diff --git a/README.md b/README.md
index 2c97044ddc..127653ce82 100644
--- a/README.md
+++ b/README.md
@@ -718,7 +718,7 @@ Sub2API supports both Grok subscription accounts through xAI OAuth and standard
 - Codex CLI style Responses WebSocket ingress is accepted on the Responses targets and bridged to xAI HTTP/SSE Responses upstream
 - Text models: `grok-4.5`, `grok-4.3`, `grok-build-0.1`, `grok-composer-2.5-fast`, `grok-4.20-0309-reasoning`, `grok-4.20-0309-non-reasoning`, and `grok-4.20-multi-agent-0309`
 - Media targets for Grok groups: `/v1/images/generations`, `/images/generations`, `/v1/images/edits`, `/images/edits`, `/v1/videos/generations`, `/videos/generations`, `/v1/videos/edits`, `/videos/edits`, `/v1/videos/extensions`, `/videos/extensions`, `/v1/videos/{request_id}`, and `/videos/{request_id}`. Generation, editing, and extension requests require the group image-generation permission.
-- Media models: `grok-imagine`, `grok-imagine-image-quality`, `grok-imagine-image`, `grok-imagine-edit`, `grok-imagine-video`, and `grok-imagine-video-1.5`
+- Media models: `grok-imagine`, `grok-imagine-image-quality`, `grok-imagine-image`, `grok-imagine-image-2.0`, `grok-imagine-edit`, `grok-imagine-video`, and `grok-imagine-video-1.5`
 - JSON image-edit and video-generation requests accept image references in `image`, `images`, `reference_images`, and `mask` objects. Use `url` for xAI-compatible payloads; the legacy `image_url` field remains accepted and is normalized to `url` before forwarding.
 - Out of scope for this provider: TTS, transcription, browser automation, cookies, and Grok web scraping
 
diff --git a/backend/internal/config/config.go b/backend/internal/config/config.go
index 0fa0058a5d..048f2d76e9 100644
--- a/backend/internal/config/config.go
+++ b/backend/internal/config/config.go
@@ -934,6 +934,9 @@ type GatewayConfig struct {
 	// OpenAIResponseHeaderTimeout: OpenAI/Codex 上游等待响应头的超时时间（秒），0表示无超时
 	// OpenAI/Codex 请求可能在上游排队较久；默认不使用通用响应头超时截断。
 	OpenAIResponseHeaderTimeout int `mapstructure:"openai_response_header_timeout"`
+	// GrokResponseHeaderTimeout bounds the pre-first-byte wait for xAI/Grok.
+	// A zero value uses the provider-safe default instead of the generic gateway timeout.
+	GrokResponseHeaderTimeout int `mapstructure:"grok_response_header_timeout"`
 	// OpenAIFirstOutputTimeoutSeconds: native HTTP Responses 首个语义输出超时（秒），0表示禁用。
 	OpenAIFirstOutputTimeoutSeconds int `mapstructure:"openai_first_output_timeout_seconds"`
 	// OpenAIHighEffortFirstOutputTimeoutSeconds: high/xhigh/max 推理的首个语义输出超时（秒）。
@@ -2327,6 +2330,7 @@ func setDefaults() {
 	// Gateway
 	viper.SetDefault("gateway.response_header_timeout", 600) // 600秒(10分钟)等待上游响应头，LLM高负载时可能排队较久
 	viper.SetDefault("gateway.openai_response_header_timeout", 0)
+	viper.SetDefault("gateway.grok_response_header_timeout", 120)
 	viper.SetDefault("gateway.openai_first_output_timeout_seconds", 0)
 	viper.SetDefault("gateway.openai_high_effort_first_output_timeout_seconds", 0)
 	viper.SetDefault("gateway.log_upstream_error_body", true)
@@ -3241,6 +3245,9 @@ func (c *Config) Validate() error {
 	if c.Gateway.OpenAIResponseHeaderTimeout < 0 {
 		return fmt.Errorf("gateway.openai_response_header_timeout must be non-negative")
 	}
+	if c.Gateway.GrokResponseHeaderTimeout < 0 || c.Gateway.GrokResponseHeaderTimeout > 1800 {
+		return fmt.Errorf("gateway.grok_response_header_timeout must be between 0-1800 seconds")
+	}
 	if c.Gateway.OpenAIFirstOutputTimeoutSeconds < 0 || c.Gateway.OpenAIFirstOutputTimeoutSeconds > 600 ||
 		(c.Gateway.OpenAIFirstOutputTimeoutSeconds > 0 && c.Gateway.OpenAIFirstOutputTimeoutSeconds < 30) {
 		return fmt.Errorf("gateway.openai_first_output_timeout_seconds must be 0 or between 30-600 seconds")
diff --git a/backend/internal/handler/failover_loop.go b/backend/internal/handler/failover_loop.go
index 8b4b389a3f..45bea91b26 100644
--- a/backend/internal/handler/failover_loop.go
+++ b/backend/internal/handler/failover_loop.go
@@ -56,10 +56,13 @@ const (
 const profitVetoExhaustedMessage = "No available accounts: all candidates rejected by group profit control"
 
 func sameAccountRetryDelayFor(failoverErr *service.UpstreamFailoverError, retryCount int) time.Duration {
-	if failoverErr != nil && failoverErr.SameAccountRetryDelay > 0 {
+	if failoverErr == nil {
+		return sameAccountRetryDelay
+	}
+	if failoverErr.SameAccountRetryDelay > 0 {
 		return failoverErr.SameAccountRetryDelay
 	}
-	if failoverErr == nil || !failoverErr.RequestScopedTransient || retryCount <= 1 {
+	if !failoverErr.RequestScopedTransient || retryCount <= 1 {
 		return sameAccountRetryDelay
 	}
 
@@ -77,10 +80,37 @@ func sameAccountRetryAllowed(failoverErr *service.UpstreamFailoverError, retryCo
 	if failoverErr == nil || !failoverErr.RetryableOnSameAccount {
 		return false
 	}
-	if !failoverErr.SameAccountRetryDeadline.IsZero() {
-		return time.Now().Before(failoverErr.SameAccountRetryDeadline)
+	if !sameAccountRetryDeadlineAllows(failoverErr) {
+		return false
 	}
-	return retryCount < retryLimit
+	// Deadline-window retries (OAuth 429) may pass retryLimit=0 and are not
+	// bound to pool_mode_retry_count. Pool-mode callers pass a positive limit.
+	if !failoverErr.SameAccountRetryDeadline.IsZero() && retryLimit <= 0 {
+		return true
+	}
+	if failoverErr.SameAccountRetryMax > 0 && (retryLimit <= 0 || failoverErr.SameAccountRetryMax < retryLimit) {
+		retryLimit = failoverErr.SameAccountRetryMax
+	}
+	return retryLimit > 0 && retryCount < retryLimit
+}
+
+// sameAccountRetryDeadlineAllows prevents a retry from starting after the
+// service-provided same-account retry window has elapsed.
+func sameAccountRetryDeadlineAllows(failoverErr *service.UpstreamFailoverError) bool {
+	return failoverErr == nil || failoverErr.SameAccountRetryDeadline.IsZero() || time.Now().Before(failoverErr.SameAccountRetryDeadline)
+}
+
+// effectiveSameAccountRetryLimit applies an error-specific cap without
+// overriding an explicit account setting of zero (which disables retries).
+func effectiveSameAccountRetryLimit(failoverErr *service.UpstreamFailoverError, account *service.Account) int {
+	if account == nil {
+		return 0
+	}
+	limit := account.GetPoolModeRetryCount()
+	if limit > 0 && failoverErr != nil && failoverErr.SameAccountRetryMax > 0 && failoverErr.SameAccountRetryMax < limit {
+		return failoverErr.SameAccountRetryMax
+	}
+	return limit
 }
 
 // FailoverState 跨循环迭代共享的 failover 状态
@@ -171,14 +201,18 @@ func (s *FailoverState) HandleFailoverError(
 	}
 
 	// 同账号重试不算切换账号，粘性会话仅在实际切换时强制缓存计费。
-	sameAccountRetry := sameAccountRetryAllowed(failoverErr, s.SameAccountRetryCount[accountID], retryLimit)
+	retryCount := s.SameAccountRetryCount[accountID]
+	if failoverErr.SameAccountRetryMax > 0 && failoverErr.SameAccountRetryMax < retryLimit {
+		retryLimit = failoverErr.SameAccountRetryMax
+	}
+	sameAccountRetry := failoverErr.RetryableOnSameAccount && retryLimit > 0 && retryCount < retryLimit && sameAccountRetryDeadlineAllows(failoverErr)
 	if needForceCacheBilling(s.hasBoundSession, failoverErr, sameAccountRetry) {
 		s.ForceCacheBilling = true
 	}
 
 	// 同账号重试：对 RetryableOnSameAccount 的临时性错误，先在同一账号上重试。
 	// 重试次数上限 retryLimit 由调用方传入（账号级 pool_mode_retry_count 配置）。
-	if sameAccountRetryAllowed(failoverErr, s.SameAccountRetryCount[accountID], retryLimit) {
+	if sameAccountRetry {
 		s.SameAccountRetryCount[accountID]++
 		retryDelay := sameAccountRetryDelayFor(failoverErr, s.SameAccountRetryCount[accountID])
 		logger.FromContext(ctx).Warn("gateway.failover_same_account_retry",
diff --git a/backend/internal/handler/failover_loop_test.go b/backend/internal/handler/failover_loop_test.go
index 6ab0079b8c..20ce869c01 100644
--- a/backend/internal/handler/failover_loop_test.go
+++ b/backend/internal/handler/failover_loop_test.go
@@ -75,6 +75,23 @@ func TestSameAccountRetryAllowedUsesDeadlineInsteadOfPoolCount(t *testing.T) {
 	require.False(t, sameAccountRetryAllowed(err, 0, 100))
 }
 
+func TestSameAccountRetryDeadlineAllows(t *testing.T) {
+	require.True(t, sameAccountRetryDeadlineAllows(&service.UpstreamFailoverError{}))
+	require.True(t, sameAccountRetryDeadlineAllows(&service.UpstreamFailoverError{
+		SameAccountRetryDeadline: time.Now().Add(time.Second),
+	}))
+	require.False(t, sameAccountRetryDeadlineAllows(&service.UpstreamFailoverError{
+		SameAccountRetryDeadline: time.Now().Add(-time.Second),
+	}))
+}
+
+func TestEffectiveSameAccountRetryLimitHonorsErrorCapAndDisabledAccount(t *testing.T) {
+	account := &service.Account{Type: service.AccountTypeAPIKey, Credentials: map[string]any{"pool_mode": true, "pool_mode_retry_count": float64(3)}}
+	require.Equal(t, 1, effectiveSameAccountRetryLimit(&service.UpstreamFailoverError{SameAccountRetryMax: 1}, account))
+	account.Credentials["pool_mode_retry_count"] = float64(0)
+	require.Equal(t, 0, effectiveSameAccountRetryLimit(&service.UpstreamFailoverError{SameAccountRetryMax: 1}, account))
+}
+
 // ---------------------------------------------------------------------------
 // Helper
 // ---------------------------------------------------------------------------
@@ -335,7 +352,7 @@ func TestHandleFailoverError_CacheBilling(t *testing.T) {
 		require.Zero(t, fs.SwitchCount)
 	})
 
-	t.Run("deadline允许超过计数上限时仍不强制缓存计费", func(t *testing.T) {
+	t.Run("deadline存在但计数已耗尽时按切换处理并强制缓存计费", func(t *testing.T) {
 		mock := &mockTempUnscheduler{}
 		fs := NewFailoverState(3, true)
 		fs.SameAccountRetryCount[100] = maxSameAccountRetries
@@ -345,8 +362,8 @@ func TestHandleFailoverError_CacheBilling(t *testing.T) {
 
 		fs.HandleFailoverError(context.Background(), mock, 100, "openai", maxSameAccountRetries, err)
 
