ func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	to := sdktranslator.FromString("gemini")
 	translatedReq := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-	translatedReq, err := thinking.ApplyThinking(translatedReq, req.Model, "gemini")
+	translatedReq, err := thinking.ApplyThinking(translatedReq, req.Model, from.String(), to.String())
 	if err != nil {
 		return cliproxyexecutor.Response{}, err
 	}

internal/runtime/executor/gemini_vertex_executor.go

@@ -170,7 +170,7 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "gemini")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -272,7 +272,7 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "gemini")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -375,7 +375,7 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "gemini")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, err
 	}
@@ -494,7 +494,7 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "gemini")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, err
 	}
@@ -605,7 +605,7 @@ func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context
 
 	translatedReq := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-	translatedReq, err := thinking.ApplyThinking(translatedReq, req.Model, "gemini")
+	translatedReq, err := thinking.ApplyThinking(translatedReq, req.Model, from.String(), to.String())
 	if err != nil {
 		return cliproxyexecutor.Response{}, err
 	}
@@ -689,7 +689,7 @@ func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *
 
 	translatedReq := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 
-	translatedReq, err := thinking.ApplyThinking(translatedReq, req.Model, "gemini")
+	translatedReq, err := thinking.ApplyThinking(translatedReq, req.Model, from.String(), to.String())
 	if err != nil {
 		return cliproxyexecutor.Response{}, err
 	}

internal/runtime/executor/github_copilot_executor.go

@@ -23,6 +23,7 @@ import (
 const (
 	githubCopilotBaseURL       = "https://api.githubcopilot.com"
 	githubCopilotChatPath      = "/chat/completions"
+	githubCopilotResponsesPath = "/responses"
 	githubCopilotAuthType      = "github-copilot"
 	githubCopilotTokenCacheTTL = 25 * time.Minute
 	// tokenExpiryBuffer is the time before expiry when we should refresh the token.
@@ -106,7 +107,11 @@ func (e *GitHubCopilotExecutor) Execute(ctx context.Context, auth *cliproxyauth.
 	defer reporter.trackFailure(ctx, &err)
 
 	from := opts.SourceFormat
+	useResponses := useGitHubCopilotResponsesEndpoint(from)
 	to := sdktranslator.FromString("openai")
+	if useResponses {
+		to = sdktranslator.FromString("openai-response")
+	}
 	originalPayload := bytes.Clone(req.Payload)
 	if len(opts.OriginalRequest) > 0 {
 		originalPayload = bytes.Clone(opts.OriginalRequest)
@@ -117,7 +122,11 @@ func (e *GitHubCopilotExecutor) Execute(ctx context.Context, auth *cliproxyauth.
 	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "stream", false)
 
-	url := githubCopilotBaseURL + githubCopilotChatPath
+	path := githubCopilotChatPath
+	if useResponses {
+		path = githubCopilotResponsesPath
+	}
+	url := githubCopilotBaseURL + path
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
 	if err != nil {
 		return resp, err
@@ -172,6 +181,9 @@ func (e *GitHubCopilotExecutor) Execute(ctx context.Context, auth *cliproxyauth.
 	appendAPIResponseChunk(ctx, e.cfg, data)
 
 	detail := parseOpenAIUsage(data)
+	if useResponses && detail.TotalTokens == 0 {
+		detail = parseOpenAIResponsesUsage(data)
+	}
 	if detail.TotalTokens > 0 {
 		reporter.publish(ctx, detail)
 	}
@@ -194,7 +206,11 @@ func (e *GitHubCopilotExecutor) ExecuteStream(ctx context.Context, auth *cliprox
 	defer reporter.trackFailure(ctx, &err)
 
 	from := opts.SourceFormat
+	useResponses := useGitHubCopilotResponsesEndpoint(from)
 	to := sdktranslator.FromString("openai")
+	if useResponses {
+		to = sdktranslator.FromString("openai-response")
+	}
 	originalPayload := bytes.Clone(req.Payload)
 	if len(opts.OriginalRequest) > 0 {
 		originalPayload = bytes.Clone(opts.OriginalRequest)
@@ -205,9 +221,15 @@ func (e *GitHubCopilotExecutor) ExecuteStream(ctx context.Context, auth *cliprox
 	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "stream", true)
 	// Enable stream options for usage stats in stream
-	body, _ = sjson.SetBytes(body, "stream_options.include_usage", true)
+	if !useResponses {
+		body, _ = sjson.SetBytes(body, "stream_options.include_usage", true)
+	}
 
-	url := githubCopilotBaseURL + githubCopilotChatPath
+	path := githubCopilotChatPath
+	if useResponses {
+		path = githubCopilotResponsesPath
+	}
+	url := githubCopilotBaseURL + path
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
 	if err != nil {
 		return nil, err
@@ -283,6 +305,10 @@ func (e *GitHubCopilotExecutor) ExecuteStream(ctx context.Context, auth *cliprox
 				}
 				if detail, ok := parseOpenAIStreamUsage(line); ok {
 					reporter.publish(ctx, detail)
+				} else if useResponses {
+					if detail, ok := parseOpenAIResponsesStreamUsage(line); ok {
+						reporter.publish(ctx, detail)
+					}
 				}
 			}
 
@@ -393,6 +419,10 @@ func (e *GitHubCopilotExecutor) normalizeModel(_ string, body []byte) []byte {
 	return body
 }
 
+func useGitHubCopilotResponsesEndpoint(sourceFormat sdktranslator.Format) bool {
+	return sourceFormat.String() == "openai-response"
+}
+
 // isHTTPSuccess checks if the status code indicates success (2xx).
 func isHTTPSuccess(statusCode int) bool {
 	return statusCode >= 200 && statusCode < 300

internal/runtime/executor/iflow_executor.go

@@ -92,7 +92,7 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "iflow")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), "iflow")
 	if err != nil {
 		return resp, err
 	}
@@ -190,7 +190,7 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "iflow")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), "iflow")
 	if err != nil {
 		return nil, err
 	}

internal/runtime/executor/kiro_executor.go

@@ -53,6 +53,7 @@ const (
 	kiroIDEUserAgent     = "aws-sdk-js/1.0.18 ua/2.1 os/darwin#25.0.0 lang/js md/nodejs#20.16.0 api/codewhispererstreaming#1.0.18 m/E KiroIDE-0.2.13-66c23a8c5d15afabec89ef9954ef52a119f10d369df04d548fc6c1eac694b0d1"
 	kiroIDEAmzUserAgent  = "aws-sdk-js/1.0.18 KiroIDE-0.2.13-66c23a8c5d15afabec89ef9954ef52a119f10d369df04d548fc6c1eac694b0d1"
 	kiroIDEAgentModeSpec = "spec"
+	kiroAgentModeVibe    = "vibe"
 )
 
 // Real-time usage estimation configuration
@@ -98,7 +99,7 @@ var kiroEndpointConfigs = []kiroEndpointConfig{
 		Name:      "CodeWhisperer",
 	},
 	{
-		URL:       "https://q.us-east-1.amazonaws.com/",
+		URL:       "https://q.us-east-1.amazonaws.com/generateAssistantResponse",
 		Origin:    "CLI",
 		AmzTarget: "AmazonQDeveloperStreamingService.SendMessage",
 		Name:      "AmazonQ",
@@ -232,7 +233,9 @@ func (e *KiroExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth
 	} else {
 		req.Header.Set("User-Agent", kiroUserAgent)
 		req.Header.Set("X-Amz-User-Agent", kiroFullUserAgent)
+		req.Header.Set("x-amzn-kiro-agent-mode", kiroAgentModeVibe)
 	}
+	req.Header.Set("x-amzn-codewhisperer-optout", "true")
 	req.Header.Set("Amz-Sdk-Request", "attempt=1; max=3")
 	req.Header.Set("Amz-Sdk-Invocation-Id", uuid.New().String())
 	req.Header.Set("Authorization", "Bearer "+accessToken)
@@ -350,7 +353,9 @@ func (e *KiroExecutor) executeWithRetry(ctx context.Context, auth *cliproxyauth.
 			} else {
 				httpReq.Header.Set("User-Agent", kiroUserAgent)
 				httpReq.Header.Set("X-Amz-User-Agent", kiroFullUserAgent)
+				httpReq.Header.Set("x-amzn-kiro-agent-mode", kiroAgentModeVibe)
 			}
+			httpReq.Header.Set("x-amzn-codewhisperer-optout", "true")
 			httpReq.Header.Set("Amz-Sdk-Request", "attempt=1; max=3")
 			httpReq.Header.Set("Amz-Sdk-Invocation-Id", uuid.New().String())
 
@@ -683,7 +688,9 @@ func (e *KiroExecutor) executeStreamWithRetry(ctx context.Context, auth *cliprox
 			} else {
 				httpReq.Header.Set("User-Agent", kiroUserAgent)
 				httpReq.Header.Set("X-Amz-User-Agent", kiroFullUserAgent)
+				httpReq.Header.Set("x-amzn-kiro-agent-mode", kiroAgentModeVibe)
 			}
+			httpReq.Header.Set("x-amzn-codewhisperer-optout", "true")
 			httpReq.Header.Set("Amz-Sdk-Request", "attempt=1; max=3")
 			httpReq.Header.Set("Amz-Sdk-Invocation-Id", uuid.New().String())
 

internal/runtime/executor/openai_compat_executor.go

@@ -92,7 +92,7 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), opts.Stream)
 	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", translated, originalTranslated)
 
-	translated, err = thinking.ApplyThinking(translated, req.Model, "openai")
+	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -187,7 +187,7 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", translated, originalTranslated)
 
-	translated, err = thinking.ApplyThinking(translated, req.Model, "openai")
+	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, err
 	}
@@ -297,7 +297,7 @@ func (e *OpenAICompatExecutor) CountTokens(ctx context.Context, auth *cliproxyau
 
 	modelForCounting := baseModel
 
-	translated, err := thinking.ApplyThinking(translated, req.Model, "openai")
+	translated, err := thinking.ApplyThinking(translated, req.Model, from.String(), to.String())
 	if err != nil {
 		return cliproxyexecutor.Response{}, err
 	}

internal/runtime/executor/payload_helpers.go

@@ -25,6 +25,7 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 	if model == "" {
 		return payload
 	}
+	candidates := payloadModelCandidates(cfg, model, protocol)
 	out := payload
 	source := original
 	if len(source) == 0 {
@@ -34,7 +35,7 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 	// Apply default rules: first write wins per field across all matching rules.
 	for i := range rules.Default {
 		rule := &rules.Default[i]
-		if !payloadRuleMatchesModel(rule, model, protocol) {
+		if !payloadRuleMatchesModels(rule, protocol, candidates) {
 			continue
 		}
 		for path, value := range rule.Params {
@@ -59,7 +60,7 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 	// Apply default raw rules: first write wins per field across all matching rules.
 	for i := range rules.DefaultRaw {
 		rule := &rules.DefaultRaw[i]
-		if !payloadRuleMatchesModel(rule, model, protocol) {
+		if !payloadRuleMatchesModels(rule, protocol, candidates) {
 			continue
 		}
 		for path, value := range rule.Params {
@@ -88,7 +89,7 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 	// Apply override rules: last write wins per field across all matching rules.
 	for i := range rules.Override {
 		rule := &rules.Override[i]
-		if !payloadRuleMatchesModel(rule, model, protocol) {
+		if !payloadRuleMatchesModels(rule, protocol, candidates) {
 			continue
 		}
 		for path, value := range rule.Params {
@@ -106,7 +107,7 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 	// Apply override raw rules: last write wins per field across all matching rules.
 	for i := range rules.OverrideRaw {
 		rule := &rules.OverrideRaw[i]
-		if !payloadRuleMatchesModel(rule, model, protocol) {
+		if !payloadRuleMatchesModels(rule, protocol, candidates) {
 			continue
 		}
 		for path, value := range rule.Params {
@@ -128,6 +129,18 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 	return out
 }
 
+func payloadRuleMatchesModels(rule *config.PayloadRule, protocol string, models []string) bool {
+	if rule == nil || len(models) == 0 {
+		return false
+	}
+	for _, model := range models {
+		if payloadRuleMatchesModel(rule, model, protocol) {
+			return true
+		}
+	}
+	return false
+}
+
 func payloadRuleMatchesModel(rule *config.PayloadRule, model, protocol string) bool {
 	if rule == nil {
 		return false
@@ -150,6 +163,65 @@ func payloadRuleMatchesModel(rule *config.PayloadRule, model, protocol string) b
 	return false
 }
 
+func payloadModelCandidates(cfg *config.Config, model, protocol string) []string {
+	model = strings.TrimSpace(model)
+	if model == "" {
+		return nil
+	}
+	candidates := []string{model}
+	if cfg == nil {
+		return candidates
+	}
+	aliases := payloadModelAliases(cfg, model, protocol)
+	if len(aliases) == 0 {
+		return candidates
+	}
+	seen := map[string]struct{}{strings.ToLower(model): struct{}{}}
+	for _, alias := range aliases {
+		alias = strings.TrimSpace(alias)
+		if alias == "" {
+			continue
+		}
+		key := strings.ToLower(alias)
+		if _, ok := seen[key]; ok {
+			continue
+		}
+		seen[key] = struct{}{}
+		candidates = append(candidates, alias)
+	}
+	return candidates
+}
+
+func payloadModelAliases(cfg *config.Config, model, protocol string) []string {
+	if cfg == nil {
+		return nil
+	}
+	model = strings.TrimSpace(model)
+	if model == "" {
+		return nil
+	}
+	channel := strings.ToLower(strings.TrimSpace(protocol))
+	if channel == "" {
+		return nil
+	}
+	entries := cfg.OAuthModelAlias[channel]
+	if len(entries) == 0 {
+		return nil
+	}
+	aliases := make([]string, 0, 2)
+	for _, entry := range entries {
+		if !strings.EqualFold(strings.TrimSpace(entry.Name), model) {
+			continue
+		}
+		alias := strings.TrimSpace(entry.Alias)
+		if alias == "" {
+			continue
+		}
+		aliases = append(aliases, alias)
+	}
+	return aliases
+}
+
 // buildPayloadPath combines an optional root path with a relative parameter path.
 // When root is empty, the parameter path is used as-is. When root is non-empty,
 // the parameter path is treated as relative to root.