-		require.False(t, fs.ForceCacheBilling)
-		require.Zero(t, fs.SwitchCount)
+		require.True(t, fs.ForceCacheBilling)
+		require.Equal(t, 1, fs.SwitchCount)
 	})
 
 	t.Run("同账号重试耗尽并实际切换时设置ForceCacheBilling", func(t *testing.T) {
diff --git a/backend/internal/handler/grok_audio.go b/backend/internal/handler/grok_audio.go
index 4ac1ae53bf..28ae40a39b 100644
--- a/backend/internal/handler/grok_audio.go
+++ b/backend/internal/handler/grok_audio.go
@@ -44,39 +44,84 @@ func (h *OpenAIGatewayHandler) GrokRealtime(c *gin.Context) {
 		return
 	}
 
-	selection, _, err := h.gatewayService.SelectAccountWithSchedulerForCapability(
-		c.Request.Context(),
-		apiKey.GroupID,
-		"",
-		"",
-		"grok-4.5",
-		nil,
-		service.OpenAIUpstreamTransportHTTPSSE,
-		// Grok only advertises chat_completions + media capabilities on HEAD.
-		service.OpenAIEndpointCapabilityChatCompletions,
-		false,
-		false,
-		false,
-		service.PlatformGrok,
-	)
-	if err != nil || selection == nil || selection.Account == nil {
-		h.errorResponse(c, http.StatusServiceUnavailable, "api_error", "No available Grok accounts")
-		return
-	}
-
-	var streamStarted bool
 	reqLog := requestLogger(c, "handler.openai_gateway.grok_realtime")
-	release, slotStatus := h.acquireResponsesAccountSlot(c, apiKey.GroupID, "", selection, true, &streamStarted, reqLog)
-	if slotStatus != openAISlotAcquireOK {
+	model := c.Query("model")
+	if strings.TrimSpace(model) == "" {
+		model = "grok-voice-latest"
+	}
+	// Keep the HTTP response uncommitted while selecting and probing an account.
+	// Realtime is not an HTTP streaming response; using reqStream=true here would
+	// let the wait queue flush an SSE ping before the WebSocket handshake succeeds.
+	failed := map[int64]struct{}{}
+	var selection *service.AccountSelectionResult
+	var release func()
+	var token string
+	var upstream *service.GrokRealtimeUpstream
+	var candidateSeen bool
+	for attempts := 0; attempts < 4; attempts++ {
+		// Realtime's voice model is not a text-model capability. Passing a
+		// concrete text model here would reject accounts mapped only to an
+		// older/default text model before the upstream handshake can decide.
+		// An empty requested model keeps account selection capability-based;
+		// the actual voice model remains in the upstream WS query below.
+		candidate, _, selectErr := h.gatewayService.SelectAccountWithSchedulerForCapability(
+			c.Request.Context(), apiKey.GroupID, "", "", "", failed,
+			service.OpenAIUpstreamTransportHTTPSSE,
+			service.OpenAIEndpointCapabilityChatCompletions,
+			false, false, false, service.PlatformGrok,
+		)
+		if selectErr != nil || candidate == nil || candidate.Account == nil {
+			break
+		}
+		candidateSeen = true
+		account := candidate.Account
+		var streamStarted bool
+		var slotStatus openAISlotAcquireResult
+		release, slotStatus = h.acquireResponsesAccountSlot(c, apiKey.GroupID, "", candidate, false, &streamStarted, reqLog)
+		if slotStatus != openAISlotAcquireOK {
+			if slotStatus == openAISlotAcquireFailed {
+				return
+			}
+			failed[account.ID] = struct{}{}
+			continue
+		}
+		var credErr error
+		token, _, credErr = h.gatewayService.GetRequestCredential(c.Request.Context(), c, account)
+		if credErr != nil {
+			release()
+			release = nil
+			failed[account.ID] = struct{}{}
+			continue
+		}
+		probeCtx, cancelProbe := context.WithTimeout(c.Request.Context(), service.DefaultGrokRealtimeDialTimeout)
+		candidateUpstream, openErr := h.gatewayService.OpenGrokRealtime(probeCtx, account, token, model)
+		cancelProbe()
+		if openErr != nil {
+			reqLog.Warn("grok_realtime.pre_accept_failed", zap.Int64("account_id", account.ID), zap.Error(openErr))
+			statusCode := http.StatusBadGateway
+			var dialErr *service.GrokRealtimeDialError
+			if errors.As(openErr, &dialErr) && dialErr.StatusCode > 0 {
+				statusCode = dialErr.StatusCode
+			}
+			h.gatewayService.HandleGrokRealtimeUpstreamError(c.Request.Context(), account, statusCode, []byte(openErr.Error()))
+			release()
+			release = nil
+			failed[account.ID] = struct{}{}
+			continue
+		}
+		selection, upstream = candidate, candidateUpstream
+		break
+	}
+	if selection == nil || selection.Account == nil || release == nil || upstream == nil {
+		if !candidateSeen {
+			h.errorResponse(c, http.StatusServiceUnavailable, "api_error", "No available Grok accounts")
+		} else {
+			h.errorResponse(c, http.StatusBadGateway, "upstream_error", "Grok realtime upstream unavailable")
+		}
 		return
 	}
 	defer release()
-
-	token, _, err := h.gatewayService.GetRequestCredential(c.Request.Context(), c, selection.Account)
-	if err != nil {
-		h.errorResponse(c, http.StatusBadGateway, "upstream_error", "Grok credential unavailable")
-		return
-	}
+	defer func() { _ = upstream.Close() }()
 
 	conn, err := coderws.Accept(c.Writer, c.Request, &coderws.AcceptOptions{CompressionMode: coderws.CompressionContextTakeover})
 	if err != nil {
@@ -84,12 +129,8 @@ func (h *OpenAIGatewayHandler) GrokRealtime(c *gin.Context) {
 	}
 	defer func() { _ = conn.CloseNow() }()
 
-	model := c.Query("model")
-	if strings.TrimSpace(model) == "" {
-		model = "grok-voice-latest"
-	}
 	started := time.Now()
-	audioObserved, proxyErr := h.gatewayService.ProxyGrokRealtime(c.Request.Context(), c, conn, selection.Account, token, model)
+	audioObserved, proxyErr := h.gatewayService.ProxyGrokRealtimeConn(c.Request.Context(), c, conn, upstream)
 	elapsed := time.Since(started)
 	if proxyErr != nil {
 		reqLog.Info("grok_realtime.proxy_failed", zap.Error(proxyErr))
diff --git a/backend/internal/handler/grok_media.go b/backend/internal/handler/grok_media.go
index ba0b601cfa..89845d0908 100644
--- a/backend/internal/handler/grok_media.go
+++ b/backend/internal/handler/grok_media.go
@@ -361,19 +361,21 @@ func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.
 					return
 				}
 				if failoverErr.RetryableOnSameAccount {
-					retryLimit := account.GetPoolModeRetryCount()
-					if sameAccountRetryCount[account.ID] < retryLimit {
+					retryLimit := effectiveSameAccountRetryLimit(failoverErr, account)
+					if sameAccountRetryCount[account.ID] < retryLimit && sameAccountRetryDeadlineAllows(failoverErr) {
 						sameAccountRetryCount[account.ID]++
+						retryDelay := sameAccountRetryDelayFor(failoverErr, sameAccountRetryCount[account.ID])
 						reqLog.Warn("grok_media.pool_mode_same_account_retry",
 							zap.Int64("account_id", account.ID),
 							zap.Int("upstream_status", failoverErr.StatusCode),
 							zap.Int("retry_limit", retryLimit),
 							zap.Int("retry_count", sameAccountRetryCount[account.ID]),
+							zap.Duration("retry_delay", retryDelay),
 						)
 						select {
 						case <-requestCtx.Done():
 							return
-						case <-time.After(sameAccountRetryDelay):
+						case <-time.After(retryDelay):
 						}
 						continue
 					}
diff --git a/backend/internal/handler/openai_chat_completions.go b/backend/internal/handler/openai_chat_completions.go
index 6918cfa198..5eaa8b3514 100644
--- a/backend/internal/handler/openai_chat_completions.go
+++ b/backend/internal/handler/openai_chat_completions.go
@@ -330,8 +330,8 @@ func (h *OpenAIGatewayHandler) ChatCompletions(c *gin.Context) {
 					}
 					// Pool mode: retry on the same account
 					if failoverErr.RetryableOnSameAccount {
-						retryLimit := account.GetPoolModeRetryCount()
-						if sameAccountRetryAllowed(failoverErr, sameAccountRetryCount[account.ID], retryLimit) {
+						retryLimit := effectiveSameAccountRetryLimit(failoverErr, account)
+						if sameAccountRetryCount[account.ID] < retryLimit && sameAccountRetryDeadlineAllows(failoverErr) {
 							sameAccountRetryCount[account.ID]++
 							retryDelay := sameAccountRetryDelayFor(failoverErr, sameAccountRetryCount[account.ID])
 							reqLog.Warn("openai_chat_completions.pool_mode_same_account_retry",
diff --git a/backend/internal/handler/openai_gateway_handler.go b/backend/internal/handler/openai_gateway_handler.go
index 9f6ed2e257..dd111267c4 100644
--- a/backend/internal/handler/openai_gateway_handler.go
+++ b/backend/internal/handler/openai_gateway_handler.go
@@ -754,8 +754,8 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 					}
 					// 池模式：同账号重试
 					if failoverErr.RetryableOnSameAccount {
-						retryLimit := account.GetPoolModeRetryCount()
-						if sameAccountRetryAllowed(failoverErr, sameAccountRetryCount[account.ID], retryLimit) {
+						retryLimit := effectiveSameAccountRetryLimit(failoverErr, account)
+						if sameAccountRetryCount[account.ID] < retryLimit && sameAccountRetryDeadlineAllows(failoverErr) {
 							sameAccountRetryCount[account.ID]++
 							retryDelay := sameAccountRetryDelayFor(failoverErr, sameAccountRetryCount[account.ID])
 							reqLog.Warn("openai.pool_mode_same_account_retry",
@@ -1309,8 +1309,8 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
 					}
 					// 池模式：同账号重试
 					if failoverErr.RetryableOnSameAccount {
-						retryLimit := account.GetPoolModeRetryCount()
-						if sameAccountRetryAllowed(failoverErr, sameAccountRetryCount[account.ID], retryLimit) {
+						retryLimit := effectiveSameAccountRetryLimit(failoverErr, account)
+						if sameAccountRetryCount[account.ID] < retryLimit && sameAccountRetryDeadlineAllows(failoverErr) {
 							sameAccountRetryCount[account.ID]++
 							retryDelay := sameAccountRetryDelayFor(failoverErr, sameAccountRetryCount[account.ID])
 							reqLog.Warn("openai_messages.pool_mode_same_account_retry",
diff --git a/backend/internal/handler/openai_gateway_handler_test.go b/backend/internal/handler/openai_gateway_handler_test.go
index f3d8f77477..6cc0eaffe7 100644
--- a/backend/internal/handler/openai_gateway_handler_test.go
+++ b/backend/internal/handler/openai_gateway_handler_test.go
@@ -672,7 +672,7 @@ func TestResolveOpenAIMessagesDispatchMappedModel(t *testing.T) {
 				Platform: service.PlatformGrok,
 			},
 		}
-		require.Equal(t, "grok-4.5", resolveOpenAIMessagesDispatchMappedModel(nil, apiKey, "claude-sonnet-4-5"))
+		require.Equal(t, "grok-4.6", resolveOpenAIMessagesDispatchMappedModel(nil, apiKey, "claude-sonnet-4-5"))
 		require.Empty(t, resolveOpenAIMessagesDispatchMappedModel(nil, apiKey, "grok"))
 	})
 
diff --git a/backend/internal/handler/openai_images.go b/backend/internal/handler/openai_images.go
index cb1f42bc1e..bac19329e2 100644
--- a/backend/internal/handler/openai_images.go
+++ b/backend/internal/handler/openai_images.go
@@ -308,8 +308,8 @@ func (h *OpenAIGatewayHandler) Images(c *gin.Context) {
 						return
 					}
 					if failoverErr.RetryableOnSameAccount {
-						retryLimit := account.GetPoolModeRetryCount()
-						if sameAccountRetryAllowed(failoverErr, sameAccountRetryCount[account.ID], retryLimit) {
+						retryLimit := effectiveSameAccountRetryLimit(failoverErr, account)
+						if sameAccountRetryCount[account.ID] < retryLimit && sameAccountRetryDeadlineAllows(failoverErr) {
 							sameAccountRetryCount[account.ID]++
 							retryDelay := sameAccountRetryDelayFor(failoverErr, sameAccountRetryCount[account.ID])
 							reqLog.Warn("openai.images.pool_mode_same_account_retry",
diff --git a/backend/internal/pkg/xai/models.go b/backend/internal/pkg/xai/models.go
index 990a6d28e0..6fab34dc39 100644
--- a/backend/internal/pkg/xai/models.go
+++ b/backend/internal/pkg/xai/models.go
@@ -51,24 +51,25 @@ type Model struct {
 // DefaultTextModel is the built-in fallback for empty model fields and Grok
 // text aliases (e.g. "grok", "grok-latest"). Operators may override the runtime
 // default via settings key grok_default_text_model.
-const DefaultTextModel = "grok-4.5"
+const DefaultTextModel = "grok-4.6"
 
 // Official Imagine model IDs (https://docs.x.ai/docs/models).
 const (
 	DefaultImagineImageQualityModel  = "grok-imagine-image-quality"
 	DefaultImagineImageFastModel     = "grok-imagine-image"
+	DefaultImagineImage20Model       = "grok-imagine-image-2.0"
 	DefaultImagineVideoModel         = "grok-imagine-video"
-	DefaultImagineVideo15LegacyModel = "grok-imagine-video-1.5"
-	DefaultImagineVideo15Model       = "grok-imagine-video-1.5-preview"
+	DefaultImagineVideo15Model       = "grok-imagine-video-1.5"
+	DefaultImagineVideo15LegacyModel = "grok-imagine-video-1.5-preview"
 )
 
 // ModelMappingOptions controls optional expansions of the default mapping.
 // Cross-client wildcards (gpt-*/claude-*) default ON via settings
 // grok_cross_client_model_map_enabled so Codex/Claude clients keep working
-// against Grok groups (map to DefaultText / grok-4.5). Operators may disable.
+// against Grok groups (map to DefaultText / grok-4.6). Operators may disable.
 type ModelMappingOptions struct {
 	// DefaultText is the target for empty models and optional cross-client maps.
-	// Empty → DefaultTextModel (grok-4.5).
+	// Empty → DefaultTextModel (grok-4.6).
 	DefaultText string
 	// EnableCrossClientMap merges gpt-*/codex-*/o*/claude-* → DefaultText.
 	EnableCrossClientMap bool
@@ -86,8 +87,6 @@ var defaultModels = []Model{
 	{ID: "grok-4.6", Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok 4.6"},
 	{ID: "grok-4.5", Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok 4.5"},
 	{ID: "grok-4.3", Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok 4.3"},
-	{ID: "grok-3-mini", Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok 3 Mini"},
-	{ID: "grok-3-mini-fast", Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok 3 Mini Fast"},
 	{ID: "grok-build-0.1", Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok Build 0.1"},
 	{ID: "grok-composer-2.5-fast", Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok Composer 2.5 Fast"},
 	{ID: "grok-4.20-0309-reasoning", Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok 4.20 Reasoning"},
@@ -96,9 +95,9 @@ var defaultModels = []Model{
 	// Imagine
 	{ID: DefaultImagineImageQualityModel, Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok Imagine Image Quality"},
 	{ID: DefaultImagineImageFastModel, Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok Imagine Image"},
+	{ID: DefaultImagineImage20Model, Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok Imagine Image 2.0"},
 	{ID: DefaultImagineVideoModel, Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok Imagine Video"},
-	{ID: DefaultImagineVideo15Model, Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok Imagine Video 1.5 Preview"},
-	{ID: DefaultImagineVideo15LegacyModel, Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok Imagine Video 1.5 Legacy"},
+	{ID: DefaultImagineVideo15Model, Object: "model", Type: "model", OwnedBy: "xai", DisplayName: "Grok Imagine Video 1.5"},
 }
 
 // grokTextResponsesModelAliases is the source of truth for Grok text models
@@ -109,14 +108,14 @@ var grokTextResponsesModelAliases = map[string]string{
 	"grok-latest":                  DefaultTextModel,
 	"grok-4.6":                     "grok-4.6",
 	"grok-4.6-latest":              "grok-4.6",
-	"grok-4.5":                     DefaultTextModel,
-	"grok-4.5-latest":              DefaultTextModel,
+	"grok-4.5":                     "grok-4.5",
+	"grok-4.5-latest":              "grok-4.5",
 	"grok-4.3":                     "grok-4.3",
 	"grok-4.3-latest":              "grok-4.3",
 	"grok-3-mini":                  "grok-3-mini",
 	"grok-3-mini-fast":             "grok-3-mini-fast",
 	"grok-build":                   "grok-build-0.1",
-	"grok-build-latest":            DefaultTextModel,
+	"grok-build-latest":            "grok-build-0.1",
 	"grok-build-0.1":               "grok-build-0.1",
 	"grok-composer-2.5-fast":       "grok-composer-2.5-fast",
 	"grok-composer":                "grok-composer-2.5-fast",
@@ -162,7 +161,7 @@ func ModelMappingWithOptions(opts ModelMappingOptions) map[string]string {
 	}
 	for alias, canonical := range grokTextResponsesModelAliases {
 		// Remap aliases that pointed at DefaultTextModel constant to runtime default.
-		if canonical == DefaultTextModel {
+		if (alias == "grok" || alias == "grok-latest") && canonical == DefaultTextModel {
 			mapping[alias] = defaultText
 		} else {
 			mapping[alias] = canonical
@@ -179,7 +178,7 @@ func ModelMappingWithOptions(opts ModelMappingOptions) map[string]string {
 	// Keep official IDs as identity so client-requested model strings are not
 	// rewritten on the wire (pricing still canonicalizes 1.5* via CanonicalImagineVideoModel).
 	mapping["grok-imagine-video"] = DefaultImagineVideoModel
-	mapping["grok-imagine-video-1.5"] = DefaultImagineVideo15LegacyModel
+	mapping["grok-imagine-video-1.5"] = DefaultImagineVideo15Model
 	mapping["grok-imagine-video-1.5-preview"] = DefaultImagineVideo15Model
 	// Informal alias only:
 	mapping["grok-video-1.5"] = DefaultImagineVideo15Model
@@ -273,7 +272,7 @@ func ResolveGrokTextResponsesModelID(model string, defaultText ...string) string
 	}
 	normalized := strings.ToLower(StripGrokProviderPrefix(trimmed))
 	if canonical, ok := grokTextResponsesModelAliases[normalized]; ok {
-		if canonical == DefaultTextModel {
+		if (normalized == "grok" || normalized == "grok-latest") && canonical == DefaultTextModel {
 			return fallback
 		}
 		return canonical
diff --git a/backend/internal/pkg/xai/models_test.go b/backend/internal/pkg/xai/models_test.go
index a01d993ccb..fb4860ee4d 100644
--- a/backend/internal/pkg/xai/models_test.go
+++ b/backend/internal/pkg/xai/models_test.go
@@ -12,14 +12,14 @@ func TestDefaultModelMappingExcludesCrossClientWildcards(t *testing.T) {
 	SetRuntimeModelMappingOptions(ModelMappingOptions{})
 	mapping := DefaultModelMapping()
 
-	require.Equal(t, "grok-4.5", mapping["grok"])
-	require.Equal(t, "grok-4.5", mapping["grok-latest"])
+	require.Equal(t, "grok-4.6", mapping["grok"])
+	require.Equal(t, "grok-4.6", mapping["grok-latest"])
 	require.Equal(t, "grok-build-0.1", mapping["grok-build"])
-	require.Equal(t, DefaultTextModel, mapping["grok-build-latest"])
+	require.Equal(t, "grok-build-0.1", mapping["grok-build-latest"])
 	require.Equal(t, DefaultImagineImageQualityModel, mapping["grok-imagine-edit"])
-	require.Equal(t, DefaultImagineVideo15LegacyModel, mapping["grok-imagine-video-1.5"])
+	require.Equal(t, DefaultImagineVideo15Model, mapping["grok-imagine-video-1.5"])
 	require.Equal(t, DefaultImagineVideo15Model, mapping["grok-imagine-video-1.5-preview"])
-	require.Equal(t, "grok-4.5", mapping["xai/grok"])
+	require.Equal(t, "grok-4.6", mapping["xai/grok"])
 
 	// Cross-vendor wildcards must stay opt-in.
 	_, hasGPT := mapping["gpt-*"]
@@ -68,7 +68,19 @@ func TestDefaultModelsIncludesGrok46(t *testing.T) {
 