internal/runtime/executor/qwen_executor.go

@@ -86,7 +86,7 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "openai")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return resp, err
 	}
@@ -172,7 +172,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
-	body, err = thinking.ApplyThinking(body, req.Model, "openai")
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String())
 	if err != nil {
 		return nil, err
 	}

internal/runtime/executor/usage_helpers.go

@@ -236,6 +236,44 @@ func parseOpenAIStreamUsage(line []byte) (usage.Detail, bool) {
 	return detail, true
 }
 
+func parseOpenAIResponsesUsageDetail(usageNode gjson.Result) usage.Detail {
+	detail := usage.Detail{
+		InputTokens:  usageNode.Get("input_tokens").Int(),
+		OutputTokens: usageNode.Get("output_tokens").Int(),
+		TotalTokens:  usageNode.Get("total_tokens").Int(),
+	}
+	if detail.TotalTokens == 0 {
+		detail.TotalTokens = detail.InputTokens + detail.OutputTokens
+	}
+	if cached := usageNode.Get("input_tokens_details.cached_tokens"); cached.Exists() {
+		detail.CachedTokens = cached.Int()
+	}
+	if reasoning := usageNode.Get("output_tokens_details.reasoning_tokens"); reasoning.Exists() {
+		detail.ReasoningTokens = reasoning.Int()
+	}
+	return detail
+}
+
+func parseOpenAIResponsesUsage(data []byte) usage.Detail {
+	usageNode := gjson.ParseBytes(data).Get("usage")
+	if !usageNode.Exists() {
+		return usage.Detail{}
+	}
+	return parseOpenAIResponsesUsageDetail(usageNode)
+}
+
+func parseOpenAIResponsesStreamUsage(line []byte) (usage.Detail, bool) {
+	payload := jsonPayload(line)
+	if len(payload) == 0 || !gjson.ValidBytes(payload) {
+		return usage.Detail{}, false
+	}
+	usageNode := gjson.GetBytes(payload, "usage")
+	if !usageNode.Exists() {
+		return usage.Detail{}, false
+	}
+	return parseOpenAIResponsesUsageDetail(usageNode), true
+}
+
 func parseClaudeUsage(data []byte) usage.Detail {
 	usageNode := gjson.ParseBytes(data).Get("usage")
 	if !usageNode.Exists() {

internal/thinking/apply.go

@@ -2,6 +2,8 @@
 package thinking
 
 import (
+	"strings"
+
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	log "github.com/sirupsen/logrus"
 	"github.com/tidwall/gjson"
@@ -59,7 +61,8 @@ func IsUserDefinedModel(modelInfo *registry.ModelInfo) bool {
 // Parameters:
 //   - body: Original request body JSON
 //   - model: Model name, optionally with thinking suffix (e.g., "claude-sonnet-4-5(16384)")
-//   - provider: Provider name (gemini, gemini-cli, antigravity, claude, openai, codex, iflow)
+//   - fromFormat: Source request format (e.g., openai, codex, gemini)
+//   - toFormat: Target provider format for the request body (gemini, gemini-cli, antigravity, claude, openai, codex, iflow)
 //
 // Returns:
 //   - Modified request body JSON with thinking configuration applied
@@ -76,16 +79,21 @@ func IsUserDefinedModel(modelInfo *registry.ModelInfo) bool {
 // Example:
 //
 //	// With suffix - suffix config takes priority
-//	result, err := thinking.ApplyThinking(body, "gemini-2.5-pro(8192)", "gemini")
+//	result, err := thinking.ApplyThinking(body, "gemini-2.5-pro(8192)", "gemini", "gemini")
 //
 //	// Without suffix - uses body config
-//	result, err := thinking.ApplyThinking(body, "gemini-2.5-pro", "gemini")
-func ApplyThinking(body []byte, model string, provider string) ([]byte, error) {
+//	result, err := thinking.ApplyThinking(body, "gemini-2.5-pro", "gemini", "gemini")
+func ApplyThinking(body []byte, model string, fromFormat string, toFormat string) ([]byte, error) {
+	providerFormat := strings.ToLower(strings.TrimSpace(toFormat))
+	fromFormat = strings.ToLower(strings.TrimSpace(fromFormat))
+	if fromFormat == "" {
+		fromFormat = providerFormat
+	}
 	// 1. Route check: Get provider applier
-	applier := GetProviderApplier(provider)
+	applier := GetProviderApplier(providerFormat)
 	if applier == nil {
 		log.WithFields(log.Fields{
-			"provider": provider,
+			"provider": providerFormat,
 			"model":    model,
 		}).Debug("thinking: unknown provider, passthrough |")
 		return body, nil
@@ -100,19 +108,19 @@ func ApplyThinking(body []byte, model string, provider string) ([]byte, error) {
 	// Unknown models are treated as user-defined so thinking config can still be applied.
 	// The upstream service is responsible for validating the configuration.
 	if IsUserDefinedModel(modelInfo) {
-		return applyUserDefinedModel(body, modelInfo, provider, suffixResult)
+		return applyUserDefinedModel(body, modelInfo, fromFormat, providerFormat, suffixResult)
 	}
 	if modelInfo.Thinking == nil {
-		config := extractThinkingConfig(body, provider)
+		config := extractThinkingConfig(body, providerFormat)
 		if hasThinkingConfig(config) {
 			log.WithFields(log.Fields{
 				"model":    baseModel,
-				"provider": provider,
+				"provider": providerFormat,
 			}).Debug("thinking: model does not support thinking, stripping config |")
-			return StripThinkingConfig(body, provider), nil
+			return StripThinkingConfig(body, providerFormat), nil
 		}
 		log.WithFields(log.Fields{
-			"provider": provider,
+			"provider": providerFormat,
 			"model":    baseModel,
 		}).Debug("thinking: model does not support thinking, passthrough |")
 		return body, nil
@@ -121,19 +129,19 @@ func ApplyThinking(body []byte, model string, provider string) ([]byte, error) {
 	// 4. Get config: suffix priority over body
 	var config ThinkingConfig
 	if suffixResult.HasSuffix {
-		config = parseSuffixToConfig(suffixResult.RawSuffix, provider, model)
+		config = parseSuffixToConfig(suffixResult.RawSuffix, providerFormat, model)
 		log.WithFields(log.Fields{
-			"provider": provider,
+			"provider": providerFormat,
 			"model":    model,
 			"mode":     config.Mode,
 			"budget":   config.Budget,
 			"level":    config.Level,
 		}).Debug("thinking: config from model suffix |")
 	} else {
-		config = extractThinkingConfig(body, provider)
+		config = extractThinkingConfig(body, providerFormat)
 		if hasThinkingConfig(config) {
 			log.WithFields(log.Fields{
-				"provider": provider,
+				"provider": providerFormat,
 				"model":    modelInfo.ID,
 				"mode":     config.Mode,
 				"budget":   config.Budget,
@@ -144,17 +152,17 @@ func ApplyThinking(body []byte, model string, provider string) ([]byte, error) {
 
 	if !hasThinkingConfig(config) {
 		log.WithFields(log.Fields{
-			"provider": provider,
+			"provider": providerFormat,
 			"model":    modelInfo.ID,
 		}).Debug("thinking: no config found, passthrough |")
 		return body, nil
 	}
 
 	// 5. Validate and normalize configuration
-	validated, err := ValidateConfig(config, modelInfo, provider)
+	validated, err := ValidateConfig(config, modelInfo, fromFormat, providerFormat, suffixResult.HasSuffix)
 	if err != nil {
 		log.WithFields(log.Fields{
-			"provider": provider,
+			"provider": providerFormat,
 			"model":    modelInfo.ID,
 			"error":    err.Error(),
 		}).Warn("thinking: validation failed |")
@@ -167,14 +175,14 @@ func ApplyThinking(body []byte, model string, provider string) ([]byte, error) {
 	// Defensive check: ValidateConfig should never return (nil, nil)
 	if validated == nil {
 		log.WithFields(log.Fields{
-			"provider": provider,
+			"provider": providerFormat,
 			"model":    modelInfo.ID,
 		}).Warn("thinking: ValidateConfig returned nil config without error, passthrough |")
 		return body, nil
 	}
 
 	log.WithFields(log.Fields{
-		"provider": provider,
+		"provider": providerFormat,
 		"model":    modelInfo.ID,
 		"mode":     validated.Mode,
 		"budget":   validated.Budget,
@@ -228,7 +236,7 @@ func parseSuffixToConfig(rawSuffix, provider, model string) ThinkingConfig {
 
 // applyUserDefinedModel applies thinking configuration for user-defined models
 // without ThinkingSupport validation.
-func applyUserDefinedModel(body []byte, modelInfo *registry.ModelInfo, provider string, suffixResult SuffixResult) ([]byte, error) {
+func applyUserDefinedModel(body []byte, modelInfo *registry.ModelInfo, fromFormat, toFormat string, suffixResult SuffixResult) ([]byte, error) {
 	// Get model ID for logging
 	modelID := ""
 	if modelInfo != nil {
@@ -240,39 +248,57 @@ func applyUserDefinedModel(body []byte, modelInfo *registry.ModelInfo, provider
 	// Get config: suffix priority over body
 	var config ThinkingConfig
 	if suffixResult.HasSuffix {
-		config = parseSuffixToConfig(suffixResult.RawSuffix, provider, modelID)
+		config = parseSuffixToConfig(suffixResult.RawSuffix, toFormat, modelID)
 	} else {
-		config = extractThinkingConfig(body, provider)
+		config = extractThinkingConfig(body, toFormat)
 	}
 
 	if !hasThinkingConfig(config) {
 		log.WithFields(log.Fields{
 			"model":    modelID,
-			"provider": provider,
+			"provider": toFormat,
 		}).Debug("thinking: user-defined model, passthrough (no config) |")
 		return body, nil
 	}
 
-	applier := GetProviderApplier(provider)
+	applier := GetProviderApplier(toFormat)
 	if applier == nil {
 		log.WithFields(log.Fields{
 			"model":    modelID,
-			"provider": provider,
+			"provider": toFormat,
 		}).Debug("thinking: user-defined model, passthrough (unknown provider) |")
 		return body, nil
 	}
 
 	log.WithFields(log.Fields{
-		"provider": provider,
+		"provider": toFormat,
 		"model":    modelID,
 		"mode":     config.Mode,
 		"budget":   config.Budget,
 		"level":    config.Level,
 	}).Debug("thinking: applying config for user-defined model (skip validation)")
 
+	config = normalizeUserDefinedConfig(config, fromFormat, toFormat)
 	return applier.Apply(body, config, modelInfo)
 }
 
+func normalizeUserDefinedConfig(config ThinkingConfig, fromFormat, toFormat string) ThinkingConfig {
+	if config.Mode != ModeLevel {
+		return config
+	}
+	if !isBudgetBasedProvider(toFormat) || !isLevelBasedProvider(fromFormat) {
+		return config
+	}
+	budget, ok := ConvertLevelToBudget(string(config.Level))
+	if !ok {
+		return config
+	}
+	config.Mode = ModeBudget
+	config.Budget = budget
+	config.Level = ""
+	return config
+}
+
 // extractThinkingConfig extracts provider-specific thinking config from request body.
 func extractThinkingConfig(body []byte, provider string) ThinkingConfig {
 	if len(body) == 0 || !gjson.ValidBytes(body) {
@@ -289,7 +315,11 @@ func extractThinkingConfig(body []byte, provider string) ThinkingConfig {
 	case "codex":
 		return extractCodexConfig(body)
 	case "iflow":
-		return extractIFlowConfig(body)
+		config := extractIFlowConfig(body)
+		if hasThinkingConfig(config) {
+			return config
+		}
+		return extractOpenAIConfig(body)
 	default:
 		return ThinkingConfig{}
 	}

internal/thinking/errors.go

@@ -24,6 +24,10 @@ const (
 	// Example: using level with a budget-only model
 	ErrLevelNotSupported ErrorCode = "LEVEL_NOT_SUPPORTED"
 
+	// ErrBudgetOutOfRange indicates the budget value is outside model range.
+	// Example: budget 64000 exceeds max 20000
+	ErrBudgetOutOfRange ErrorCode = "BUDGET_OUT_OF_RANGE"
+
 	// ErrProviderMismatch indicates the provider does not match the model.
 	// Example: applying Claude format to a Gemini model
 	ErrProviderMismatch ErrorCode = "PROVIDER_MISMATCH"

internal/thinking/strip.go

@@ -27,28 +27,32 @@ func StripThinkingConfig(body []byte, provider string) []byte {
 		return body
 	}
 
+	var paths []string
 	switch provider {
 	case "claude":
-		result, _ := sjson.DeleteBytes(body, "thinking")
-		return result
+		paths = []string{"thinking"}
 	case "gemini":
-		result, _ := sjson.DeleteBytes(body, "generationConfig.thinkingConfig")
-		return result
+		paths = []string{"generationConfig.thinkingConfig"}
 	case "gemini-cli", "antigravity":
-		result, _ := sjson.DeleteBytes(body, "request.generationConfig.thinkingConfig")
-		return result
+		paths = []string{"request.generationConfig.thinkingConfig"}
 	case "openai":
-		result, _ := sjson.DeleteBytes(body, "reasoning_effort")
-		return result
+		paths = []string{"reasoning_effort"}
 	case "codex":
-		result, _ := sjson.DeleteBytes(body, "reasoning.effort")
-		return result
+		paths = []string{"reasoning.effort"}
 	case "iflow":
-		result, _ := sjson.DeleteBytes(body, "chat_template_kwargs.enable_thinking")
-		result, _ = sjson.DeleteBytes(result, "chat_template_kwargs.clear_thinking")
-		result, _ = sjson.DeleteBytes(result, "reasoning_split")
-		return result
+		paths = []string{
+			"chat_template_kwargs.enable_thinking",
+			"chat_template_kwargs.clear_thinking",
+			"reasoning_split",
+			"reasoning_effort",
+		}
 	default:
 		return body
 	}
+
+	result := body
+	for _, path := range paths {
+		result, _ = sjson.DeleteBytes(result, path)
+	}
+	return result
 }

internal/thinking/validate.go

@@ -9,64 +9,6 @@ import (
 	log "github.com/sirupsen/logrus"
 )
 