 func TestResolveGrokTextResponsesModelID(t *testing.T) {
 	t.Parallel()
-	require.Equal(t, "grok-4.5", ResolveGrokTextResponsesModelID(""))
+	require.Equal(t, "grok-4.6", ResolveGrokTextResponsesModelID(""))
 	require.Equal(t, "grok-4.3", ResolveGrokTextResponsesModelID("grok", "grok-4.3"))
 	require.Equal(t, "grok-4.20-multi-agent-0309", ResolveGrokTextResponsesModelID("grok-4.20-multi-agent"))
 }
+
+func TestExplicitGrok45DoesNotFollowRuntimeDefault(t *testing.T) {
+	require.Equal(t, "grok-4.5", ResolveGrokTextResponsesModelID("grok-4.5", "grok-4.6"))
+	require.Equal(t, "grok-4.5", ResolveGrokTextResponsesModelID("grok-4.5-latest", "grok-4.6"))
+	require.Equal(t, "grok-4.6", ResolveGrokTextResponsesModelID("grok", "grok-4.6"))
+}
+
+func TestBareGrokAliasesFollowGrok46Default(t *testing.T) {
+	require.Equal(t, "grok-4.6", ResolveGrokTextResponsesModelID("grok"))
+	require.Equal(t, "grok-4.6", ResolveGrokTextResponsesModelID("grok-latest"))
+	require.Equal(t, "grok-build-0.1", ResolveGrokTextResponsesModelID("grok-build-latest"))
+}
diff --git a/backend/internal/pkg/xai/oauth_test.go b/backend/internal/pkg/xai/oauth_test.go
index 90036ff8f2..c7b1577be4 100644
--- a/backend/internal/pkg/xai/oauth_test.go
+++ b/backend/internal/pkg/xai/oauth_test.go
@@ -350,14 +350,14 @@ func TestDefaultModelMappingIncludesGrokAliases(t *testing.T) {
 	t.Cleanup(func() { SetRuntimeModelMappingOptions(original) })
 	SetRuntimeModelMappingOptions(ModelMappingOptions{})
 	mapping := DefaultModelMapping()
-	require.Equal(t, "grok-4.5", mapping["grok"])
-	require.Equal(t, "grok-4.5", mapping["grok-latest"])
+	require.Equal(t, "grok-4.6", mapping["grok"])
+	require.Equal(t, "grok-4.6", mapping["grok-latest"])
 	require.Equal(t, "grok-4.6", mapping["grok-4.6"])
 	require.Equal(t, "grok-4.6", mapping["grok-4.6-latest"])
 	require.Equal(t, "grok-4.5", mapping["grok-4.5"])
 	require.Equal(t, "grok-4.5", mapping["grok-4.5-latest"])
 	require.Equal(t, "grok-build-0.1", mapping["grok-build"])
-	require.Equal(t, "grok-4.5", mapping["grok-build-latest"])
+	require.Equal(t, "grok-build-0.1", mapping["grok-build-latest"])
 	require.Equal(t, "grok-composer-2.5-fast", mapping["grok-composer"])
 	require.Equal(t, "grok-composer-2.5-fast", mapping["composer-2.5"])
 	require.Equal(t, "grok-4.20-0309-reasoning", mapping["grok-4.20-reasoning"])
@@ -368,7 +368,7 @@ func TestDefaultModelMappingIncludesGrokAliases(t *testing.T) {
 	require.Equal(t, DefaultImagineImageQualityModel, mapping["grok-imagine-image-quality"])
 	require.Equal(t, DefaultImagineImageQualityModel, mapping["grok-imagine-edit"])
 	require.Equal(t, DefaultImagineVideoModel, mapping["grok-imagine-video"])
-	require.Equal(t, DefaultImagineVideo15LegacyModel, mapping["grok-imagine-video-1.5"])
+	require.Equal(t, DefaultImagineVideo15Model, mapping["grok-imagine-video-1.5"])
 	require.Equal(t, DefaultImagineVideo15Model, mapping["grok-imagine-video-1.5-preview"])
 	_, hasGPT := mapping["gpt-*"]
 	require.False(t, hasGPT, "cross-client wildcards must be opt-in")
diff --git a/backend/internal/pkg/xai/usage.go b/backend/internal/pkg/xai/usage.go
new file mode 100644
index 0000000000..77aa382900
--- /dev/null
+++ b/backend/internal/pkg/xai/usage.go
@@ -0,0 +1,25 @@
+package xai
+
+// IncludeIndependentReasoningTokens adds reasoning tokens to billed output
+// only when total_tokens proves they are not already folded into output.
+//
+// Official xAI Chat Completions example: prompt=32, completion=9,
+// reasoning=94, total=135. Responses example: input=32, output=9,
+// reasoning=110, total=151. OpenAI's canonical completion_tokens already
+// includes reasoning and total equals input+output.
+func IncludeIndependentReasoningTokens(input, output, total, reasoning int64) int64 {
+	if input < 0 || output < 0 || reasoning <= 0 || total <= 0 {
+		return output
+	}
+	if total == input+output {
+		return output
+	}
+	gap := total - input - output
+	if gap <= 0 {
+		return output
+	}
+	if reasoning < gap {
+		gap = reasoning
+	}
+	return output + gap
+}
diff --git a/backend/internal/pkg/xai/usage_test.go b/backend/internal/pkg/xai/usage_test.go
new file mode 100644
index 0000000000..79b8f8639c
--- /dev/null
+++ b/backend/internal/pkg/xai/usage_test.go
@@ -0,0 +1,30 @@
+package xai
+
+import "testing"
+
+func TestIncludeIndependentReasoningTokens(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name                         string
+		input, output, total, reason int64
+		want                         int64
+	}{
+		{name: "xAI chat example", input: 32, output: 9, total: 135, reason: 94, want: 103},
+		{name: "xAI responses example", input: 32, output: 9, total: 151, reason: 110, want: 119},
+		{name: "OpenAI inclusive output", input: 32, output: 103, total: 135, reason: 94, want: 103},
+		{name: "gap smaller than reasoning", input: 10, output: 20, total: 33, reason: 5, want: 23},
+		{name: "absent total", input: 10, output: 20, total: 0, reason: 5, want: 20},
+		{name: "total already complete", input: 10, output: 20, total: 30, reason: 5, want: 20},
+		{name: "inconsistent undershoot", input: 10, output: 20, total: 25, reason: 5, want: 20},
+		{name: "zero reasoning", input: 10, output: 20, total: 35, reason: 0, want: 20},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+			got := IncludeIndependentReasoningTokens(tt.input, tt.output, tt.total, tt.reason)
+			if got != tt.want {
+				t.Fatalf("got %d, want %d", got, tt.want)
+			}
+		})
+	}
+}
diff --git a/backend/internal/repository/group_usage_rollup_trigger_integration_test.go b/backend/internal/repository/group_usage_rollup_trigger_integration_test.go
index db182be495..32016e3c0a 100644
--- a/backend/internal/repository/group_usage_rollup_trigger_integration_test.go
+++ b/backend/internal/repository/group_usage_rollup_trigger_integration_test.go
@@ -151,6 +151,7 @@ func TestGroupUsageRollupTriggerSerializesInsertTransactionAcrossMidnight(t *tes
 
 	insertTx := beginGroupUsageRollupTriggerTestTx(t, ctx, schema)
 	defer func() { _ = insertTx.Rollback() }()
+	require.NoError(t, setGroupUsageRollupTriggerTimeZone(ctx, insertTx, "Asia/Shanghai"))
 	var insertBackendPID int
 	require.NoError(t, insertTx.QueryRowContext(ctx, "SELECT pg_backend_pid()").Scan(&insertBackendPID))
 
@@ -206,6 +207,7 @@ func TestGroupUsageRollupTriggerKeepsWatermarkForTodayInsert(t *testing.T) {
 
 	tx := beginGroupUsageRollupTriggerTestTx(t, ctx, schema)
 	defer func() { _ = tx.Rollback() }()
+	require.NoError(t, setGroupUsageRollupTriggerTimeZone(ctx, tx, "Asia/Shanghai"))
 	_, err := tx.ExecContext(ctx, `
 		INSERT INTO groups (id) VALUES (10);
 		INSERT INTO users (id) VALUES (1);
@@ -476,6 +478,11 @@ func setGroupUsageRollupTriggerSearchPath(ctx context.Context, tx *sql.Tx, quote
 	return err
 }
 
+func setGroupUsageRollupTriggerTimeZone(ctx context.Context, tx *sql.Tx, name string) error {
+	_, err := tx.ExecContext(ctx, "SET LOCAL TIME ZONE "+pq.QuoteLiteral(name))
+	return err
+}
+
 func waitForGroupUsageRollupStateLock(
 	ctx context.Context,
 	backendPID int,
diff --git a/backend/internal/repository/http_upstream.go b/backend/internal/repository/http_upstream.go
index 904ff09eb9..bdd6478cff 100644
--- a/backend/internal/repository/http_upstream.go
+++ b/backend/internal/repository/http_upstream.go
@@ -99,6 +99,7 @@ const (
 	upstreamProtocolModeOpenAIH1         = "openai_h1"
 	upstreamProtocolModeOpenAIH2         = "openai_h2"
 	upstreamProtocolModeOpenAIH1Fallback = "openai_h1_fallback"
+	upstreamProtocolModeGrok             = "grok"
 )
 
 var errUpstreamClientLimitReached = errors.New("upstream client cache limit reached")
@@ -899,12 +900,20 @@ func (s *httpUpstreamService) resolvePoolSettings(isolation string, accountConcu
 }
 
 func (s *httpUpstreamService) applyProfilePoolSettings(settings poolSettings, profile service.HTTPUpstreamProfile) poolSettings {
-	if profile != service.HTTPUpstreamProfileOpenAI {
-		return settings
-	}
-	settings.responseHeaderTimeout = 0
-	if s != nil && s.cfg != nil && s.cfg.Gateway.OpenAIResponseHeaderTimeout > 0 {
-		settings.responseHeaderTimeout = time.Duration(s.cfg.Gateway.OpenAIResponseHeaderTimeout) * time.Second
+	switch profile {
+	case service.HTTPUpstreamProfileOpenAI:
+		settings.responseHeaderTimeout = 0
+		if s != nil && s.cfg != nil && s.cfg.Gateway.OpenAIResponseHeaderTimeout > 0 {
+			settings.responseHeaderTimeout = time.Duration(s.cfg.Gateway.OpenAIResponseHeaderTimeout) * time.Second
+		}
+	case service.HTTPUpstreamProfileGrok:
+		// Grok can stall before its first byte under capacity pressure. Keep the
+		// generic 600s gateway timeout from turning one request into a 10-minute
+		// resource hold; streaming after headers is unaffected.
+		settings.responseHeaderTimeout = 120 * time.Second
+		if s != nil && s.cfg != nil {
+			settings.responseHeaderTimeout = time.Duration(s.cfg.Gateway.GrokResponseHeaderTimeout) * time.Second
+		}
 	}
 	return settings
 }
@@ -983,6 +992,9 @@ func (s *httpUpstreamService) resolveOpenAIHTTP2Settings() openAIHTTP2Settings {
 }
 
 func (s *httpUpstreamService) resolveProtocolMode(profile service.HTTPUpstreamProfile, proxyKey string, parsedProxy *url.URL) string {
+	if profile == service.HTTPUpstreamProfileGrok {
+		return upstreamProtocolModeGrok
+	}
 	if profile != service.HTTPUpstreamProfileOpenAI {
 		return upstreamProtocolModeDefault
 	}
diff --git a/backend/internal/server/api_contract_test.go b/backend/internal/server/api_contract_test.go
index 7f3b25f763..2bca68d3be 100644
--- a/backend/internal/server/api_contract_test.go
+++ b/backend/internal/server/api_contract_test.go
@@ -886,7 +886,7 @@ func TestAPIContracts(t *testing.T) {
 						"invitation_code_enabled": false,
 						"home_content": "",
 					"hide_ccs_import_button": false,
-					"grok_default_text_model": "grok-4.5",
+					"grok_default_text_model": "grok-4.6",
 					"grok_default_base_url_mode": "cli",
 					"grok_cross_client_model_map_enabled": true,
 					"purchase_subscription_enabled": false,
@@ -1169,7 +1169,7 @@ func TestAPIContracts(t *testing.T) {
 					"doc_url": "",
 					"home_content": "",
 					"hide_ccs_import_button": false,
-					"grok_default_text_model": "grok-4.5",
+					"grok_default_text_model": "grok-4.6",
 					"grok_default_base_url_mode": "cli",
 					"grok_cross_client_model_map_enabled": true,
 					"purchase_subscription_enabled": false,
diff --git a/backend/internal/service/account_test_service.go b/backend/internal/service/account_test_service.go
index 49d5c76295..68c59f25af 100644
--- a/backend/internal/service/account_test_service.go
+++ b/backend/internal/service/account_test_service.go
@@ -101,7 +101,7 @@ const (
 	AccountTestModeGrokRealtime = "realtime"
 
 	defaultGrokRealtimeTestModel = "grok-voice-latest"
-	grokRealtimeProbeTimeout     = 12 * time.Second
+	grokRealtimeProbeTimeout     = DefaultGrokRealtimeDialTimeout
 )
 
 // isOpenAIImageModel checks if the model is an OpenAI image generation model (e.g. gpt-image-2).
diff --git a/backend/internal/service/billing_service.go b/backend/internal/service/billing_service.go
index af8c694495..6345f2c40e 100644
--- a/backend/internal/service/billing_service.go
+++ b/backend/internal/service/billing_service.go
@@ -667,7 +667,8 @@ func (s *BillingService) initFallbackPricing() {
 		SupportsCacheBreakdown:  false,
 	}
 
-	// xAI Grok 4.5: $2 input / $0.30 cached input / $6 output below 200k.
+	// xAI Grok 4.5: $2 input / $0.30 cached input / $6 output below 200k;
+	// long-context rates are $4 / $0.60 / $12 (>=200k prompt tokens).
 	s.fallbackPrices["grok-4.5"] = &ModelPricing{
 		InputPricePerToken:            2e-6,
 		OutputPricePerToken:           6e-6,
@@ -679,9 +680,8 @@ func (s *BillingService) initFallbackPricing() {
 		LongContextOutputMultiplier:   2,
 	}
 
-	// xAI Grok 4.6 (docs.x.ai/developers/models: $2 input / $0.50 cached input /
-	// $6 output per MTok under 200k prompt tokens; ≥200k is 2× on input,
-	// cached input, and output).
+	// xAI Grok 4.6: $2 input / $0.50 cached input / $6 output below 200k;
+	// long-context rates are $4 / $1 / $12 (>=200k prompt tokens).
 	s.fallbackPrices["grok-4.6"] = &ModelPricing{
 		InputPricePerToken:            2e-6,
 		OutputPricePerToken:           6e-6,
@@ -693,7 +693,8 @@ func (s *BillingService) initFallbackPricing() {
 		LongContextOutputMultiplier:   2,
 	}
 
-	// xAI Grok 4.3: $1.25 input / $0.20 cached / $2.50 output below 200k.
+	// xAI Grok 4.3: $1.25 input / $0.20 cached / $2.50 output below 200k;
+	// long-context rates are $2.50 / $0.40 / $5.
 	s.fallbackPrices["grok-4.3"] = &ModelPricing{
 		InputPricePerToken:            1.25e-6,
 		OutputPricePerToken:           2.5e-6,
@@ -704,6 +705,33 @@ func (s *BillingService) initFallbackPricing() {
 		LongContextInputMultiplier:    2,
 		LongContextOutputMultiplier:   2,
 	}
+	// Grok 4.20 variants share the official $1.25 / $0.20 / $2.50 card
+	// (and $2.50 / $0.40 / $5 long-context rates) with Grok 4.3.
+	s.fallbackPrices["grok-4.20"] = &ModelPricing{
+		InputPricePerToken:            1.25e-6,
+		OutputPricePerToken:           2.5e-6,
+		CacheReadPricePerToken:        0.2e-6,
+		SupportsCacheBreakdown:        false,
+		LongContextInputThreshold:     200000,
+		LongContextThresholdInclusive: true,
+		LongContextInputMultiplier:    2,
+		LongContextOutputMultiplier:   2,
+	}
+
+	// Keep legacy Grok 3 Mini requests on their own historical xAI price card;
+	// otherwise the generic Grok fallback bills them as Grok 4.5.
+	s.fallbackPrices["grok-3-mini"] = &ModelPricing{
+		InputPricePerToken:     0.30e-6,
+		OutputPricePerToken:    0.50e-6,
+		CacheReadPricePerToken: 0.075e-6,
+		SupportsCacheBreakdown: false,
+	}
+	s.fallbackPrices["grok-3-mini-fast"] = &ModelPricing{
+		InputPricePerToken:     0.60e-6,
+		OutputPricePerToken:    4e-6,
+		CacheReadPricePerToken: 0.15e-6,
+		SupportsCacheBreakdown: false,
+	}
 	// xAI Grok Build 0.1 (official docs: $1 input / $0.20 cached input /
 	// $2 output per MTok). Composer is available only through Grok Build and
 	// has no standalone public API rate card, so its aliases use this coding
@@ -909,17 +937,22 @@ func (s *BillingService) getFallbackPricing(model string) *ModelPricing {
 	}
 
 	switch modelLower {
-	case "grok", "grok-latest", "grok-4.5", "grok-4.5-latest":
-		return s.fallbackPrices["grok-4.5"]
-	case "grok-4.6", "grok-4.6-latest":
+	case "grok", "grok-latest", "grok-4.6", "grok-4.6-latest":
 		return s.fallbackPrices["grok-4.6"]
-	case "grok-4.3",
-		"grok-4.20-0309-reasoning",
+	case "grok-4.5", "grok-4.5-latest":
+		return s.fallbackPrices["grok-4.5"]
+	case "grok-3-mini":
+		return s.fallbackPrices["grok-3-mini"]
+	case "grok-3-mini-fast":
+		return s.fallbackPrices["grok-3-mini-fast"]
+	case "grok-4.3":
+		return s.fallbackPrices["grok-4.3"]
+	case "grok-4.20-0309-reasoning",
 		"grok-4.20-0309-non-reasoning",
 		"grok-4.20-multi-agent-0309",
 		"grok-4.20-reasoning",
 		"grok-4.20-non-reasoning":
-		return s.fallbackPrices["grok-4.3"]
+		return s.fallbackPrices["grok-4.20"]
 	case "grok-build", "grok-build-latest", "grok-build-0.1", "grok-composer", "grok-composer-2.5-fast", "composer-2.5":
 		return s.fallbackPrices["grok-build-0.1"]
 	}
@@ -937,7 +970,7 @@ func (s *BillingService) grokUnknownTextFamilyFallback(model string) *ModelPrici
 	if s == nil || !isGrokUnknownTextFamilyModel(model) {
 		return nil
 	}
-	return s.fallbackPrices["grok-4.5"]
+	return s.fallbackPrices["grok-4.6"]
 }
 
 func isGrokUnknownTextFamilyModel(model string) bool {
diff --git a/backend/internal/service/billing_service_test.go b/backend/internal/service/billing_service_test.go
index 1e801d8e95..7248878a2e 100644
--- a/backend/internal/service/billing_service_test.go
+++ b/backend/internal/service/billing_service_test.go
@@ -1179,7 +1179,7 @@ func TestCalculateCostWithLongContext_PropagatesError(t *testing.T) {
 func TestGetModelPricing_Grok45OfficialFallback(t *testing.T) {
 	svc := newTestBillingService()
 
-	for _, model := range []string{"grok", "grok-latest", "grok-4.5", "grok-4.5-latest"} {
+	for _, model := range []string{"grok-4.5", "grok-4.5-latest"} {
 		model := model
 		t.Run(model, func(t *testing.T) {
 			pricing, err := svc.GetModelPricing(model)
@@ -1192,6 +1192,17 @@ func TestGetModelPricing_Grok45OfficialFallback(t *testing.T) {
 	}
 }
 
+func TestGetModelPricing_GrokBareAliasesUseGrok46(t *testing.T) {
+	svc := newTestBillingService()
+	for _, model := range []string{"grok", "grok-latest"} {
+		pricing, err := svc.GetModelPricing(model)
+		require.NoError(t, err)
+		require.InDelta(t, 2e-6, pricing.InputPricePerToken, 1e-12)
+		require.InDelta(t, 0.5e-6, pricing.CacheReadPricePerToken, 1e-12)
+		require.InDelta(t, 6e-6, pricing.OutputPricePerToken, 1e-12)
+	}
+}
+
 func TestGetModelPricing_Grok46OfficialFallback(t *testing.T) {
 	svc := newTestBillingService()
 
@@ -1211,6 +1222,25 @@ func TestGetModelPricing_Grok46OfficialFallback(t *testing.T) {
 	}
 }
 
+func TestGetModelPricing_GrokOfficialFamilyCards(t *testing.T) {
+	svc := newTestBillingService()
+	for _, tc := range []struct {
+		model                 string
+		input, cached, output float64
+	}{
+		{"grok-4.3", 1.25e-6, 0.2e-6, 2.5e-6},
+		{"grok-4.20-0309-reasoning", 1.25e-6, 0.2e-6, 2.5e-6},
+		{"grok-build-0.1", 1e-6, 0.2e-6, 2e-6},
+	} {
+		p, err := svc.GetModelPricing(tc.model)
+		require.NoError(t, err, tc.model)
+		require.InDelta(t, tc.input, p.InputPricePerToken, 1e-12)
+		require.InDelta(t, tc.cached, p.CacheReadPricePerToken, 1e-12)
+		require.InDelta(t, tc.output, p.OutputPricePerToken, 1e-12)
+		require.Equal(t, 200000, p.LongContextInputThreshold)
+	}
+}
+
 func TestCalculateCostUnified_GroupLongContextToggleUsesPresetLadder(t *testing.T) {
 	svc := newTestBillingService()
 	resolver := NewModelPricingResolver(nil, svc)
@@ -1234,9 +1264,9 @@ func TestCalculateCostUnified_GroupLongContextToggleUsesPresetLadder(t *testing.
 	require.InDelta(t, disabled.OutputCost*2, enabled.OutputCost, 1e-12)
 }
 
-func TestGetModelPricing_UnknownGrokTextFallsBackToGrok45(t *testing.T) {
+func TestGetModelPricing_UnknownGrokTextFallsBackToGrok46(t *testing.T) {
 	svc := newTestBillingService()
-	baseline, err := svc.GetModelPricing("grok-4.5")
+	baseline, err := svc.GetModelPricing("grok-4.6")
 	require.NoError(t, err)
 
 	for _, model := range []string{"grok-5", "grok-5-latest", "x-ai/grok-7", "grok-4.7-beta"} {
diff --git a/backend/internal/service/gateway_service.go b/backend/internal/service/gateway_service.go
index 7203d618da..fd39ce2bc3 100644
--- a/backend/internal/service/gateway_service.go
+++ b/backend/internal/service/gateway_service.go
@@ -671,14 +671,15 @@ type GatewayFailureReason string
 // source-compatible and preserves their legacy retry-next-account behavior.
 type UpstreamFailoverError struct {
 	StatusCode               int
-	ResponseBody             []byte      // 上游响应体，用于错误透传规则匹配
-	ResponseHeaders          http.Header // 上游响应头，用于透传 cf-ray/cf-mitigated/content-type 等诊断信息
-	ForceCacheBilling        bool        // Antigravity 粘性会话切换时设为 true
-	RetryableOnSameAccount   bool        // 临时性错误（如 Google 间歇性 400、空响应），应在同一账号上重试 N 次再切换
-	SameAccountRetryDelay    time.Duration
-	SameAccountRetryDeadline time.Time
-	RequestScopedTransient   bool // 故障因素与账号无关（如上游按客户端身份/模型容量降载）：可同账号重试，但不得据此对账号做临时封禁
-	SafeToFailoverAfterWrite bool // 仅写出 SSE 注释等非语义字节时，仍可在同一客户端流中切换账号
+	ResponseBody             []byte        // 上游响应体，用于错误透传规则匹配
+	ResponseHeaders          http.Header   // 上游响应头，用于透传 cf-ray/cf-mitigated/content-type 等诊断信息
+	ForceCacheBilling        bool          // Antigravity 粘性会话切换时设为 true
+	RetryableOnSameAccount   bool          // 临时性错误（如 Google 间歇性 400、空响应），应在同一账号上重试 N 次再切换
+	SameAccountRetryDelay    time.Duration // 同账号重试的最小间隔；零值使用 handler 默认值
+	SameAccountRetryDeadline time.Time     // 同账号重试截止时间；零值表示仅受 retryLimit 限制
+	SameAccountRetryMax      int           // 可选的错误级同账号重试上限，低于 handler 默认预算时优先采用
+	RequestScopedTransient   bool          // 故障因素与账号无关（如上游按客户端身份/模型容量降载）：可同账号重试，但不得据此对账号做临时封禁
+	SafeToFailoverAfterWrite bool          // 仅写出 SSE 注释等非语义字节时，仍可在同一客户端流中切换账号
 	Stage                    GatewayFailureStage
 	Scope                    GatewayFailureScope
 	Reason                   GatewayFailureReason
diff --git a/backend/internal/service/grok_audio.go b/backend/internal/service/grok_audio.go
index 06c8410b8d..0cd138cad5 100644
--- a/backend/internal/service/grok_audio.go
+++ b/backend/internal/service/grok_audio.go
@@ -16,6 +16,10 @@ import (
 	"github.com/tidwall/gjson"
 )
 