-// ClampBudget clamps a budget value to the model's supported range.
-//
-// Logging:
-//   - Warn when value=0 but ZeroAllowed=false
-//   - Debug when value is clamped to min/max
-//
-// Fields: provider, model, original_value, clamped_to, min, max
-func ClampBudget(value int, modelInfo *registry.ModelInfo, provider string) int {
-	model := "unknown"
-	support := (*registry.ThinkingSupport)(nil)
-	if modelInfo != nil {
-		if modelInfo.ID != "" {
-			model = modelInfo.ID
-		}
-		support = modelInfo.Thinking
-	}
-	if support == nil {
-		return value
-	}
-
-	// Auto value (-1) passes through without clamping.
-	if value == -1 {
-		return value
-	}
-
-	min := support.Min
-	max := support.Max
-	if value == 0 && !support.ZeroAllowed {
-		log.WithFields(log.Fields{
-			"provider":       provider,
-			"model":          model,
-			"original_value": value,
-			"clamped_to":     min,
-			"min":            min,
-			"max":            max,
-		}).Warn("thinking: budget zero not allowed |")
-		return min
-	}
-
-	// Some models are level-only and do not define numeric budget ranges.
-	if min == 0 && max == 0 {
-		return value
-	}
-
-	if value < min {
-		if value == 0 && support.ZeroAllowed {
-			return 0
-		}
-		logClamp(provider, model, value, min, min, max)
-		return min
-	}
-	if value > max {
-		logClamp(provider, model, value, max, min, max)
-		return max
-	}
-	return value
-}
-
 // ValidateConfig validates a thinking configuration against model capabilities.
 //
 // This function performs comprehensive validation:
@@ -74,10 +16,16 @@ func ClampBudget(value int, modelInfo *registry.ModelInfo, provider string) int
 //   - Auto-converts between Budget and Level formats based on model capability
 //   - Validates that requested level is in the model's supported levels list
 //   - Clamps budget values to model's allowed range
+//   - When converting Budget -> Level for level-only models, clamps the derived standard level to the nearest supported level
+//     (special values none/auto are preserved)
+//   - When config comes from a model suffix, strict budget validation is disabled (we clamp instead of error)
 //
 // Parameters:
 //   - config: The thinking configuration to validate
 //   - support: Model's ThinkingSupport properties (nil means no thinking support)
+//   - fromFormat: Source provider format (used to determine strict validation rules)
+//   - toFormat: Target provider format
+//   - fromSuffix: Whether config was sourced from model suffix
 //
 // Returns:
 //   - Normalized ThinkingConfig with clamped values
@@ -87,9 +35,8 @@ func ClampBudget(value int, modelInfo *registry.ModelInfo, provider string) int
 //   - Budget-only model + Level config → Level converted to Budget
 //   - Level-only model + Budget config → Budget converted to Level
 //   - Hybrid model → preserve original format
-func ValidateConfig(config ThinkingConfig, modelInfo *registry.ModelInfo, provider string) (*ThinkingConfig, error) {
-	normalized := config
-
+func ValidateConfig(config ThinkingConfig, modelInfo *registry.ModelInfo, fromFormat, toFormat string, fromSuffix bool) (*ThinkingConfig, error) {
+	fromFormat, toFormat = strings.ToLower(strings.TrimSpace(fromFormat)), strings.ToLower(strings.TrimSpace(toFormat))
 	model := "unknown"
 	support := (*registry.ThinkingSupport)(nil)
 	if modelInfo != nil {
@@ -103,101 +50,108 @@ func ValidateConfig(config ThinkingConfig, modelInfo *registry.ModelInfo, provid
 		if config.Mode != ModeNone {
 			return nil, NewThinkingErrorWithModel(ErrThinkingNotSupported, "thinking not supported for this model", model)
 		}
-		return &normalized, nil
+		return &config, nil
 	}
 
+	allowClampUnsupported := isBudgetBasedProvider(fromFormat) && isLevelBasedProvider(toFormat)
+	strictBudget := !fromSuffix && fromFormat != "" && isSameProviderFamily(fromFormat, toFormat)
+	budgetDerivedFromLevel := false
+
 	capability := detectModelCapability(modelInfo)
 	switch capability {
 	case CapabilityBudgetOnly:
-		if normalized.Mode == ModeLevel {
-			if normalized.Level == LevelAuto {
+		if config.Mode == ModeLevel {
+			if config.Level == LevelAuto {
 				break
 			}
-			budget, ok := ConvertLevelToBudget(string(normalized.Level))
+			budget, ok := ConvertLevelToBudget(string(config.Level))
 			if !ok {
-				return nil, NewThinkingError(ErrUnknownLevel, fmt.Sprintf("unknown level: %s", normalized.Level))
+				return nil, NewThinkingError(ErrUnknownLevel, fmt.Sprintf("unknown level: %s", config.Level))
 			}
-			normalized.Mode = ModeBudget
-			normalized.Budget = budget
-			normalized.Level = ""
+			config.Mode = ModeBudget
+			config.Budget = budget
+			config.Level = ""
+			budgetDerivedFromLevel = true
 		}
 	case CapabilityLevelOnly:
-		if normalized.Mode == ModeBudget {
-			level, ok := ConvertBudgetToLevel(normalized.Budget)
+		if config.Mode == ModeBudget {
+			level, ok := ConvertBudgetToLevel(config.Budget)
 			if !ok {
-				return nil, NewThinkingError(ErrUnknownLevel, fmt.Sprintf("budget %d cannot be converted to a valid level", normalized.Budget))
+				return nil, NewThinkingError(ErrUnknownLevel, fmt.Sprintf("budget %d cannot be converted to a valid level", config.Budget))
 			}
-			normalized.Mode = ModeLevel
-			normalized.Level = ThinkingLevel(level)
-			normalized.Budget = 0
+			// When converting Budget -> Level for level-only models, clamp the derived standard level
+			// to the nearest supported level. Special values (none/auto) are preserved.
+			config.Mode = ModeLevel
+			config.Level = clampLevel(ThinkingLevel(level), modelInfo, toFormat)
+			config.Budget = 0
 		}
 	case CapabilityHybrid:
 	}
 
-	if normalized.Mode == ModeLevel && normalized.Level == LevelNone {
-		normalized.Mode = ModeNone
-		normalized.Budget = 0
-		normalized.Level = ""
+	if config.Mode == ModeLevel && config.Level == LevelNone {
+		config.Mode = ModeNone
+		config.Budget = 0
+		config.Level = ""
 	}
-	if normalized.Mode == ModeLevel && normalized.Level == LevelAuto {
-		normalized.Mode = ModeAuto
-		normalized.Budget = -1
-		normalized.Level = ""
+	if config.Mode == ModeLevel && config.Level == LevelAuto {
+		config.Mode = ModeAuto
+		config.Budget = -1
+		config.Level = ""
 	}
-	if normalized.Mode == ModeBudget && normalized.Budget == 0 {
-		normalized.Mode = ModeNone
-		normalized.Level = ""
+	if config.Mode == ModeBudget && config.Budget == 0 {
+		config.Mode = ModeNone
+		config.Level = ""
 	}
 
-	if len(support.Levels) > 0 && normalized.Mode == ModeLevel {
-		if !isLevelSupported(string(normalized.Level), support.Levels) {
-			validLevels := normalizeLevels(support.Levels)
-			message := fmt.Sprintf("level %q not supported, valid levels: %s", strings.ToLower(string(normalized.Level)), strings.Join(validLevels, ", "))
-			return nil, NewThinkingError(ErrLevelNotSupported, message)
+	if len(support.Levels) > 0 && config.Mode == ModeLevel {
+		if !isLevelSupported(string(config.Level), support.Levels) {
+			if allowClampUnsupported {
+				config.Level = clampLevel(config.Level, modelInfo, toFormat)
+			}
+			if !isLevelSupported(string(config.Level), support.Levels) {
+				// User explicitly specified an unsupported level - return error
+				// (budget-derived levels may be clamped based on source format)
+				validLevels := normalizeLevels(support.Levels)
+				message := fmt.Sprintf("level %q not supported, valid levels: %s", strings.ToLower(string(config.Level)), strings.Join(validLevels, ", "))
+				return nil, NewThinkingError(ErrLevelNotSupported, message)
+			}
+		}
+	}
+
+	if strictBudget && config.Mode == ModeBudget && !budgetDerivedFromLevel {
+		min, max := support.Min, support.Max
+		if min != 0 || max != 0 {
+			if config.Budget < min || config.Budget > max || (config.Budget == 0 && !support.ZeroAllowed) {
+				message := fmt.Sprintf("budget %d out of range [%d,%d]", config.Budget, min, max)
+				return nil, NewThinkingError(ErrBudgetOutOfRange, message)
+			}
 		}
 	}
 
 	// Convert ModeAuto to mid-range if dynamic not allowed
-	if normalized.Mode == ModeAuto && !support.DynamicAllowed {
-		normalized = convertAutoToMidRange(normalized, support, provider, model)
+	if config.Mode == ModeAuto && !support.DynamicAllowed {
+		config = convertAutoToMidRange(config, support, toFormat, model)
 	}
 
-	if normalized.Mode == ModeNone && provider == "claude" {
+	if config.Mode == ModeNone && toFormat == "claude" {
 		// Claude supports explicit disable via thinking.type="disabled".
 		// Keep Budget=0 so applier can omit budget_tokens.
-		normalized.Budget = 0
-		normalized.Level = ""
+		config.Budget = 0
+		config.Level = ""
 	} else {
-		switch normalized.Mode {
+		switch config.Mode {
 		case ModeBudget, ModeAuto, ModeNone:
-			normalized.Budget = ClampBudget(normalized.Budget, modelInfo, provider)
+			config.Budget = clampBudget(config.Budget, modelInfo, toFormat)
 		}
 
 		// ModeNone with clamped Budget > 0: set Level to lowest for Level-only/Hybrid models
 		// This ensures Apply layer doesn't need to access support.Levels
-		if normalized.Mode == ModeNone && normalized.Budget > 0 && len(support.Levels) > 0 {
-			normalized.Level = ThinkingLevel(support.Levels[0])
+		if config.Mode == ModeNone && config.Budget > 0 && len(support.Levels) > 0 {
+			config.Level = ThinkingLevel(support.Levels[0])
 		}
 	}
 
-	return &normalized, nil
-}
-
-func isLevelSupported(level string, supported []string) bool {
-	for _, candidate := range supported {
-		if strings.EqualFold(level, strings.TrimSpace(candidate)) {
-			return true
-		}
-	}
-	return false
-}
-
-func normalizeLevels(levels []string) []string {
-	normalized := make([]string, 0, len(levels))
-	for _, level := range levels {
-		normalized = append(normalized, strings.ToLower(strings.TrimSpace(level)))
-	}
-	return normalized
+	return &config, nil
 }
 
 // convertAutoToMidRange converts ModeAuto to a mid-range value when dynamic is not allowed.
@@ -246,7 +200,172 @@ func convertAutoToMidRange(config ThinkingConfig, support *registry.ThinkingSupp
 	return config
 }
 
-// logClamp logs a debug message when budget clamping occurs.
+// standardLevelOrder defines the canonical ordering of thinking levels from lowest to highest.
+var standardLevelOrder = []ThinkingLevel{LevelMinimal, LevelLow, LevelMedium, LevelHigh, LevelXHigh}
+
+// clampLevel clamps the given level to the nearest supported level.
+// On tie, prefers the lower level.
+func clampLevel(level ThinkingLevel, modelInfo *registry.ModelInfo, provider string) ThinkingLevel {
+	model := "unknown"
+	var supported []string
+	if modelInfo != nil {
+		if modelInfo.ID != "" {
+			model = modelInfo.ID
+		}
+		if modelInfo.Thinking != nil {
+			supported = modelInfo.Thinking.Levels
+		}
+	}
+
+	if len(supported) == 0 || isLevelSupported(string(level), supported) {
+		return level
+	}
+
+	pos := levelIndex(string(level))
+	if pos == -1 {
+		return level
+	}
+	bestIdx, bestDist := -1, len(standardLevelOrder)+1
+
+	for _, s := range supported {
+		if idx := levelIndex(strings.TrimSpace(s)); idx != -1 {
+			if dist := abs(pos - idx); dist < bestDist || (dist == bestDist && idx < bestIdx) {
+				bestIdx, bestDist = idx, dist
+			}
+		}
+	}
+
+	if bestIdx >= 0 {
+		clamped := standardLevelOrder[bestIdx]
+		log.WithFields(log.Fields{
+			"provider":       provider,
+			"model":          model,
+			"original_value": string(level),
+			"clamped_to":     string(clamped),
+		}).Debug("thinking: level clamped |")
+		return clamped
+	}
+	return level
+}
+
+// clampBudget clamps a budget value to the model's supported range.
+func clampBudget(value int, modelInfo *registry.ModelInfo, provider string) int {
+	model := "unknown"
+	support := (*registry.ThinkingSupport)(nil)
+	if modelInfo != nil {
+		if modelInfo.ID != "" {
+			model = modelInfo.ID
+		}
+		support = modelInfo.Thinking
+	}
+	if support == nil {
+		return value
+	}
+
+	// Auto value (-1) passes through without clamping.
+	if value == -1 {
+		return value
+	}
+
+	min, max := support.Min, support.Max
+	if value == 0 && !support.ZeroAllowed {
+		log.WithFields(log.Fields{
+			"provider":       provider,
+			"model":          model,
+			"original_value": value,
+			"clamped_to":     min,
+			"min":            min,
+			"max":            max,
+		}).Warn("thinking: budget zero not allowed |")
+		return min
+	}
+
+	// Some models are level-only and do not define numeric budget ranges.
+	if min == 0 && max == 0 {
+		return value
+	}
+
+	if value < min {
+		if value == 0 && support.ZeroAllowed {
+			return 0
+		}
+		logClamp(provider, model, value, min, min, max)
+		return min
+	}
+	if value > max {
+		logClamp(provider, model, value, max, min, max)
+		return max
+	}
+	return value
+}
+
+func isLevelSupported(level string, supported []string) bool {
+	for _, s := range supported {
+		if strings.EqualFold(level, strings.TrimSpace(s)) {
+			return true
+		}
+	}
+	return false
+}
+
+func levelIndex(level string) int {
+	for i, l := range standardLevelOrder {
+		if strings.EqualFold(level, string(l)) {
+			return i
+		}
+	}
+	return -1
+}
+
+func normalizeLevels(levels []string) []string {
+	out := make([]string, len(levels))
+	for i, l := range levels {
+		out[i] = strings.ToLower(strings.TrimSpace(l))
+	}
+	return out
+}
+
+func isBudgetBasedProvider(provider string) bool {
+	switch provider {
+	case "gemini", "gemini-cli", "antigravity", "claude":
+		return true
+	default:
+		return false
+	}
+}
+
+func isLevelBasedProvider(provider string) bool {
+	switch provider {
+	case "openai", "openai-response", "codex":
+		return true
+	default:
+		return false
+	}
+}
+
+func isGeminiFamily(provider string) bool {
+	switch provider {
+	case "gemini", "gemini-cli", "antigravity":
+		return true
+	default:
+		return false
+	}
+}
+
+func isSameProviderFamily(from, to string) bool {
+	if from == to {
+		return true
+	}
+	return isGeminiFamily(from) && isGeminiFamily(to)
+}
+
+func abs(x int) int {
+	if x < 0 {
+		return -x
+	}
+	return x
+}
+
 func logClamp(provider, model string, original, clampedTo, min, max int) {
 	log.WithFields(log.Fields{
 		"provider":       provider,

internal/translator/antigravity/claude/antigravity_claude_request.go

@@ -12,7 +12,6 @@ import (
 	"strings"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/cache"
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
@@ -388,14 +387,11 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when type==enabled
 	if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() {
-		modelInfo := registry.LookupModelInfo(modelName)
-		if modelInfo != nil && modelInfo.Thinking != nil {
-			if t.Get("type").String() == "enabled" {
-				if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
-					budget := int(b.Int())
-					out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
-					out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
-				}
+		if t.Get("type").String() == "enabled" {
+			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
+				budget := int(b.Int())
+				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
+				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.includeThoughts", true)
 			}
 		}
 	}

internal/translator/antigravity/claude/antigravity_claude_request_test.go

@@ -343,8 +343,8 @@ func TestConvertClaudeRequestToAntigravity_ThinkingConfig(t *testing.T) {
 		if thinkingConfig.Get("thinkingBudget").Int() != 8000 {
 			t.Errorf("Expected thinkingBudget 8000, got %d", thinkingConfig.Get("thinkingBudget").Int())
 		}
-		if !thinkingConfig.Get("include_thoughts").Bool() {
-			t.Error("include_thoughts should be true")
+		if !thinkingConfig.Get("includeThoughts").Bool() {
+			t.Error("includeThoughts should be true")
 		}
 	} else {
 		t.Log("thinkingConfig not present - model may not be registered in test registry")

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

@@ -132,6 +132,7 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 			}
 		}
 
+		systemPartIndex := 0
 		for i := 0; i < len(arr); i++ {
 			m := arr[i]
 			role := m.Get("role").String()
@@ -141,16 +142,19 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 				// system -> request.systemInstruction as a user message style
 				if content.Type == gjson.String {
 					out, _ = sjson.SetBytes(out, "request.systemInstruction.role", "user")
-					out, _ = sjson.SetBytes(out, "request.systemInstruction.parts.0.text", content.String())
+					out, _ = sjson.SetBytes(out, fmt.Sprintf("request.systemInstruction.parts.%d.text", systemPartIndex), content.String())
+					systemPartIndex++
 				} else if content.IsObject() && content.Get("type").String() == "text" {
 					out, _ = sjson.SetBytes(out, "request.systemInstruction.role", "user")
-					out, _ = sjson.SetBytes(out, "request.systemInstruction.parts.0.text", content.Get("text").String())
+					out, _ = sjson.SetBytes(out, fmt.Sprintf("request.systemInstruction.parts.%d.text", systemPartIndex), content.Get("text").String())
+					systemPartIndex++
 				} else if content.IsArray() {
 					contents := content.Array()
 					if len(contents) > 0 {
 						out, _ = sjson.SetBytes(out, "request.systemInstruction.role", "user")
 						for j := 0; j < len(contents); j++ {
-							out, _ = sjson.SetBytes(out, fmt.Sprintf("request.systemInstruction.parts.%d.text", j), contents[j].Get("text").String())
+							out, _ = sjson.SetBytes(out, fmt.Sprintf("request.systemInstruction.parts.%d.text", systemPartIndex), contents[j].Get("text").String())
+							systemPartIndex++
 						}
 					}
 				}
@@ -165,7 +169,10 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 					for _, item := range items {
 						switch item.Get("type").String() {
 						case "text":
-							node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".text", item.Get("text").String())
+							text := item.Get("text").String()
+							if text != "" {
+								node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".text", text)
+							}
 							p++
 						case "image_url":
 							imageURL := item.Get("image_url.url").String()
@@ -209,6 +216,10 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 					for _, item := range content.Array() {
 						switch item.Get("type").String() {
 						case "text":
+							text := item.Get("text").String()
+							if text != "" {
+								node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".text", text)
+							}
 							p++
 						case "image_url":
 							// If the assistant returned an inline data URL, preserve it for history fidelity.

internal/translator/claude/gemini/claude_gemini_request.go

@@ -15,7 +15,7 @@ import (
 	"strings"
 
 	"github.com/google/uuid"
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -115,18 +115,41 @@ func ConvertGeminiRequestToClaude(modelName string, inputRawJSON []byte, stream
 			}
 		}
 		// Include thoughts configuration for reasoning process visibility
-		// Only apply for models that support thinking and use numeric budgets, not discrete levels.
+		// Translator only does format conversion, ApplyThinking handles model capability validation.
 		if thinkingConfig := genConfig.Get("thinkingConfig"); thinkingConfig.Exists() && thinkingConfig.IsObject() {
-			modelInfo := registry.LookupModelInfo(modelName)
-			if modelInfo != nil && modelInfo.Thinking != nil && len(modelInfo.Thinking.Levels) == 0 {
-				// Check for thinkingBudget first - if present, enable thinking with budget
-				if thinkingBudget := thinkingConfig.Get("thinkingBudget"); thinkingBudget.Exists() && thinkingBudget.Int() > 0 {
-					out, _ = sjson.Set(out, "thinking.type", "enabled")
-					out, _ = sjson.Set(out, "thinking.budget_tokens", thinkingBudget.Int())
-				} else if includeThoughts := thinkingConfig.Get("include_thoughts"); includeThoughts.Exists() && includeThoughts.Type == gjson.True {
-					// Fallback to include_thoughts if no budget specified
+			if thinkingLevel := thinkingConfig.Get("thinkingLevel"); thinkingLevel.Exists() {
+				level := strings.ToLower(strings.TrimSpace(thinkingLevel.String()))
+				switch level {
+				case "":
+				case "none":
+					out, _ = sjson.Set(out, "thinking.type", "disabled")
+					out, _ = sjson.Delete(out, "thinking.budget_tokens")
+				case "auto":
 					out, _ = sjson.Set(out, "thinking.type", "enabled")
+					out, _ = sjson.Delete(out, "thinking.budget_tokens")
+				default:
+					if budget, ok := thinking.ConvertLevelToBudget(level); ok {
+						out, _ = sjson.Set(out, "thinking.type", "enabled")
+						out, _ = sjson.Set(out, "thinking.budget_tokens", budget)
+					}
 				}
+			} else if thinkingBudget := thinkingConfig.Get("thinkingBudget"); thinkingBudget.Exists() {
+				budget := int(thinkingBudget.Int())
+				switch budget {
+				case 0:
+					out, _ = sjson.Set(out, "thinking.type", "disabled")
+					out, _ = sjson.Delete(out, "thinking.budget_tokens")
+				case -1:
+					out, _ = sjson.Set(out, "thinking.type", "enabled")
+					out, _ = sjson.Delete(out, "thinking.budget_tokens")
+				default:
+					out, _ = sjson.Set(out, "thinking.type", "enabled")
+					out, _ = sjson.Set(out, "thinking.budget_tokens", budget)
+				}
+			} else if includeThoughts := thinkingConfig.Get("includeThoughts"); includeThoughts.Exists() && includeThoughts.Type == gjson.True {
+				out, _ = sjson.Set(out, "thinking.type", "enabled")
+			} else if includeThoughts := thinkingConfig.Get("include_thoughts"); includeThoughts.Exists() && includeThoughts.Type == gjson.True {
+				out, _ = sjson.Set(out, "thinking.type", "enabled")
 			}
 		}
 	}

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

@@ -15,7 +15,6 @@ import (
 	"strings"
 
 	"github.com/google/uuid"
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -66,23 +65,21 @@ func ConvertOpenAIRequestToClaude(modelName string, inputRawJSON []byte, stream
 
 	root := gjson.ParseBytes(rawJSON)
 
+	// Convert OpenAI reasoning_effort to Claude thinking config.
 	if v := root.Get("reasoning_effort"); v.Exists() {
-		modelInfo := registry.LookupModelInfo(modelName)
-		if modelInfo != nil && modelInfo.Thinking != nil && len(modelInfo.Thinking.Levels) == 0 {
-			effort := strings.ToLower(strings.TrimSpace(v.String()))
-			if effort != "" {
-				budget, ok := thinking.ConvertLevelToBudget(effort)
-				if ok {
-					switch budget {
-					case 0:
-						out, _ = sjson.Set(out, "thinking.type", "disabled")
-					case -1:
+		effort := strings.ToLower(strings.TrimSpace(v.String()))
+		if effort != "" {
+			budget, ok := thinking.ConvertLevelToBudget(effort)
+			if ok {
+				switch budget {
+				case 0:
+					out, _ = sjson.Set(out, "thinking.type", "disabled")
+				case -1:
+					out, _ = sjson.Set(out, "thinking.type", "enabled")
+				default:
+					if budget > 0 {
 						out, _ = sjson.Set(out, "thinking.type", "enabled")
-					default:
-						if budget > 0 {
-							out, _ = sjson.Set(out, "thinking.type", "enabled")
-							out, _ = sjson.Set(out, "thinking.budget_tokens", budget)
-						}
+						out, _ = sjson.Set(out, "thinking.budget_tokens", budget)
 					}
 				}
 			}
@@ -141,17 +138,35 @@ func ConvertOpenAIRequestToClaude(modelName string, inputRawJSON []byte, stream
 
 	// Process messages and transform them to Claude Code format
 	if messages := root.Get("messages"); messages.Exists() && messages.IsArray() {
+		messageIndex := 0
+		systemMessageIndex := -1
 		messages.ForEach(func(_, message gjson.Result) bool {
 			role := message.Get("role").String()
 			contentResult := message.Get("content")
 
 			switch role {
-			case "system", "user", "assistant":
-				// Create Claude Code message with appropriate role mapping
-				if role == "system" {
-					role = "user"
+			case "system":
+				if systemMessageIndex == -1 {
+					systemMsg := `{"role":"user","content":[]}`
+					out, _ = sjson.SetRaw(out, "messages.-1", systemMsg)
+					systemMessageIndex = messageIndex
+					messageIndex++
 				}
-
+				if contentResult.Exists() && contentResult.Type == gjson.String && contentResult.String() != "" {
+					textPart := `{"type":"text","text":""}`
+					textPart, _ = sjson.Set(textPart, "text", contentResult.String())
+					out, _ = sjson.SetRaw(out, fmt.Sprintf("messages.%d.content.-1", systemMessageIndex), textPart)
+				} else if contentResult.Exists() && contentResult.IsArray() {
+					contentResult.ForEach(func(_, part gjson.Result) bool {
+						if part.Get("type").String() == "text" {
+							textPart := `{"type":"text","text":""}`
+							textPart, _ = sjson.Set(textPart, "text", part.Get("text").String())
+							out, _ = sjson.SetRaw(out, fmt.Sprintf("messages.%d.content.-1", systemMessageIndex), textPart)
+						}
+						return true
+					})
+				}
+			case "user", "assistant":
 				msg := `{"role":"","content":[]}`
 				msg, _ = sjson.Set(msg, "role", role)
 
@@ -230,6 +245,7 @@ func ConvertOpenAIRequestToClaude(modelName string, inputRawJSON []byte, stream
 				}
 
 				out, _ = sjson.SetRaw(out, "messages.-1", msg)
+				messageIndex++
 
 			case "tool":
 				// Handle tool result messages conversion
@@ -240,6 +256,7 @@ func ConvertOpenAIRequestToClaude(modelName string, inputRawJSON []byte, stream
 				msg, _ = sjson.Set(msg, "content.0.tool_use_id", toolCallID)
 				msg, _ = sjson.Set(msg, "content.0.content", content)
 				out, _ = sjson.SetRaw(out, "messages.-1", msg)
+				messageIndex++
 			}
 			return true
 		})