+// DefaultGrokRealtimeDialTimeout bounds the pre-accept upstream handshake.
+// The timeout only covers dialing; an established session is not interrupted.
+const DefaultGrokRealtimeDialTimeout = 12 * time.Second
+
 // supportedGrokVoiceHTTPEndpoints are xAI Voice HTTP paths we forward as-is.
 var supportedGrokVoiceHTTPEndpoints = map[string]struct{}{
 	"tts":           {},
@@ -125,35 +129,79 @@ func (s *OpenAIGatewayService) ProxyGrokRealtime(ctx context.Context, c *gin.Con
 	if account.Platform != PlatformGrok {
 		return false, fmt.Errorf("account platform %s is not supported for grok realtime", account.Platform)
 	}
-	base, err := buildGrokVoiceURL(account, s.cfg, "realtime")
-	if err != nil {
-		return false, err
-	}
-	u, err := url.Parse(base)
-	if err != nil {
-		return false, err
-	}
-	u.Scheme = "wss"
-	u.RawQuery = "model=" + url.QueryEscape(firstNonEmpty(model, "grok-voice-latest"))
-	headers := http.Header{"Authorization": []string{"Bearer " + token}}
-	// Match media/voice HTTP: CLI headers only on CLI proxy hosts.
-	if account.IsGrokOAuth() && isGrokCLIProxyTarget(u.String()) {
-		applyGrokCLIHeaders(headers)
-	}
-	if account != nil {
-		account.ApplyHeaderOverrides(headers)
-	}
-
-	dialer := s.getOpenAIWSPassthroughDialer()
-	proxyURL := ""
-	if account.ProxyID != nil && account.Proxy != nil {
-		proxyURL = account.Proxy.URL()
-	}
-	upstream, _, _, err := dialer.Dial(ctx, u.String(), headers, proxyURL)
+	upstream, err := s.OpenGrokRealtime(ctx, account, token, model)
 	if err != nil {
 		return false, err
 	}
 	defer func() { _ = upstream.Close() }()
+	return s.ProxyGrokRealtimeConn(ctx, c, client, upstream)
+}
+
+type GrokRealtimeUpstream struct{ conn openAIWSClientConn }
+
+// GrokRealtimeDialError preserves an HTTP status returned before WebSocket
+// upgrade so handlers can apply the normal Grok account policy.
+type GrokRealtimeDialError struct {
+	StatusCode int
+	Err        error
+}
+
+func (e *GrokRealtimeDialError) Error() string { return e.Err.Error() }
+func (e *GrokRealtimeDialError) Unwrap() error { return e.Err }
+
+func (u *GrokRealtimeUpstream) Close() error {
+	if u == nil || u.conn == nil {
+		return nil
+	}
+	return u.conn.Close()
+}
+
+func (s *OpenAIGatewayService) OpenGrokRealtime(ctx context.Context, account *Account, token, model string) (*GrokRealtimeUpstream, error) {
+	if s == nil || account == nil || account.Platform != PlatformGrok {
+		return nil, fmt.Errorf("grok realtime account is required")
+	}
+	base, err := buildGrokVoiceURL(account, s.cfg, "realtime")
+	if err != nil {
+		return nil, err
+	}
+	u, err := url.Parse(base)
+	if err != nil {
+		return nil, err
+	}
+	u.Scheme = "wss"
+	q := u.Query()
+	q.Set("model", firstNonEmpty(model, "grok-voice-latest"))
+	u.RawQuery = q.Encode()
+	headers := http.Header{"Authorization": []string{"Bearer " + token}}
+	if account.IsGrokOAuth() && isGrokCLIProxyTarget(u.String()) {
+		applyGrokCLIHeaders(headers)
+	}
+	account.ApplyHeaderOverrides(headers)
+	proxyURL := ""
+	if account.ProxyID != nil && account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+	conn, status, _, err := s.getOpenAIWSPassthroughDialer().Dial(ctx, u.String(), headers, proxyURL)
+	if err != nil {
+		return nil, &GrokRealtimeDialError{StatusCode: status, Err: err}
+	}
+	return &GrokRealtimeUpstream{conn: conn}, nil
+}
+
+// HandleGrokRealtimeUpstreamError applies the shared Grok account policy to a
+// failed pre-accept WebSocket handshake.
+func (s *OpenAIGatewayService) HandleGrokRealtimeUpstreamError(ctx context.Context, account *Account, statusCode int, body []byte) {
+	if statusCode <= 0 {
+		statusCode = http.StatusBadGateway
+	}
+	s.handleGrokAccountUpstreamError(ctx, account, statusCode, nil, body)
+}
+
+func (s *OpenAIGatewayService) ProxyGrokRealtimeConn(ctx context.Context, c *gin.Context, client *coderws.Conn, upstream *GrokRealtimeUpstream) (bool, error) {
+	if s == nil || client == nil || upstream == nil || upstream.conn == nil {
+		return false, fmt.Errorf("realtime connection is required")
+	}
+	conn := upstream.conn
 
 	ctx, cancel := context.WithCancel(ctx)
 	defer cancel()
@@ -163,7 +211,7 @@ func (s *OpenAIGatewayService) ProxyGrokRealtime(ctx context.Context, c *gin.Con
 	// Upstream → client
 	go func() {
 		for {
-			msg, readErr := upstream.ReadMessage(ctx)
+			msg, readErr := conn.ReadMessage(ctx)
 			if readErr != nil {
 				errCh <- readErr
 				return
@@ -197,7 +245,7 @@ func (s *OpenAIGatewayService) ProxyGrokRealtime(ctx context.Context, c *gin.Con
 				errCh <- fmt.Errorf("invalid realtime event: %w", unmarshalErr)
 				return
 			}
-			if writeErr := upstream.WriteJSON(ctx, raw); writeErr != nil {
+			if writeErr := conn.WriteJSON(ctx, raw); writeErr != nil {
 				errCh <- writeErr
 				return
 			}
@@ -207,6 +255,45 @@ func (s *OpenAIGatewayService) ProxyGrokRealtime(ctx context.Context, c *gin.Con
 	return awaitGrokRealtimeAudioObserved(errCh, &audioObserved)
 }
 
+// ProbeGrokRealtime performs the upstream WebSocket handshake without sending
+// any client-visible events. Handlers use it before accepting the downstream
+// upgrade so authentication and endpoint failures remain ordinary HTTP errors.
+func (s *OpenAIGatewayService) ProbeGrokRealtime(ctx context.Context, account *Account, token, model string) error {
+	if s == nil || account == nil {
+		return fmt.Errorf("realtime service and account are required")
+	}
+	if account.Platform != PlatformGrok {
+		return fmt.Errorf("account platform %s is not supported for grok realtime", account.Platform)
+	}
+	base, err := buildGrokVoiceURL(account, s.cfg, "realtime")
+	if err != nil {
+		return err
+	}
+	u, err := url.Parse(base)
+	if err != nil {
+		return err
+	}
+	u.Scheme = "wss"
+	q := u.Query()
+	q.Set("model", firstNonEmpty(model, "grok-voice-latest"))
+	u.RawQuery = q.Encode()
+	headers := http.Header{"Authorization": []string{"Bearer " + token}}
+	if account.IsGrokOAuth() && isGrokCLIProxyTarget(u.String()) {
+		applyGrokCLIHeaders(headers)
+	}
+	account.ApplyHeaderOverrides(headers)
+	proxyURL := ""
+	if account.ProxyID != nil && account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+	dialer := s.getOpenAIWSPassthroughDialer()
+	conn, _, _, err := dialer.Dial(ctx, u.String(), headers, proxyURL)
+	if err != nil {
+		return err
+	}
+	return conn.Close()
+}
+
 func awaitGrokRealtimeAudioObserved(errCh <-chan error, audioObserved *atomic.Bool) (bool, error) {
 	err := <-errCh
 	if audioObserved == nil {
diff --git a/backend/internal/service/grok_media.go b/backend/internal/service/grok_media.go
index 2efb2332f6..85dd1b14ec 100644
--- a/backend/internal/service/grok_media.go
+++ b/backend/internal/service/grok_media.go
@@ -59,6 +59,8 @@ type GrokMediaRequestInfo struct {
 	N               int
 	Size            string
 	SizeTier        string
+	AspectRatio     string
+	ImageResolution string
 	Resolution      string
 	DurationSeconds int
 	InputImageURLs  []string
@@ -127,6 +129,8 @@ func ParseGrokMediaRequest(contentType string, body []byte) GrokMediaRequestInfo
 	info.Prompt = strings.TrimSpace(info.Prompt)
 	info.Size = strings.TrimSpace(info.Size)
 	info.SizeTier = NormalizeImageBillingTierOrDefault(info.Size)
+	info.AspectRatio = strings.TrimSpace(info.AspectRatio)
+	info.ImageResolution = grokImagineImageResolution(info.ImageResolution)
 	info.Resolution = NormalizeVideoBillingResolutionOrDefault(info.Resolution)
 	info.DurationSeconds = NormalizeVideoBillingDurationSecondsOrDefault(info.DurationSeconds)
 	if info.N <= 0 {
@@ -142,7 +146,8 @@ func parseGrokMediaJSONRequest(body []byte, info *GrokMediaRequestInfo) {
 	info.Model = strings.TrimSpace(gjson.GetBytes(body, "model").String())
 	info.Prompt = strings.TrimSpace(gjson.GetBytes(body, "prompt").String())
 	info.Size = strings.TrimSpace(gjson.GetBytes(body, "size").String())
-	info.Resolution = strings.TrimSpace(gjson.GetBytes(body, "resolution").String())
+	info.AspectRatio = strings.TrimSpace(gjson.GetBytes(body, "aspect_ratio").String())
+	assignGrokMediaResolution(strings.TrimSpace(gjson.GetBytes(body, "resolution").String()), info)
 	if duration := gjson.GetBytes(body, "duration"); duration.Exists() && duration.Type == gjson.Number {
 		info.DurationSeconds = int(duration.Int())
 	}
@@ -255,8 +260,10 @@ func parseGrokMediaMultipartRequest(contentType string, body []byte, info *GrokM
 			info.Prompt = value
 		case "size":
 			info.Size = value
+		case "aspect_ratio":
+			info.AspectRatio = value
 		case "resolution":
-			info.Resolution = value
+			assignGrokMediaResolution(value, info)
 		case "duration":
 			if duration, err := strconv.Atoi(value); err == nil {
 				info.DurationSeconds = duration
@@ -934,6 +941,12 @@ func prepareGrokMediaForwardBody(endpoint GrokMediaEndpoint, body []byte, conten
 	if info.Size != "" {
 		payload["size"] = info.Size
 	}
+	if info.ImageResolution != "" {
+		payload["resolution"] = info.ImageResolution
+	}
+	if info.AspectRatio != "" {
+		payload["aspect_ratio"] = info.AspectRatio
+	}
 
 	images := make([]map[string]string, 0, len(info.InputImageURLs)+len(info.Uploads))
 	for _, imageURL := range info.InputImageURLs {
@@ -1106,10 +1119,7 @@ func sanitizeGrokMediaForwardBody(endpoint GrokMediaEndpoint, body []byte, conte
 	}
 	switch endpoint {
 	case GrokMediaEndpointImagesGenerations, GrokMediaEndpointImagesEdits:
-		if !gjson.GetBytes(body, "size").Exists() {
-			return body, contentType, nil
-		}
-		out, err := sjson.DeleteBytes(body, "size")
+		out, err := applyGrokImagineImageGeometry(body)
 		if err != nil {
 			return nil, "", fmt.Errorf("sanitize grok media size: %w", err)
 		}
@@ -1288,11 +1298,16 @@ func (s *OpenAIGatewayService) handleGrokMediaErrorResponse(
 		Detail:             upstreamDetail,
 	})
 	if kind == "failover" {
+		retryable, retryDelay, retryDeadline, retryMax := grokSameAccountRetryMetadata(account, resp.StatusCode, body)
 		return nil, &UpstreamFailoverError{
-			StatusCode:             resp.StatusCode,
-			ResponseBody:           body,
-			ResponseHeaders:        resp.Header.Clone(),
-			RetryableOnSameAccount: account.IsPoolMode() && account.IsPoolModeRetryableStatus(resp.StatusCode),
+			StatusCode:               resp.StatusCode,
+			ResponseBody:             body,
+			ResponseHeaders:          resp.Header.Clone(),
+			RetryableOnSameAccount:   retryable,
+			RequestScopedTransient:   retryable && resp.StatusCode == http.StatusTooManyRequests,
+			SameAccountRetryDelay:    retryDelay,
+			SameAccountRetryDeadline: retryDeadline,
+			SameAccountRetryMax:      retryMax,
 		}
 	}
 