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

@@ -10,7 +10,6 @@ import (
 	"strings"
 
 	"github.com/google/uuid"
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -54,23 +53,21 @@ func ConvertOpenAIResponsesRequestToClaude(modelName string, inputRawJSON []byte
 
 	root := gjson.ParseBytes(rawJSON)
 
+	// Convert OpenAI Responses reasoning.effort to Claude thinking config.
 	if v := root.Get("reasoning.effort"); v.Exists() {
-		modelInfo := registry.LookupModelInfo(modelName)
-		if modelInfo != nil && modelInfo.Thinking != nil && len(modelInfo.Thinking.Levels) == 0 {
-			effort := strings.ToLower(strings.TrimSpace(v.String()))
-			if effort != "" {
-				budget, ok := thinking.ConvertLevelToBudget(effort)
-				if ok {
-					switch budget {
-					case 0:
-						out, _ = sjson.Set(out, "thinking.type", "disabled")
-					case -1:
+		effort := strings.ToLower(strings.TrimSpace(v.String()))
+		if effort != "" {
+			budget, ok := thinking.ConvertLevelToBudget(effort)
+			if ok {
+				switch budget {
+				case 0:
+					out, _ = sjson.Set(out, "thinking.type", "disabled")
+				case -1:
+					out, _ = sjson.Set(out, "thinking.type", "enabled")
+				default:
+					if budget > 0 {
 						out, _ = sjson.Set(out, "thinking.type", "enabled")
-					default:
-						if budget > 0 {
-							out, _ = sjson.Set(out, "thinking.type", "enabled")
-							out, _ = sjson.Set(out, "thinking.budget_tokens", budget)
-						}
+						out, _ = sjson.Set(out, "thinking.budget_tokens", budget)
 					}
 				}
 			}

internal/translator/codex/claude/codex_claude_request.go

@@ -12,7 +12,6 @@ import (
 	"strings"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -52,7 +51,7 @@ func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 	systemsResult := rootResult.Get("system")
 	if systemsResult.IsArray() {
 		systemResults := systemsResult.Array()
-		message := `{"type":"message","role":"user","content":[]}`
+		message := `{"type":"message","role":"developer","content":[]}`
 		for i := 0; i < len(systemResults); i++ {
 			systemResult := systemResults[i]
 			systemTypeResult := systemResult.Get("type")
@@ -218,18 +217,15 @@ func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 	// Add additional configuration parameters for the Codex API.
 	template, _ = sjson.Set(template, "parallel_tool_calls", true)
 
-	// Convert thinking.budget_tokens to reasoning.effort for level-based models
-	reasoningEffort := "medium" // default
+	// Convert thinking.budget_tokens to reasoning.effort.
+	reasoningEffort := "medium"
 	if thinkingConfig := rootResult.Get("thinking"); thinkingConfig.Exists() && thinkingConfig.IsObject() {
-		modelInfo := registry.LookupModelInfo(modelName)
 		switch thinkingConfig.Get("type").String() {
 		case "enabled":
-			if modelInfo != nil && modelInfo.Thinking != nil && len(modelInfo.Thinking.Levels) > 0 {
-				if budgetTokens := thinkingConfig.Get("budget_tokens"); budgetTokens.Exists() {
-					budget := int(budgetTokens.Int())
-					if effort, ok := thinking.ConvertBudgetToLevel(budget); ok && effort != "" {
-						reasoningEffort = effort
-					}
+			if budgetTokens := thinkingConfig.Get("budget_tokens"); budgetTokens.Exists() {
+				budget := int(budgetTokens.Int())
+				if effort, ok := thinking.ConvertBudgetToLevel(budget); ok && effort != "" {
+					reasoningEffort = effort
 				}
 			}
 		case "disabled":
@@ -245,21 +241,23 @@ func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 	template, _ = sjson.Set(template, "include", []string{"reasoning.encrypted_content"})
 
 	// Add a first message to ignore system instructions and ensure proper execution.
-	inputResult := gjson.Get(template, "input")
-	if inputResult.Exists() && inputResult.IsArray() {
-		inputResults := inputResult.Array()
-		newInput := "[]"
-		for i := 0; i < len(inputResults); i++ {
-			if i == 0 {
-				firstText := inputResults[i].Get("content.0.text")
-				firstInstructions := "EXECUTE ACCORDING TO THE FOLLOWING INSTRUCTIONS!!!"
-				if firstText.Exists() && firstText.String() != firstInstructions {
-					newInput, _ = sjson.SetRaw(newInput, "-1", `{"type":"message","role":"user","content":[{"type":"input_text","text":"EXECUTE ACCORDING TO THE FOLLOWING INSTRUCTIONS!!!"}]}`)
+	if misc.GetCodexInstructionsEnabled() {
+		inputResult := gjson.Get(template, "input")
+		if inputResult.Exists() && inputResult.IsArray() {
+			inputResults := inputResult.Array()
+			newInput := "[]"
+			for i := 0; i < len(inputResults); i++ {
+				if i == 0 {
+					firstText := inputResults[i].Get("content.0.text")
+					firstInstructions := "EXECUTE ACCORDING TO THE FOLLOWING INSTRUCTIONS!!!"
+					if firstText.Exists() && firstText.String() != firstInstructions {
+						newInput, _ = sjson.SetRaw(newInput, "-1", `{"type":"message","role":"user","content":[{"type":"input_text","text":"EXECUTE ACCORDING TO THE FOLLOWING INSTRUCTIONS!!!"}]}`)
+					}
 				}
+				newInput, _ = sjson.SetRaw(newInput, "-1", inputResults[i].Raw)
 			}
-			newInput, _ = sjson.SetRaw(newInput, "-1", inputResults[i].Raw)
+			template, _ = sjson.SetRaw(template, "input", newInput)
 		}
-		template, _ = sjson.SetRaw(template, "input", newInput)
 	}
 
 	return []byte(template)

internal/translator/codex/gemini/codex_gemini_request.go

@@ -14,7 +14,6 @@ import (
 	"strings"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	"github.com/tidwall/gjson"
@@ -95,7 +94,7 @@ func ConvertGeminiRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 	// System instruction -> as a user message with input_text parts
 	sysParts := root.Get("system_instruction.parts")
 	if sysParts.IsArray() {
-		msg := `{"type":"message","role":"user","content":[]}`
+		msg := `{"type":"message","role":"developer","content":[]}`
 		arr := sysParts.Array()
 		for i := 0; i < len(arr); i++ {
 			p := arr[i]
@@ -249,22 +248,28 @@ func ConvertGeminiRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 	// Fixed flags aligning with Codex expectations
 	out, _ = sjson.Set(out, "parallel_tool_calls", true)
 
-	// Convert thinkingBudget to reasoning.effort for level-based models
-	reasoningEffort := "medium" // default
+	// Convert Gemini thinkingConfig to Codex reasoning.effort.
+	effortSet := false
 	if genConfig := root.Get("generationConfig"); genConfig.Exists() {
 		if thinkingConfig := genConfig.Get("thinkingConfig"); thinkingConfig.Exists() && thinkingConfig.IsObject() {
-			modelInfo := registry.LookupModelInfo(modelName)
-			if modelInfo != nil && modelInfo.Thinking != nil && len(modelInfo.Thinking.Levels) > 0 {
-				if thinkingBudget := thinkingConfig.Get("thinkingBudget"); thinkingBudget.Exists() {
-					budget := int(thinkingBudget.Int())
-					if effort, ok := thinking.ConvertBudgetToLevel(budget); ok && effort != "" {
-						reasoningEffort = effort
-					}
+			if thinkingLevel := thinkingConfig.Get("thinkingLevel"); thinkingLevel.Exists() {
+				effort := strings.ToLower(strings.TrimSpace(thinkingLevel.String()))
+				if effort != "" {
+					out, _ = sjson.Set(out, "reasoning.effort", effort)
+					effortSet = true
+				}
+			} else if thinkingBudget := thinkingConfig.Get("thinkingBudget"); thinkingBudget.Exists() {
+				if effort, ok := thinking.ConvertBudgetToLevel(int(thinkingBudget.Int())); ok {
+					out, _ = sjson.Set(out, "reasoning.effort", effort)
+					effortSet = true
 				}
 			}
 		}
 	}
-	out, _ = sjson.Set(out, "reasoning.effort", reasoningEffort)
+	if !effortSet {
+		// No thinking config, set default effort
+		out, _ = sjson.Set(out, "reasoning.effort", "medium")
+	}
 	out, _ = sjson.Set(out, "reasoning.summary", "auto")
 	out, _ = sjson.Set(out, "stream", true)
 	out, _ = sjson.Set(out, "store", false)

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

@@ -33,7 +33,7 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 	rawJSON := bytes.Clone(inputRawJSON)
 	userAgent := misc.ExtractCodexUserAgent(rawJSON)
 	// Start with empty JSON object
-	out := `{}`
+	out := `{"instructions":""}`
 
 	// Stream must be set to true
 	out, _ = sjson.Set(out, "stream", stream)
@@ -98,7 +98,9 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 	// Extract system instructions from first system message (string or text object)
 	messages := gjson.GetBytes(rawJSON, "messages")
 	_, instructions := misc.CodexInstructionsForModel(modelName, "", userAgent)
-	out, _ = sjson.Set(out, "instructions", instructions)
+	if misc.GetCodexInstructionsEnabled() {
+		out, _ = sjson.Set(out, "instructions", instructions)
+	}
 	// if messages.IsArray() {
 	// 	arr := messages.Array()
 	// 	for i := 0; i < len(arr); i++ {
@@ -141,7 +143,7 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 				msg := `{}`
 				msg, _ = sjson.Set(msg, "type", "message")
 				if role == "system" {
-					msg, _ = sjson.Set(msg, "role", "user")
+					msg, _ = sjson.Set(msg, "role", "developer")
 				} else {
 					msg, _ = sjson.Set(msg, "role", role)
 				}

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

@@ -74,6 +74,11 @@ func ConvertOpenAIResponsesRequestToCodex(modelName string, inputRawJSON []byte,
 	}
 
 	if hasOfficialInstructions {
+		newInput := "[]"
+		for _, item := range inputResults {
+			newInput, _ = sjson.SetRaw(newInput, "-1", item.Raw)
+		}
+		rawJSON, _ = sjson.SetRawBytes(rawJSON, "input", []byte(newInput))
 		return rawJSON
 	}
 	// log.Debugf("instructions not matched, %s\n", originalInstructions)

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

@@ -9,7 +9,6 @@ import (
 	"bytes"
 	"strings"
 
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -161,14 +160,11 @@ func ConvertClaudeRequestToCLI(modelName string, inputRawJSON []byte, _ bool) []
 
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when type==enabled
 	if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() {
-		modelInfo := registry.LookupModelInfo(modelName)
-		if modelInfo != nil && modelInfo.Thinking != nil {
-			if t.Get("type").String() == "enabled" {
-				if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
-					budget := int(b.Int())
-					out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
-					out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
-				}
+		if t.Get("type").String() == "enabled" {
+			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
+				budget := int(b.Int())
+				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
+				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.includeThoughts", true)
 			}
 		}
 	}

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

@@ -129,6 +129,7 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 			}
 		}
 
+		systemPartIndex := 0
 		for i := 0; i < len(arr); i++ {
 			m := arr[i]
 			role := m.Get("role").String()
@@ -138,16 +139,19 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 				// system -> request.systemInstruction as a user message style
 				if content.Type == gjson.String {
 					out, _ = sjson.SetBytes(out, "request.systemInstruction.role", "user")
-					out, _ = sjson.SetBytes(out, "request.systemInstruction.parts.0.text", content.String())
+					out, _ = sjson.SetBytes(out, fmt.Sprintf("request.systemInstruction.parts.%d.text", systemPartIndex), content.String())
+					systemPartIndex++
 				} else if content.IsObject() && content.Get("type").String() == "text" {
 					out, _ = sjson.SetBytes(out, "request.systemInstruction.role", "user")
-					out, _ = sjson.SetBytes(out, "request.systemInstruction.parts.0.text", content.Get("text").String())
+					out, _ = sjson.SetBytes(out, fmt.Sprintf("request.systemInstruction.parts.%d.text", systemPartIndex), content.Get("text").String())
+					systemPartIndex++
 				} else if content.IsArray() {
 					contents := content.Array()
 					if len(contents) > 0 {
 						out, _ = sjson.SetBytes(out, "request.systemInstruction.role", "user")
 						for j := 0; j < len(contents); j++ {
-							out, _ = sjson.SetBytes(out, fmt.Sprintf("request.systemInstruction.parts.%d.text", j), contents[j].Get("text").String())
+							out, _ = sjson.SetBytes(out, fmt.Sprintf("request.systemInstruction.parts.%d.text", systemPartIndex), contents[j].Get("text").String())
+							systemPartIndex++
 						}
 					}
 				}

internal/translator/gemini/claude/gemini_claude_request.go

@@ -9,7 +9,6 @@ import (
 	"bytes"
 	"strings"
 
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -153,16 +152,13 @@ func ConvertClaudeRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 	}
 
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when enabled
-	// Only apply for models that use numeric budgets, not discrete levels.
+	// Translator only does format conversion, ApplyThinking handles model capability validation.
 	if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() {
-		modelInfo := registry.LookupModelInfo(modelName)
-		if modelInfo != nil && modelInfo.Thinking != nil && len(modelInfo.Thinking.Levels) == 0 {
-			if t.Get("type").String() == "enabled" {
-				if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
-					budget := int(b.Int())
-					out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", budget)
-					out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
-				}
+		if t.Get("type").String() == "enabled" {
+			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
+				budget := int(b.Int())
+				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", budget)
+				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.includeThoughts", true)
 			}
 		}
 	}

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

@@ -129,6 +129,7 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 			}
 		}
 
+		systemPartIndex := 0
 		for i := 0; i < len(arr); i++ {
 			m := arr[i]
 			role := m.Get("role").String()
@@ -138,16 +139,19 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 				// system -> system_instruction as a user message style
 				if content.Type == gjson.String {
 					out, _ = sjson.SetBytes(out, "system_instruction.role", "user")
-					out, _ = sjson.SetBytes(out, "system_instruction.parts.0.text", content.String())
+					out, _ = sjson.SetBytes(out, fmt.Sprintf("system_instruction.parts.%d.text", systemPartIndex), content.String())
+					systemPartIndex++
 				} else if content.IsObject() && content.Get("type").String() == "text" {
 					out, _ = sjson.SetBytes(out, "system_instruction.role", "user")
-					out, _ = sjson.SetBytes(out, "system_instruction.parts.0.text", content.Get("text").String())
+					out, _ = sjson.SetBytes(out, fmt.Sprintf("system_instruction.parts.%d.text", systemPartIndex), content.Get("text").String())
+					systemPartIndex++
 				} else if content.IsArray() {
 					contents := content.Array()
 					if len(contents) > 0 {
-						out, _ = sjson.SetBytes(out, "request.systemInstruction.role", "user")
+						out, _ = sjson.SetBytes(out, "system_instruction.role", "user")
 						for j := 0; j < len(contents); j++ {
-							out, _ = sjson.SetBytes(out, fmt.Sprintf("request.systemInstruction.parts.%d.text", j), contents[j].Get("text").String())
+							out, _ = sjson.SetBytes(out, fmt.Sprintf("system_instruction.parts.%d.text", systemPartIndex), contents[j].Get("text").String())
+							systemPartIndex++
 						}
 					}
 				}
@@ -162,7 +166,10 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 					for _, item := range items {
 						switch item.Get("type").String() {
 						case "text":
-							node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".text", item.Get("text").String())
+							text := item.Get("text").String()
+							if text != "" {
+								node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".text", text)
+							}
 							p++
 						case "image_url":
 							imageURL := item.Get("image_url.url").String()
@@ -207,6 +214,10 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 					for _, item := range content.Array() {
 						switch item.Get("type").String() {
 						case "text":
+							text := item.Get("text").String()
+							if text != "" {
+								node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".text", text)
+							}
 							p++
 						case "image_url":
 							// If the assistant returned an inline data URL, preserve it for history fidelity.

internal/translator/kiro/claude/kiro_claude_request.go

@@ -240,9 +240,13 @@ func BuildKiroPayload(claudeBody []byte, modelID, profileArn, origin string, isA
 	// Process messages and build history
 	history, currentUserMsg, currentToolResults := processMessages(messages, modelID, origin)
 
-	// Build content with system prompt
+	// Build content with system prompt (only on first turn to avoid re-injection)
 	if currentUserMsg != nil {
-		currentUserMsg.Content = buildFinalContent(currentUserMsg.Content, systemPrompt, currentToolResults)
+		effectiveSystemPrompt := systemPrompt
+		if len(history) > 0 {
+			effectiveSystemPrompt = "" // Don't re-inject on subsequent turns
+		}
+		currentUserMsg.Content = buildFinalContent(currentUserMsg.Content, effectiveSystemPrompt, currentToolResults)
 
 		// Deduplicate currentToolResults
 		currentToolResults = deduplicateToolResults(currentToolResults)

internal/translator/openai/gemini/openai_gemini_request.go

@@ -77,12 +77,15 @@ func ConvertGeminiRequestToOpenAI(modelName string, inputRawJSON []byte, stream
 			}
 		}
 
-		// Convert thinkingBudget to reasoning_effort
-		// Always perform conversion to support allowCompat models that may not be in registry
+		// Map Gemini thinkingConfig to OpenAI reasoning_effort.
 		if thinkingConfig := genConfig.Get("thinkingConfig"); thinkingConfig.Exists() && thinkingConfig.IsObject() {
-			if thinkingBudget := thinkingConfig.Get("thinkingBudget"); thinkingBudget.Exists() {
-				budget := int(thinkingBudget.Int())
-				if effort, ok := thinking.ConvertBudgetToLevel(budget); ok && effort != "" {
+			if thinkingLevel := thinkingConfig.Get("thinkingLevel"); thinkingLevel.Exists() {
+				effort := strings.ToLower(strings.TrimSpace(thinkingLevel.String()))
+				if effort != "" {
+					out, _ = sjson.Set(out, "reasoning_effort", effort)
+				}
+			} else if thinkingBudget := thinkingConfig.Get("thinkingBudget"); thinkingBudget.Exists() {
+				if effort, ok := thinking.ConvertBudgetToLevel(int(thinkingBudget.Int())); ok {
 					out, _ = sjson.Set(out, "reasoning_effort", effort)
 				}
 			}

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

@@ -12,6 +12,10 @@ import (
 	"github.com/tidwall/sjson"
 )
 
+type oaiToResponsesStateReasoning struct {
+	ReasoningID   string
+	ReasoningData string
+}
 type oaiToResponsesState struct {
 	Seq            int
 	ResponseID     string
@@ -23,6 +27,7 @@ type oaiToResponsesState struct {
 	// Per-output message text buffers by index
 	MsgTextBuf   map[int]*strings.Builder
 	ReasoningBuf strings.Builder
+	Reasonings   []oaiToResponsesStateReasoning
 	FuncArgsBuf  map[int]*strings.Builder // index -> args
 	FuncNames    map[int]string           // index -> name
 	FuncCallIDs  map[int]string           // index -> call_id
@@ -63,6 +68,7 @@ func ConvertOpenAIChatCompletionsResponseToOpenAIResponses(ctx context.Context,
 			MsgItemDone:     make(map[int]bool),
 			FuncArgsDone:    make(map[int]bool),
 			FuncItemDone:    make(map[int]bool),
+			Reasonings:      make([]oaiToResponsesStateReasoning, 0),
 		}
 	}
 	st := (*param).(*oaiToResponsesState)
@@ -157,6 +163,31 @@ func ConvertOpenAIChatCompletionsResponseToOpenAIResponses(ctx context.Context,
 		st.Started = true
 	}
 
+	stopReasoning := func(text string) {
+		// Emit reasoning done events
+		textDone := `{"type":"response.reasoning_summary_text.done","sequence_number":0,"item_id":"","output_index":0,"summary_index":0,"text":""}`
+		textDone, _ = sjson.Set(textDone, "sequence_number", nextSeq())
+		textDone, _ = sjson.Set(textDone, "item_id", st.ReasoningID)
+		textDone, _ = sjson.Set(textDone, "output_index", st.ReasoningIndex)
+		textDone, _ = sjson.Set(textDone, "text", text)
+		out = append(out, emitRespEvent("response.reasoning_summary_text.done", textDone))
+		partDone := `{"type":"response.reasoning_summary_part.done","sequence_number":0,"item_id":"","output_index":0,"summary_index":0,"part":{"type":"summary_text","text":""}}`
+		partDone, _ = sjson.Set(partDone, "sequence_number", nextSeq())
+		partDone, _ = sjson.Set(partDone, "item_id", st.ReasoningID)
+		partDone, _ = sjson.Set(partDone, "output_index", st.ReasoningIndex)
+		partDone, _ = sjson.Set(partDone, "part.text", text)
+		out = append(out, emitRespEvent("response.reasoning_summary_part.done", partDone))
+		outputItemDone := `{"type":"response.output_item.done","item":{"id":"","type":"reasoning","encrypted_content":"","summary":[{"type":"summary_text","text":""}]},"output_index":0,"sequence_number":0}`
+		outputItemDone, _ = sjson.Set(outputItemDone, "sequence_number", nextSeq())
+		outputItemDone, _ = sjson.Set(outputItemDone, "item.id", st.ReasoningID)
+		outputItemDone, _ = sjson.Set(outputItemDone, "output_index", st.ReasoningIndex)
+		outputItemDone, _ = sjson.Set(outputItemDone, "item.summary.text", text)
+		out = append(out, emitRespEvent("response.output_item.done", outputItemDone))
+
+		st.Reasonings = append(st.Reasonings, oaiToResponsesStateReasoning{ReasoningID: st.ReasoningID, ReasoningData: text})
+		st.ReasoningID = ""
+	}
+
 	// choices[].delta content / tool_calls / reasoning_content
 	if choices := root.Get("choices"); choices.Exists() && choices.IsArray() {
 		choices.ForEach(func(_, choice gjson.Result) bool {
@@ -165,6 +196,10 @@ func ConvertOpenAIChatCompletionsResponseToOpenAIResponses(ctx context.Context,
 			if delta.Exists() {
 				if c := delta.Get("content"); c.Exists() && c.String() != "" {
 					// Ensure the message item and its first content part are announced before any text deltas
+					if st.ReasoningID != "" {
+						stopReasoning(st.ReasoningBuf.String())
+						st.ReasoningBuf.Reset()
+					}
 					if !st.MsgItemAdded[idx] {
 						item := `{"type":"response.output_item.added","sequence_number":0,"output_index":0,"item":{"id":"","type":"message","status":"in_progress","content":[],"role":"assistant"}}`
 						item, _ = sjson.Set(item, "sequence_number", nextSeq())
@@ -226,6 +261,10 @@ func ConvertOpenAIChatCompletionsResponseToOpenAIResponses(ctx context.Context,
 
 				// tool calls
 				if tcs := delta.Get("tool_calls"); tcs.Exists() && tcs.IsArray() {
+					if st.ReasoningID != "" {
+						stopReasoning(st.ReasoningBuf.String())
+						st.ReasoningBuf.Reset()
+					}
 					// Before emitting any function events, if a message is open for this index,
 					// close its text/content to match Codex expected ordering.
 					if st.MsgItemAdded[idx] && !st.MsgItemDone[idx] {
@@ -361,17 +400,8 @@ func ConvertOpenAIChatCompletionsResponseToOpenAIResponses(ctx context.Context,
 				}
 
 				if st.ReasoningID != "" {
-					// Emit reasoning done events
-					textDone := `{"type":"response.reasoning_summary_text.done","sequence_number":0,"item_id":"","output_index":0,"summary_index":0,"text":""}`
-					textDone, _ = sjson.Set(textDone, "sequence_number", nextSeq())
-					textDone, _ = sjson.Set(textDone, "item_id", st.ReasoningID)
-					textDone, _ = sjson.Set(textDone, "output_index", st.ReasoningIndex)
-					out = append(out, emitRespEvent("response.reasoning_summary_text.done", textDone))
-					partDone := `{"type":"response.reasoning_summary_part.done","sequence_number":0,"item_id":"","output_index":0,"summary_index":0,"part":{"type":"summary_text","text":""}}`
-					partDone, _ = sjson.Set(partDone, "sequence_number", nextSeq())
-					partDone, _ = sjson.Set(partDone, "item_id", st.ReasoningID)
-					partDone, _ = sjson.Set(partDone, "output_index", st.ReasoningIndex)
-					out = append(out, emitRespEvent("response.reasoning_summary_part.done", partDone))
+					stopReasoning(st.ReasoningBuf.String())
+					st.ReasoningBuf.Reset()
 				}
 
 				// Emit function call done events for any active function calls
@@ -485,11 +515,13 @@ func ConvertOpenAIChatCompletionsResponseToOpenAIResponses(ctx context.Context,
 				}
 				// Build response.output using aggregated buffers
 				outputsWrapper := `{"arr":[]}`
-				if st.ReasoningBuf.Len() > 0 {
-					item := `{"id":"","type":"reasoning","summary":[{"type":"summary_text","text":""}]}`
-					item, _ = sjson.Set(item, "id", st.ReasoningID)
-					item, _ = sjson.Set(item, "summary.0.text", st.ReasoningBuf.String())
-					outputsWrapper, _ = sjson.SetRaw(outputsWrapper, "arr.-1", item)
+				if len(st.Reasonings) > 0 {
+					for _, r := range st.Reasonings {
+						item := `{"id":"","type":"reasoning","summary":[{"type":"summary_text","text":""}]}`
+						item, _ = sjson.Set(item, "id", r.ReasoningID)
+						item, _ = sjson.Set(item, "summary.0.text", r.ReasoningData)
+						outputsWrapper, _ = sjson.SetRaw(outputsWrapper, "arr.-1", item)
+					}
 				}
 				// Append message items in ascending index order
 				if len(st.MsgItemAdded) > 0 {

internal/util/gemini_schema.go

@@ -19,6 +19,7 @@ func CleanJSONSchemaForAntigravity(jsonStr string) string {
 	// Phase 1: Convert and add hints
 	jsonStr = convertRefsToHints(jsonStr)
 	jsonStr = convertConstToEnum(jsonStr)
+	jsonStr = convertEnumValuesToStrings(jsonStr)
 	jsonStr = addEnumHints(jsonStr)
 	jsonStr = addAdditionalPropertiesHints(jsonStr)
 	jsonStr = moveConstraintsToDescription(jsonStr)
@@ -77,6 +78,33 @@ func convertConstToEnum(jsonStr string) string {
 	return jsonStr
 }
 