diff --git a/backend/internal/service/grok_media_image_geometry.go b/backend/internal/service/grok_media_image_geometry.go
new file mode 100644
index 0000000000..51e2a6fd69
--- /dev/null
+++ b/backend/internal/service/grok_media_image_geometry.go
@@ -0,0 +1,149 @@
+package service
+
+import (
+	"math"
+	"strconv"
+	"strings"
+
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// Official Imagine image geometry: https://docs.x.ai/developers/model-capabilities/images/generation
+var grokImagineAspectRatioValues = []struct {
+	label string
+	ratio float64
+}{
+	{"1:1", 1},
+	{"16:9", 16.0 / 9.0},
+	{"9:16", 9.0 / 16.0},
+	{"4:3", 4.0 / 3.0},
+	{"3:4", 3.0 / 4.0},
+	{"3:2", 1.5},
+	{"2:3", 2.0 / 3.0},
+	{"2:1", 2},
+	{"1:2", 0.5},
+	{"19.5:9", 19.5 / 9.0},
+	{"9:19.5", 9.0 / 19.5},
+	{"20:9", 20.0 / 9.0},
+	{"9:20", 9.0 / 20.0},
+}
+
+func applyGrokImagineImageGeometry(body []byte) ([]byte, error) {
+	size := strings.TrimSpace(gjson.GetBytes(body, "size").String())
+	resolution := grokImagineImageResolution(gjson.GetBytes(body, "resolution").String())
+	aspect := strings.TrimSpace(gjson.GetBytes(body, "aspect_ratio").String())
+	out := append([]byte(nil), body...)
+
+	if resolution == "" {
+		if derived := grokImagineImageResolutionFromSize(size); derived != "" {
+			next, err := sjson.SetBytes(out, "resolution", derived)
+			if err != nil {
+				return nil, err
+			}
+			out = next
+		}
+	} else if gjson.GetBytes(body, "resolution").String() != resolution {
+		next, err := sjson.SetBytes(out, "resolution", resolution)
+		if err != nil {
+			return nil, err
+		}
+		out = next
+	}
+
+	if aspect == "" {
+		if derived := grokImagineAspectRatioFromSize(size); derived != "" {
+			next, err := sjson.SetBytes(out, "aspect_ratio", derived)
+			if err != nil {
+				return nil, err
+			}
+			out = next
+		}
+	}
+
+	if !gjson.GetBytes(out, "size").Exists() {
+		return out, nil
+	}
+	return sjson.DeleteBytes(out, "size")
+}
+
+func assignGrokMediaResolution(value string, info *GrokMediaRequestInfo) {
+	if info == nil {
+		return
+	}
+	value = strings.TrimSpace(value)
+	if value == "" {
+		return
+	}
+	if img := grokImagineImageResolution(value); img != "" {
+		info.ImageResolution = img
+		return
+	}
+	info.Resolution = value
+}
+
+func grokImagineImageResolution(value string) string {
+	switch strings.ToLower(strings.TrimSpace(value)) {
+	case "1k":
+		return "1k"
+	case "2k":
+		return "2k"
+	default:
+		return ""
+	}
+}
+
+func grokImagineImageResolutionFromSize(size string) string {
+	if explicit := grokImagineImageResolution(size); explicit != "" {
+		return explicit
+	}
+	tier, ok := ClassifyImageBillingTier(size)
+	if !ok {
+		return ""
+	}
+	if tier == ImageBillingSize1K {
+		return "1k"
+	}
+	return "2k"
+}
+
+func grokImagineAspectRatioFromSize(size string) string {
+	width, height, ok := parseImageBillingDimensions(strings.TrimSpace(size))
+	if !ok || width <= 0 || height <= 0 {
+		return ""
+	}
+	div := grokImagineGCD(width, height)
+	exact := strconv.Itoa(width/div) + ":" + strconv.Itoa(height/div)
+	for _, candidate := range grokImagineAspectRatioValues {
+		if candidate.label == exact {
+			return exact
+		}
+	}
+	ratio := float64(width) / float64(height)
+	bestLabel := ""
+	bestDelta := math.MaxFloat64
+	for _, candidate := range grokImagineAspectRatioValues {
+		delta := math.Abs(ratio - candidate.ratio)
+		if delta < bestDelta {
+			bestDelta = delta
+			bestLabel = candidate.label
+		}
+	}
+	return bestLabel
+}
+
+func grokImagineGCD(a, b int) int {
+	if a < 0 {
+		a = -a
+	}
+	if b < 0 {
+		b = -b
+	}
+	for b != 0 {
+		a, b = b, a%b
+	}
+	if a == 0 {
+		return 1
+	}
+	return a
+}
diff --git a/backend/internal/service/grok_media_image_geometry_test.go b/backend/internal/service/grok_media_image_geometry_test.go
new file mode 100644
index 0000000000..cac26c3bd1
--- /dev/null
+++ b/backend/internal/service/grok_media_image_geometry_test.go
@@ -0,0 +1,60 @@
+package service
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+func TestApplyGrokImagineImageGeometryMapsOpenAISize(t *testing.T) {
+	t.Parallel()
+
+	out, err := applyGrokImagineImageGeometry([]byte(`{"model":"grok-imagine-image-2.0","prompt":"hi","size":"1152x1536"}`))
+	require.NoError(t, err)
+	require.False(t, gjson.GetBytes(out, "size").Exists())
+	require.Equal(t, "2k", gjson.GetBytes(out, "resolution").String())
+	require.Equal(t, "3:4", gjson.GetBytes(out, "aspect_ratio").String())
+}
+
+func TestApplyGrokImagineImageGeometryKeepsClientGeometry(t *testing.T) {
+	t.Parallel()
+
+	out, err := applyGrokImagineImageGeometry([]byte(`{"size":"1024x1024","resolution":"2K","aspect_ratio":"16:9"}`))
+	require.NoError(t, err)
+	require.False(t, gjson.GetBytes(out, "size").Exists())
+	require.Equal(t, "2k", gjson.GetBytes(out, "resolution").String())
+	require.Equal(t, "16:9", gjson.GetBytes(out, "aspect_ratio").String())
+}
+
+func TestSanitizeGrokMediaForwardBodyConvertsImageSize(t *testing.T) {
+	t.Parallel()
+
+	out, contentType, err := sanitizeGrokMediaForwardBody(
+		GrokMediaEndpointImagesGenerations,
+		[]byte(`{"model":"grok-imagine-image","prompt":"hi","size":"1024x1024"}`),
+		"application/json",
+	)
+	require.NoError(t, err)
+	require.Equal(t, "application/json", contentType)
+	require.False(t, gjson.GetBytes(out, "size").Exists())
+	require.Equal(t, "1k", gjson.GetBytes(out, "resolution").String())
+	require.Equal(t, "1:1", gjson.GetBytes(out, "aspect_ratio").String())
+}
+
+func TestParseGrokMediaRequestKeepsImageResolutionOutOfVideoNormalize(t *testing.T) {
+	t.Parallel()
+
+	info := ParseGrokMediaRequest("application/json", []byte(`{"model":"grok-imagine-image-2.0","resolution":"2K","aspect_ratio":"16:9"}`))
+	require.Equal(t, "2k", info.ImageResolution)
+	require.Equal(t, "16:9", info.AspectRatio)
+	require.Equal(t, VideoBillingResolution480P, info.Resolution)
+}
+
+func TestGrokImagineAspectRatioFromSize(t *testing.T) {
+	t.Parallel()
+	require.Equal(t, "1:1", grokImagineAspectRatioFromSize("1024x1024"))
+	require.Equal(t, "3:4", grokImagineAspectRatioFromSize("1152x1536"))
+	require.Equal(t, "4:3", grokImagineAspectRatioFromSize("1536x1152"))
+	require.Equal(t, "16:9", grokImagineAspectRatioFromSize("1792x1024"))
+}
diff --git a/backend/internal/service/grok_stream_idle.go b/backend/internal/service/grok_stream_idle.go
index 3bdfa27fc9..ddfa9f80bd 100644
--- a/backend/internal/service/grok_stream_idle.go
+++ b/backend/internal/service/grok_stream_idle.go
@@ -32,7 +32,16 @@ func grokStreamIdleFailoverError(account *Account, idle time.Duration) *Upstream
 		StatusCode:               502,
 		ResponseBody:             []byte(`{"error":{"code":"empty_upstream","message":"` + strings.ReplaceAll(msg, `"`, `'`) + `"}}`),
 		SafeToFailoverAfterWrite: true,
-		// Allow pool-mode retries; normal OAuth switches account via handler.
-		RetryableOnSameAccount: account != nil && account.IsPoolMode(),
+		// An idle upstream stream is transient and should get the configured
+		// same-account retry budget before switching credentials. This applies
+		// to both pooled and dedicated Grok accounts; the handler still enforces
+		// the request's retry limit.
+		RetryableOnSameAccount: account != nil && account.Platform == PlatformGrok,
+		RequestScopedTransient: true,
+		SameAccountRetryMax:    1,
+		// Permit at most one same-account replay after the idle failure. The
+		// deadline is anchored at failure time, so a hung stream cannot consume
+		// the normal three-attempt budget before failover.
+		SameAccountRetryDeadline: time.Now().Add(idle),
 	}
 }
diff --git a/backend/internal/service/grok_stream_idle_test.go b/backend/internal/service/grok_stream_idle_test.go
index e3c6d16943..7110dac9f2 100644
--- a/backend/internal/service/grok_stream_idle_test.go
+++ b/backend/internal/service/grok_stream_idle_test.go
@@ -21,5 +21,16 @@ func TestGrokStreamIdleFailoverError(t *testing.T) {
 	require.NotNil(t, err)
 	require.Equal(t, 502, err.StatusCode)
 	require.True(t, err.SafeToFailoverAfterWrite)
+	require.True(t, err.RetryableOnSameAccount)
+	require.True(t, err.RequestScopedTransient)
+	require.Equal(t, 1, err.SameAccountRetryMax)
 	require.Contains(t, string(err.ResponseBody), "empty_upstream")
+	require.WithinDuration(t, time.Now().Add(180*time.Second), err.SameAccountRetryDeadline, 2*time.Second)
+}
+
+func TestGrokStreamIdleFailoverErrorRequiresGrokAccount(t *testing.T) {
+	openAI := &Account{ID: 2, Platform: PlatformOpenAI, Type: AccountTypeOAuth}
+	err := grokStreamIdleFailoverError(openAI, time.Second)
+	require.False(t, err.RetryableOnSameAccount)
+	require.True(t, err.RequestScopedTransient)
 }
diff --git a/backend/internal/service/grok_upstream_errors.go b/backend/internal/service/grok_upstream_errors.go
index 918bea6613..816ac2fb97 100644
--- a/backend/internal/service/grok_upstream_errors.go
+++ b/backend/internal/service/grok_upstream_errors.go
@@ -116,8 +116,9 @@ func isGrokAccountAccessCode(value string) bool {
 		"subscription_required",
 		"entitlement_required",
 		"not_entitled",
-		"plan_required",
-		"permission_denied":
+		"plan_required":
+		// permission-denied is omitted: xAI reuses it for both entitlement
+		// refusals and request-scoped safety blocks, so the message decides.
 		return true
 	default:
 		return false
@@ -173,6 +174,7 @@ func grokContentPolicyMessage(value string) bool {
 		"prompt violates policy",
 		"input violates content policy",
 		"input violates policy",
+		"violates usage guidelines",
 	} {
 		if strings.Contains(lower, phrase) {
 			return true
@@ -207,7 +209,7 @@ func (s *OpenAIGatewayService) shouldFailoverGrokUpstreamError(statusCode int, r
 	}
 	decision := classifyGrokUpstreamFailure(statusCode, responseBody, "")
 	switch decision.Class {
-	case GrokFailureFreeUsage, GrokFailureEmptyUpstream, GrokFailureBilling, GrokFailureModelCapacity:
+	case GrokFailureFreeUsage, GrokFailureEmptyUpstream, GrokFailureBilling, GrokFailureModelCapacity, GrokFailureCompatibility:
 		return decision.ShouldFailover
 	}
 	return s.shouldFailoverUpstreamError(statusCode)
diff --git a/backend/internal/service/grok_upstream_errors_test.go b/backend/internal/service/grok_upstream_errors_test.go
index 0dee756b72..3bd5ad3e73 100644
--- a/backend/internal/service/grok_upstream_errors_test.go
+++ b/backend/internal/service/grok_upstream_errors_test.go
@@ -83,6 +83,24 @@ func TestIsGrokContentPolicyRejection(t *testing.T) {
 			body:   `{"error":{"code":"policy_violation","message":"request blocked by policy"}}`,
 			want:   true,
 		},
+		{
+			name:   "permission-denied usage guidelines is request scoped",
+			status: http.StatusForbidden,
+			body:   `{"code":"permission-denied","error":"Content violates usage guidelines. "}`,
+			want:   true,
+		},
+		{
+			name:   "permission-denied entitlement stays on the account path",
+			status: http.StatusForbidden,
+			body:   `{"code":"permission-denied","error":"Access to the chat endpoint is denied"}`,
+			want:   false,
+		},
+		{
+			name:   "structured account code overrides usage guidelines phrase",
+			status: http.StatusForbidden,
+			body:   `{"error":{"code":"account_suspended","message":"Content violates usage guidelines."}}`,
+			want:   false,
+		},
 		{
 			name:   "wrong status",
 			status: http.StatusBadRequest,
@@ -279,6 +297,21 @@ func TestGrokContentPolicySSEErrorDoesNotMutateOrFailover(t *testing.T) {
 	require.False(t, svc.isOpenAIAccountRuntimeBlocked(account))
 }
 