+// convertEnumValuesToStrings ensures all enum values are strings.
+// 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") {
+		arr := gjson.Get(jsonStr, p)
+		if !arr.IsArray() {
+			continue
+		}
+
+		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)
+		}
+	}
+	return jsonStr
+}
+
 func addEnumHints(jsonStr string) string {
 	for _, p := range findPaths(jsonStr, "enum") {
 		arr := gjson.Get(jsonStr, p)

internal/util/gemini_schema_test.go

@@ -818,3 +818,54 @@ func TestCleanJSONSchemaForAntigravity_MultipleFormats(t *testing.T) {
 		t.Errorf("date-time format hint should be added, got: %s", result)
 	}
 }
+
+func TestCleanJSONSchemaForAntigravity_NumericEnumToString(t *testing.T) {
+	// Gemini API requires enum values to be strings, not numbers
+	input := `{
+		"type": "object",
+		"properties": {
+			"priority": {"type": "integer", "enum": [0, 1, 2]},
+			"level": {"type": "number", "enum": [1.5, 2.5, 3.5]},
+			"status": {"type": "string", "enum": ["active", "inactive"]}
+		}
+	}`
+
+	result := CleanJSONSchemaForAntigravity(input)
+
+	// Numeric enum values should be converted to strings
+	if strings.Contains(result, `"enum":[0,1,2]`) {
+		t.Errorf("Integer enum values should be converted to strings, got: %s", result)
+	}
+	if strings.Contains(result, `"enum":[1.5,2.5,3.5]`) {
+		t.Errorf("Float enum values should be converted to strings, got: %s", result)
+	}
+	// Should contain string versions
+	if !strings.Contains(result, `"0"`) || !strings.Contains(result, `"1"`) || !strings.Contains(result, `"2"`) {
+		t.Errorf("Integer enum values should be converted to string format, got: %s", result)
+	}
+	// String enum values should remain unchanged
+	if !strings.Contains(result, `"active"`) || !strings.Contains(result, `"inactive"`) {
+		t.Errorf("String enum values should remain unchanged, got: %s", result)
+	}
+}
+
+func TestCleanJSONSchemaForAntigravity_BooleanEnumToString(t *testing.T) {
+	// Boolean enum values should also be converted to strings
+	input := `{
+		"type": "object",
+		"properties": {
+			"enabled": {"type": "boolean", "enum": [true, false]}
+		}
+	}`
+
+	result := CleanJSONSchemaForAntigravity(input)
+
+	// Boolean enum values should be converted to strings
+	if strings.Contains(result, `"enum":[true,false]`) {
+		t.Errorf("Boolean enum values should be converted to strings, got: %s", result)
+	}
+	// Should contain string versions "true" and "false"
+	if !strings.Contains(result, `"true"`) || !strings.Contains(result, `"false"`) {
+		t.Errorf("Boolean enum values should be converted to string format, got: %s", result)
+	}
+}

sdk/api/handlers/gemini/gemini_handlers.go

@@ -56,8 +56,12 @@ func (h *GeminiAPIHandler) GeminiModels(c *gin.Context) {
 		for k, v := range model {
 			normalizedModel[k] = v
 		}
-		if name, ok := normalizedModel["name"].(string); ok && name != "" && !strings.HasPrefix(name, "models/") {
-			normalizedModel["name"] = "models/" + name
+		if name, ok := normalizedModel["name"].(string); ok && name != "" {
+			if !strings.HasPrefix(name, "models/") {
+				normalizedModel["name"] = "models/" + name
+			}
+			normalizedModel["displayName"] = name
+			normalizedModel["description"] = name
 		}
 		if _, ok := normalizedModel["supportedGenerationMethods"]; !ok {
 			normalizedModel["supportedGenerationMethods"] = defaultMethods
@@ -85,94 +89,35 @@ func (h *GeminiAPIHandler) GeminiGetHandler(c *gin.Context) {
 		return
 	}
 	action := strings.TrimPrefix(request.Action, "/")
-	switch action {
-	case "gemini-3-pro-preview":
-		c.JSON(http.StatusOK, gin.H{
-			"name":             "models/gemini-3-pro-preview",
-			"version":          "3",
-			"displayName":      "Gemini 3 Pro Preview",
-			"description":      "Gemini 3 Pro Preview",
-			"inputTokenLimit":  1048576,
-			"outputTokenLimit": 65536,
-			"supportedGenerationMethods": []string{
-				"generateContent",
-				"countTokens",
-				"createCachedContent",
-				"batchGenerateContent",
-			},
-			"temperature":    1,
-			"topP":           0.95,
-			"topK":           64,
-			"maxTemperature": 2,
-			"thinking":       true,
-		},
-		)
-	case "gemini-2.5-pro":
-		c.JSON(http.StatusOK, gin.H{
-			"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",
-			},
-			"temperature":    1,
-			"topP":           0.95,
-			"topK":           64,
-			"maxTemperature": 2,
-			"thinking":       true,
-		},
-		)
-	case "gemini-2.5-flash":
-		c.JSON(http.StatusOK, gin.H{
-			"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",
-			},
-			"temperature":    1,
-			"topP":           0.95,
-			"topK":           64,
-			"maxTemperature": 2,
-			"thinking":       true,
-		})
-	case "gpt-5":
-		c.JSON(http.StatusOK, gin.H{
-			"name":             "gpt-5",
-			"version":          "001",
-			"displayName":      "GPT 5",
-			"description":      "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			"inputTokenLimit":  400000,
-			"outputTokenLimit": 128000,
-			"supportedGenerationMethods": []string{
-				"generateContent",
-			},
-			"temperature":    1,
-			"topP":           0.95,
-			"topK":           64,
-			"maxTemperature": 2,
-			"thinking":       true,
-		})
-	default:
-		c.JSON(http.StatusNotFound, handlers.ErrorResponse{
-			Error: handlers.ErrorDetail{
-				Message: "Not Found",
-				Type:    "not_found",
-			},
-		})
+
+	// Get dynamic models from the global registry and find the matching one
+	availableModels := h.Models()
+	var targetModel map[string]any
+
+	for _, model := range availableModels {
+		name, _ := model["name"].(string)
+		// Match name with or without 'models/' prefix
+		if name == action || name == "models/"+action {
+			targetModel = model
+			break
+		}
 	}
+
+	if targetModel != nil {
+		// Ensure the name has 'models/' prefix in the output if it's a Gemini model
+		if name, ok := targetModel["name"].(string); ok && name != "" && !strings.HasPrefix(name, "models/") {
+			targetModel["name"] = "models/" + name
+		}
+		c.JSON(http.StatusOK, targetModel)
+		return
+	}
+
+	c.JSON(http.StatusNotFound, handlers.ErrorResponse{
+		Error: handlers.ErrorDetail{
+			Message: "Not Found",
+			Type:    "not_found",
+		},
+	})
 }
 
 // GeminiHandler handles POST requests for Gemini API operations.

sdk/api/handlers/openai/endpoint_compat.go

@@ -0,0 +1,37 @@
+package openai
+
+import "github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+
+const (
+	openAIChatEndpoint      = "/chat/completions"
+	openAIResponsesEndpoint = "/responses"
+)
+
+func resolveEndpointOverride(modelName, requestedEndpoint string) (string, bool) {
+	if modelName == "" {
+		return "", false
+	}
+	info := registry.GetGlobalRegistry().GetModelInfo(modelName)
+	if info == nil || len(info.SupportedEndpoints) == 0 {
+		return "", false
+	}
+	if endpointListContains(info.SupportedEndpoints, requestedEndpoint) {
+		return "", false
+	}
+	if requestedEndpoint == openAIChatEndpoint && endpointListContains(info.SupportedEndpoints, openAIResponsesEndpoint) {
+		return openAIResponsesEndpoint, true
+	}
+	if requestedEndpoint == openAIResponsesEndpoint && endpointListContains(info.SupportedEndpoints, openAIChatEndpoint) {
+		return openAIChatEndpoint, true
+	}
+	return "", false
+}
+
+func endpointListContains(items []string, value string) bool {
+	for _, item := range items {
+		if item == value {
+			return true
+		}
+	}
+	return false
+}

sdk/api/handlers/openai/openai_handlers.go

@@ -17,6 +17,7 @@ import (
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/constant"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+	codexconverter "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/codex/openai/chat-completions"
 	responsesconverter "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/openai/openai/responses"
 	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers"
 	"github.com/tidwall/gjson"
@@ -112,6 +113,23 @@ func (h *OpenAIAPIHandler) ChatCompletions(c *gin.Context) {
 	streamResult := gjson.GetBytes(rawJSON, "stream")
 	stream := streamResult.Type == gjson.True
 
+	modelName := gjson.GetBytes(rawJSON, "model").String()
+	if overrideEndpoint, ok := resolveEndpointOverride(modelName, openAIChatEndpoint); ok && overrideEndpoint == openAIResponsesEndpoint {
+		originalChat := rawJSON
+		if shouldTreatAsResponsesFormat(rawJSON) {
+			// Already responses-style payload; no conversion needed.
+		} else {
+			rawJSON = codexconverter.ConvertOpenAIRequestToCodex(modelName, rawJSON, stream)
+		}
+		stream = gjson.GetBytes(rawJSON, "stream").Bool()
+		if stream {
+			h.handleStreamingResponseViaResponses(c, rawJSON, originalChat)
+		} else {
+			h.handleNonStreamingResponseViaResponses(c, rawJSON, originalChat)
+		}
+		return
+	}
+
 	// Some clients send OpenAI Responses-format payloads to /v1/chat/completions.
 	// Convert them to Chat Completions so downstream translators preserve tool metadata.
 	if shouldTreatAsResponsesFormat(rawJSON) {
@@ -245,6 +263,76 @@ func convertCompletionsRequestToChatCompletions(rawJSON []byte) []byte {
 	return []byte(out)
 }
 
+func convertResponsesObjectToChatCompletion(ctx context.Context, modelName string, originalChatJSON, responsesRequestJSON, responsesPayload []byte) []byte {
+	if len(responsesPayload) == 0 {
+		return nil
+	}
+	wrapped := wrapResponsesPayloadAsCompleted(responsesPayload)
+	if len(wrapped) == 0 {
+		return nil
+	}
+	var param any
+	converted := codexconverter.ConvertCodexResponseToOpenAINonStream(ctx, modelName, originalChatJSON, responsesRequestJSON, wrapped, &param)
+	if converted == "" {
+		return nil
+	}
+	return []byte(converted)
+}
+
+func wrapResponsesPayloadAsCompleted(payload []byte) []byte {
+	if gjson.GetBytes(payload, "type").Exists() {
+		return payload
+	}
+	if gjson.GetBytes(payload, "object").String() != "response" {
+		return payload
+	}
+	wrapped := `{"type":"response.completed","response":{}}`
+	wrapped, _ = sjson.SetRaw(wrapped, "response", string(payload))
+	return []byte(wrapped)
+}
+
+func writeConvertedResponsesChunk(c *gin.Context, ctx context.Context, modelName string, originalChatJSON, responsesRequestJSON, chunk []byte, param *any) {
+	outputs := codexconverter.ConvertCodexResponseToOpenAI(ctx, modelName, originalChatJSON, responsesRequestJSON, chunk, param)
+	for _, out := range outputs {
+		if out == "" {
+			continue
+		}
+		_, _ = fmt.Fprintf(c.Writer, "data: %s\n\n", out)
+	}
+}
+
+func (h *OpenAIAPIHandler) forwardResponsesAsChatStream(c *gin.Context, flusher http.Flusher, cancel func(error), data <-chan []byte, errs <-chan *interfaces.ErrorMessage, ctx context.Context, modelName string, originalChatJSON, responsesRequestJSON []byte, param *any) {
+	h.ForwardStream(c, flusher, cancel, data, errs, handlers.StreamForwardOptions{
+		WriteChunk: func(chunk []byte) {
+			outputs := codexconverter.ConvertCodexResponseToOpenAI(ctx, modelName, originalChatJSON, responsesRequestJSON, chunk, param)
+			for _, out := range outputs {
+				if out == "" {
+					continue
+				}
+				_, _ = fmt.Fprintf(c.Writer, "data: %s\n\n", out)
+			}
+		},
+		WriteTerminalError: func(errMsg *interfaces.ErrorMessage) {
+			if errMsg == nil {
+				return
+			}
+			status := http.StatusInternalServerError
+			if errMsg.StatusCode > 0 {
+				status = errMsg.StatusCode
+			}
+			errText := http.StatusText(status)
+			if errMsg.Error != nil && errMsg.Error.Error() != "" {
+				errText = errMsg.Error.Error()
+			}
+			body := handlers.BuildErrorResponseBody(status, errText)
+			_, _ = fmt.Fprintf(c.Writer, "data: %s\n\n", string(body))
+		},
+		WriteDone: func() {
+			_, _ = fmt.Fprintf(c.Writer, "data: [DONE]\n\n")
+		},
+	})
+}
+
 // convertChatCompletionsResponseToCompletions converts chat completions API response back to completions format.
 // This ensures the completions endpoint returns data in the expected format.
 //
@@ -435,6 +523,30 @@ func (h *OpenAIAPIHandler) handleNonStreamingResponse(c *gin.Context, rawJSON []
 	cliCancel()
 }
 
+func (h *OpenAIAPIHandler) handleNonStreamingResponseViaResponses(c *gin.Context, rawJSON []byte, originalChatJSON []byte) {
+	c.Header("Content-Type", "application/json")
+
+	modelName := gjson.GetBytes(rawJSON, "model").String()
+	cliCtx, cliCancel := h.GetContextWithCancel(h, c, context.Background())
+	resp, errMsg := h.ExecuteWithAuthManager(cliCtx, OpenaiResponse, modelName, rawJSON, h.GetAlt(c))
+	if errMsg != nil {
+		h.WriteErrorResponse(c, errMsg)
+		cliCancel(errMsg.Error)
+		return
+	}
+	converted := convertResponsesObjectToChatCompletion(cliCtx, modelName, originalChatJSON, rawJSON, resp)
+	if converted == nil {
+		h.WriteErrorResponse(c, &interfaces.ErrorMessage{
+			StatusCode: http.StatusInternalServerError,
+			Error:      fmt.Errorf("failed to convert response to chat completion format"),
+		})
+		cliCancel(fmt.Errorf("response conversion failed"))
+		return
+	}
+	_, _ = c.Writer.Write(converted)
+	cliCancel()
+}
+
 // handleStreamingResponse handles streaming responses for Gemini models.
 // It establishes a streaming connection with the backend service and forwards
 // the response chunks to the client in real-time using Server-Sent Events.
@@ -509,6 +621,67 @@ func (h *OpenAIAPIHandler) handleStreamingResponse(c *gin.Context, rawJSON []byt
 	}
 }
 
+func (h *OpenAIAPIHandler) handleStreamingResponseViaResponses(c *gin.Context, rawJSON []byte, originalChatJSON []byte) {
+	flusher, ok := c.Writer.(http.Flusher)
+	if !ok {
+		c.JSON(http.StatusInternalServerError, handlers.ErrorResponse{
+			Error: handlers.ErrorDetail{
+				Message: "Streaming not supported",
+				Type:    "server_error",
+			},
+		})
+		return
+	}
+
+	modelName := gjson.GetBytes(rawJSON, "model").String()
+	cliCtx, cliCancel := h.GetContextWithCancel(h, c, context.Background())
+	dataChan, errChan := h.ExecuteStreamWithAuthManager(cliCtx, OpenaiResponse, modelName, rawJSON, h.GetAlt(c))
+	var param any
+
+	setSSEHeaders := func() {
+		c.Header("Content-Type", "text/event-stream")
+		c.Header("Cache-Control", "no-cache")
+		c.Header("Connection", "keep-alive")
+		c.Header("Access-Control-Allow-Origin", "*")
+	}
+
+	// Peek for first usable chunk
+	for {
+		select {
+		case <-c.Request.Context().Done():
+			cliCancel(c.Request.Context().Err())
+			return
+		case errMsg, ok := <-errChan:
+			if !ok {
+				errChan = nil
+				continue
+			}
+			h.WriteErrorResponse(c, errMsg)
+			if errMsg != nil {
+				cliCancel(errMsg.Error)
+			} else {
+				cliCancel(nil)
+			}
+			return
+		case chunk, ok := <-dataChan:
+			if !ok {
+				setSSEHeaders()
+				_, _ = fmt.Fprintf(c.Writer, "data: [DONE]\n\n")
+				flusher.Flush()
+				cliCancel(nil)
+				return
+			}
+
+			setSSEHeaders()
+			writeConvertedResponsesChunk(c, cliCtx, modelName, originalChatJSON, rawJSON, chunk, &param)
+			flusher.Flush()
+
+			h.forwardResponsesAsChatStream(c, flusher, func(err error) { cliCancel(err) }, dataChan, errChan, cliCtx, modelName, originalChatJSON, rawJSON, &param)
+			return
+		}
+	}
+}
+
 // handleCompletionsNonStreamingResponse handles non-streaming completions responses.
 // It converts completions request to chat completions format, sends to backend,
 // then converts the response back to completions format before sending to client.

sdk/api/handlers/openai/openai_responses_handlers.go

@@ -16,6 +16,7 @@ import (
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/constant"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+	responsesconverter "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/openai/openai/responses"
 	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers"
 	"github.com/tidwall/gjson"
 )
@@ -83,7 +84,21 @@ func (h *OpenAIResponsesAPIHandler) Responses(c *gin.Context) {
 
 	// Check if the client requested a streaming response.
 	streamResult := gjson.GetBytes(rawJSON, "stream")
-	if streamResult.Type == gjson.True {
+	stream := streamResult.Type == gjson.True
+
+	modelName := gjson.GetBytes(rawJSON, "model").String()
+	if overrideEndpoint, ok := resolveEndpointOverride(modelName, openAIResponsesEndpoint); ok && overrideEndpoint == openAIChatEndpoint {
+		chatJSON := responsesconverter.ConvertOpenAIResponsesRequestToOpenAIChatCompletions(modelName, rawJSON, stream)
+		stream = gjson.GetBytes(chatJSON, "stream").Bool()
+		if stream {
+			h.handleStreamingResponseViaChat(c, rawJSON, chatJSON)
+		} else {
+			h.handleNonStreamingResponseViaChat(c, rawJSON, chatJSON)
+		}
+		return
+	}
+
+	if stream {
 		h.handleStreamingResponse(c, rawJSON)
 	} else {
 		h.handleNonStreamingResponse(c, rawJSON)
@@ -116,6 +131,31 @@ func (h *OpenAIResponsesAPIHandler) handleNonStreamingResponse(c *gin.Context, r
 	cliCancel()
 }
 
+func (h *OpenAIResponsesAPIHandler) handleNonStreamingResponseViaChat(c *gin.Context, originalResponsesJSON, chatJSON []byte) {
+	c.Header("Content-Type", "application/json")
+
+	modelName := gjson.GetBytes(chatJSON, "model").String()
+	cliCtx, cliCancel := h.GetContextWithCancel(h, c, context.Background())
+	resp, errMsg := h.ExecuteWithAuthManager(cliCtx, OpenAI, modelName, chatJSON, "")
+	if errMsg != nil {
+		h.WriteErrorResponse(c, errMsg)
+		cliCancel(errMsg.Error)
+		return
+	}
+	var param any
+	converted := responsesconverter.ConvertOpenAIChatCompletionsResponseToOpenAIResponsesNonStream(cliCtx, modelName, originalResponsesJSON, originalResponsesJSON, resp, &param)
+	if converted == "" {
+		h.WriteErrorResponse(c, &interfaces.ErrorMessage{
+			StatusCode: http.StatusInternalServerError,
+			Error:      fmt.Errorf("failed to convert chat completion response to responses format"),
+		})
+		cliCancel(fmt.Errorf("response conversion failed"))
+		return
+	}
+	_, _ = c.Writer.Write([]byte(converted))
+	cliCancel()
+}
+
 // handleStreamingResponse handles streaming responses for Gemini models.
 // It establishes a streaming connection with the backend service and forwards
 // the response chunks to the client in real-time using Server-Sent Events.
@@ -196,6 +236,116 @@ func (h *OpenAIResponsesAPIHandler) handleStreamingResponse(c *gin.Context, rawJ
 	}
 }
 
+func (h *OpenAIResponsesAPIHandler) handleStreamingResponseViaChat(c *gin.Context, originalResponsesJSON, chatJSON []byte) {
+	flusher, ok := c.Writer.(http.Flusher)
+	if !ok {
+		c.JSON(http.StatusInternalServerError, handlers.ErrorResponse{
+			Error: handlers.ErrorDetail{
+				Message: "Streaming not supported",
+				Type:    "server_error",
+			},
+		})
+		return
+	}
+
+	modelName := gjson.GetBytes(chatJSON, "model").String()
+	cliCtx, cliCancel := h.GetContextWithCancel(h, c, context.Background())
+	dataChan, errChan := h.ExecuteStreamWithAuthManager(cliCtx, OpenAI, modelName, chatJSON, "")
+	var param any
+
+	setSSEHeaders := func() {
+		c.Header("Content-Type", "text/event-stream")
+		c.Header("Cache-Control", "no-cache")
+		c.Header("Connection", "keep-alive")
+		c.Header("Access-Control-Allow-Origin", "*")
+	}
+
+	for {
+		select {
+		case <-c.Request.Context().Done():
+			cliCancel(c.Request.Context().Err())
+			return
+		case errMsg, ok := <-errChan:
+			if !ok {
+				errChan = nil
+				continue
+			}
+			h.WriteErrorResponse(c, errMsg)
+			if errMsg != nil {
+				cliCancel(errMsg.Error)
+			} else {
+				cliCancel(nil)
+			}
+			return
+		case chunk, ok := <-dataChan:
+			if !ok {
+				setSSEHeaders()
+				_, _ = c.Writer.Write([]byte("\n"))
+				flusher.Flush()
+				cliCancel(nil)
+				return
+			}
+
+			setSSEHeaders()
+			writeChatAsResponsesChunk(c, cliCtx, modelName, originalResponsesJSON, chunk, &param)
+			flusher.Flush()
+
+			h.forwardChatAsResponsesStream(c, flusher, func(err error) { cliCancel(err) }, dataChan, errChan, cliCtx, modelName, originalResponsesJSON, &param)
+			return
+		}
+	}
+}
+
+func writeChatAsResponsesChunk(c *gin.Context, ctx context.Context, modelName string, originalResponsesJSON, chunk []byte, param *any) {
+	outputs := responsesconverter.ConvertOpenAIChatCompletionsResponseToOpenAIResponses(ctx, modelName, originalResponsesJSON, originalResponsesJSON, chunk, param)
+	for _, out := range outputs {
+		if out == "" {
+			continue
+		}
+		if bytes.HasPrefix([]byte(out), []byte("event:")) {
+			_, _ = c.Writer.Write([]byte("\n"))
+		}
+		_, _ = c.Writer.Write([]byte(out))
+		_, _ = c.Writer.Write([]byte("\n"))
+	}
+}
+
+func (h *OpenAIResponsesAPIHandler) forwardChatAsResponsesStream(c *gin.Context, flusher http.Flusher, cancel func(error), data <-chan []byte, errs <-chan *interfaces.ErrorMessage, ctx context.Context, modelName string, originalResponsesJSON []byte, param *any) {
+	h.ForwardStream(c, flusher, cancel, data, errs, handlers.StreamForwardOptions{
+		WriteChunk: func(chunk []byte) {
+			outputs := responsesconverter.ConvertOpenAIChatCompletionsResponseToOpenAIResponses(ctx, modelName, originalResponsesJSON, originalResponsesJSON, chunk, param)
+			for _, out := range outputs {
+				if out == "" {
+					continue
+				}
+				if bytes.HasPrefix([]byte(out), []byte("event:")) {
+					_, _ = c.Writer.Write([]byte("\n"))
+				}
+				_, _ = c.Writer.Write([]byte(out))
+				_, _ = c.Writer.Write([]byte("\n"))
+			}
+		},
+		WriteTerminalError: func(errMsg *interfaces.ErrorMessage) {
+			if errMsg == nil {
+				return
+			}
+			status := http.StatusInternalServerError
+			if errMsg.StatusCode > 0 {
+				status = errMsg.StatusCode
+			}
+			errText := http.StatusText(status)
+			if errMsg.Error != nil && errMsg.Error.Error() != "" {
+				errText = errMsg.Error.Error()
+			}
+			body := handlers.BuildErrorResponseBody(status, errText)
+			_, _ = fmt.Fprintf(c.Writer, "\nevent: error\ndata: %s\n\n", string(body))
+		},
+		WriteDone: func() {
+			_, _ = c.Writer.Write([]byte("\n"))
+		},
+	})
+}
+
 func (h *OpenAIResponsesAPIHandler) forwardResponsesStream(c *gin.Context, flusher http.Flusher, cancel func(error), data <-chan []byte, errs <-chan *interfaces.ErrorMessage) {
 	h.ForwardStream(c, flusher, cancel, data, errs, handlers.StreamForwardOptions{
 		WriteChunk: func(chunk []byte) {

sdk/cliproxy/auth/oauth_model_alias.go

@@ -221,7 +221,7 @@ func modelAliasChannel(auth *Auth) string {
 // and auth kind. Returns empty string if the provider/authKind combination doesn't support
 // OAuth model alias (e.g., API key authentication).
 //
-// Supported channels: gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow.
+// Supported channels: gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow, kiro, github-copilot.
 func OAuthModelAliasChannel(provider, authKind string) string {
 	provider = strings.ToLower(strings.TrimSpace(provider))
 	authKind = strings.ToLower(strings.TrimSpace(authKind))
@@ -245,7 +245,7 @@ func OAuthModelAliasChannel(provider, authKind string) string {
 			return ""
 		}
 		return "codex"
-	case "gemini-cli", "aistudio", "antigravity", "qwen", "iflow":
+	case "gemini-cli", "aistudio", "antigravity", "qwen", "iflow", "kiro", "github-copilot":
 		return provider
 	default:
 		return ""

sdk/cliproxy/auth/oauth_model_alias_test.go

@@ -43,6 +43,15 @@ func TestResolveOAuthUpstreamModel_SuffixPreservation(t *testing.T) {
 			input:   "gemini-2.5-pro",
 			want:    "gemini-2.5-pro-exp-03-25",
 		},
+		{
+			name: "kiro alias resolves",
+			aliases: map[string][]internalconfig.OAuthModelAlias{
+				"kiro": {{Name: "kiro-claude-sonnet-4-5", Alias: "sonnet"}},
+			},
+			channel: "kiro",
+			input:   "sonnet",
+			want:    "kiro-claude-sonnet-4-5",
+		},
 		{
 			name: "config suffix takes priority",
 			aliases: map[string][]internalconfig.OAuthModelAlias{
@@ -152,6 +161,8 @@ func createAuthForChannel(channel string) *Auth {
 		return &Auth{Provider: "qwen"}
 	case "iflow":
 		return &Auth{Provider: "iflow"}
+	case "kiro":
+		return &Auth{Provider: "kiro"}
 	default:
 		return &Auth{Provider: channel}
 	}

sdk/cliproxy/auth/types.go

@@ -227,6 +227,18 @@ func (a *Auth) AccountInfo() (string, string) {
 		}
 	}
 
+	// For GitHub provider, return username
+	if strings.ToLower(a.Provider) == "github" {
+		if a.Metadata != nil {
+			if username, ok := a.Metadata["username"].(string); ok {
+				username = strings.TrimSpace(username)
+				if username != "" {
+					return "oauth", username
+				}
+			}
+		}
+	}
+
 	// Check metadata for email first (OAuth-style auth)
 	if a.Metadata != nil {
 		if v, ok := a.Metadata["email"].(string); ok {

sdk/cliproxy/service.go

@@ -1222,6 +1222,9 @@ func rewriteModelInfoName(name, oldID, newID string) string {
 	if strings.EqualFold(oldID, newID) {
 		return name
 	}
+	if strings.EqualFold(trimmed, oldID) {
+		return newID
+	}
 	if strings.HasSuffix(trimmed, "/"+oldID) {
 		prefix := strings.TrimSuffix(trimmed, oldID)
 		return prefix + newID