+func TestGrokPermissionDeniedContentRefusalDoesNotMutateOrFailover(t *testing.T) {
+	repo := &grokQuotaAccountRepo{}
+	svc := &OpenAIGatewayService{accountRepo: repo}
+	account := &Account{ID: 4785, Platform: PlatformGrok, Type: AccountTypeOAuth}
+	body := []byte(`{"code":"permission-denied","error":"Content violates usage guidelines. "}`)
+
+	svc.handleGrokAccountUpstreamError(context.Background(), account, http.StatusForbidden, nil, body)
+
+	require.Zero(t, repo.tempUnschedCalls)
+	require.Zero(t, repo.rateLimitedCalls)
+	require.Zero(t, repo.updateCalls)
+	require.False(t, svc.isOpenAIAccountRuntimeBlocked(account))
+	require.False(t, svc.shouldFailoverGrokUpstreamError(http.StatusForbidden, body))
+}
+
 func TestHandleGrokAccountUpstreamErrorEntitlement403KeepsDefaultCooldown(t *testing.T) {
 	repo := &grokQuotaAccountRepo{}
 	svc := &OpenAIGatewayService{accountRepo: repo}
diff --git a/backend/internal/service/grok_upstream_failure.go b/backend/internal/service/grok_upstream_failure.go
index e7bed08bf0..53d9d5500a 100644
--- a/backend/internal/service/grok_upstream_failure.go
+++ b/backend/internal/service/grok_upstream_failure.go
@@ -26,6 +26,10 @@ const (
 	GrokFailureRateLimit     GrokUpstreamFailureClass = "rate_limit"
 	GrokFailureAuth          GrokUpstreamFailureClass = "auth_error"
 	GrokFailureServer        GrokUpstreamFailureClass = "server_error"
+	// GrokFailureCompatibility represents a request-history/body-shape that is
+	// incompatible with the selected account or upstream replay contract. It
+	// is account-independent: fail over, but never quarantine the pool.
+	GrokFailureCompatibility GrokUpstreamFailureClass = "compatibility_error"
 )
 
 // GrokUpstreamFailureDecision is a pure classification result. Callers map it
@@ -111,6 +115,20 @@ func classifyGrokUpstreamFailure(statusCode int, responseBody []byte, requestedM
 		}
 	}
 
+	// Responses replay/compaction payloads can be rejected by one Grok
+	// deployment while the same request is valid on another account. Treat
+	// these precise decoder/content-shape failures as account compatibility
+	// errors, rather than durable account health failures.
+	if isGrokCompatibilityError(statusCode, low, code) {
+		return GrokUpstreamFailureDecision{
+			Class:          GrokFailureCompatibility,
+			Model:          model,
+			ShouldFailover: true,
+			ShouldCooldown: false,
+			Reason:         firstNonEmpty(text, "grok response compatibility error"),
+		}
+	}
+
 	// Empty HTTP 200 / empty model output (often rewritten to synthetic 502).
 	if isGrokEmptyModelOutputText(low) || isGrokEmptyModelOutputCode(code) {
 		return GrokUpstreamFailureDecision{
@@ -129,7 +147,7 @@ func classifyGrokUpstreamFailure(statusCode int, responseBody []byte, requestedM
 		return GrokUpstreamFailureDecision{
 			Class:          GrokFailureModelCapacity,
 			Model:          model,
-			Cooldown:       3 * time.Minute,
+			Cooldown:       time.Minute,
 			ShouldCooldown: true,
 			ShouldFailover: true,
 			BlockModel:     false,
@@ -355,7 +373,8 @@ func isGrokBillingQuotaText(low string) bool {
 	if strings.Contains(low, "payment") && (strings.Contains(low, "required") || strings.Contains(low, "fail")) {
 		return true
 	}
-	if strings.Contains(low, "spending limit") || strings.Contains(low, "run out of credits") || strings.Contains(low, "out of credits") {
+	if strings.Contains(low, "spending limit") || strings.Contains(low, "run out of credits") || strings.Contains(low, "out of credits") ||
+		(strings.Contains(low, "need a grok subscription") || strings.Contains(low, "need grok subscription")) {
 		return true
 	}
 	if strings.Contains(low, "余额不足") || strings.Contains(low, "欠费") || strings.Contains(low, "需要付费") {
@@ -364,6 +383,86 @@ func isGrokBillingQuotaText(low string) bool {
 	return false
 }
 
+// grokRetryableOnSameAccount marks transient 429 classes for the shared
+// failover loop. Capacity and ordinary throttles are request/model pressure,
+// not evidence that the credential is invalid, so a bounded retry on the same
+// account is preferable before switching accounts. Free-usage and billing
+// exhaustion deliberately skip same-account retry and fail over immediately.
+func grokRetryableOnSameAccount(account *Account, statusCode int, responseBody []byte) bool {
+	if account == nil || !account.IsGrok() {
+		return false
+	}
+	decision := classifyGrokUpstreamFailure(statusCode, responseBody, "")
+	switch decision.Class {
+	case GrokFailureFreeUsage, GrokFailureBilling, GrokFailureCompatibility:
+		// Quota/entitlement exhaustion is account state, not transient
+		// pressure. Retrying the same account only repeats the failure.
+		return false
+	case GrokFailureModelCapacity:
+		if statusCode == http.StatusTooManyRequests {
+			return true
+		}
+	}
+	return account.IsPoolMode() && account.IsPoolModeRetryableStatus(statusCode)
+}
+
+func grokSameAccountRetryMetadata(account *Account, statusCode int, responseBody []byte) (bool, time.Duration, time.Time, int) {
+	if !grokRetryableOnSameAccount(account, statusCode, responseBody) {
+		return false, 0, time.Time{}, 0
+	}
+	decision := classifyGrokUpstreamFailure(statusCode, responseBody, "")
+	if decision.Class != GrokFailureModelCapacity {
+		return true, 0, time.Time{}, 0
+	}
+	// The error is reconstructed after every upstream attempt, so a deadline
+	// stored on the error cannot provide a request-wide window. Cap capacity
+	// retries explicitly to one replay; this remains effective even when the
+	// first attempt itself takes longer than the nominal 30-second window.
+	return true, 500 * time.Millisecond, time.Now().Add(30 * time.Second), 1
+}
+
+// shouldMarkGrokTeamModelRateLimit controls the process-local sibling-account
+// overlay. Model-capacity responses are request pressure, not a team quota;
+// marking them would hide healthy sibling credentials while the bounded
+// same-account retry is still in progress. Ordinary 429s and free-usage
+// exhaustion retain the existing quota/team isolation behavior.
+func shouldMarkGrokTeamModelRateLimit(statusCode int, responseBody []byte) bool {
+	decision := classifyGrokUpstreamFailure(statusCode, responseBody, "")
+	if decision.Class == GrokFailureModelCapacity {
+		return false
+	}
+	return statusCode == http.StatusTooManyRequests || decision.Class == GrokFailureFreeUsage
+}
+
+func isGrokCompatibilityError(statusCode int, low, code string) bool {
+	if statusCode != http.StatusBadRequest && statusCode != http.StatusUnprocessableEntity {
+		return false
+	}
+	combined := strings.ToLower(strings.TrimSpace(low + " " + code))
+	// Compaction blobs are account/session-bound and frequently fail with 400
+	// or 422 after a reconnect. Also cover xAI's JSON decoder shape errors.
+	for _, phrase := range []string{
+		"could not decode the compaction blob",
+		"cannot decode the compaction blob",
+		"decode the compaction blob",
+		"ensure it is unmodified from the compact response",
+		"compaction blob",
+	} {
+		if strings.Contains(combined, phrase) {
+			return true
+		}
+	}
+	for _, marker := range []string{
+		"invalid_compaction",
+		"compaction_decode_error",
+	} {
+		if strings.Contains(combined, marker) {
+			return true
+		}
+	}
+	return false
+}
+
 func isGrokModelCapacityText(low string) bool {
 	return strings.Contains(low, "capacity") ||
 		strings.Contains(low, "overloaded") ||
@@ -489,10 +588,11 @@ func (s *OpenAIGatewayService) applyGrokUpstreamFailureDecision(
 	case GrokFailureEmptyUpstream:
 		reason = "grok empty model output"
 	case GrokFailureModelCapacity:
-		if persistGrokTransientModelCooldown(account, decision) {
-			return true
-		}
-		reason = "grok model capacity"
+		// Capacity is scoped to the requested model. Never persist an account-wide
+		// unschedulable state for this transient class; the failover loop performs
+		// a bounded same-account retry before selecting another account.
+		_ = persistGrokTransientModelCooldown(account, decision)
+		return true
 	case GrokFailureRateLimit:
 		// Pure 429 without free-usage language keeps the existing rate-limit
 		// snapshot path (Retry-After / quota headers). Body-only rate-limit
@@ -501,6 +601,11 @@ func (s *OpenAIGatewayService) applyGrokUpstreamFailureDecision(
 		return false
 	case GrokFailureServer:
 		reason = "grok upstream temporary error"
+	case GrokFailureCompatibility:
+		// Deliberately no account mutation. The caller uses ShouldFailover to
+		// retry another account; cooling a pool for a request-shape mismatch
+		// would remove healthy accounts.
+		return true
 	default:
 		return false
 	}
diff --git a/backend/internal/service/grok_upstream_failure_test.go b/backend/internal/service/grok_upstream_failure_test.go
index 0bd1063577..4f6514cde1 100644
--- a/backend/internal/service/grok_upstream_failure_test.go
+++ b/backend/internal/service/grok_upstream_failure_test.go
@@ -54,6 +54,14 @@ func TestClassifyGrokUpstreamFailure_EmptyUpstream(t *testing.T) {
 	require.Equal(t, 4*time.Minute, d.Cooldown)
 }
 
+func TestClassifyGrokUpstreamFailure_ModelCapacityUsesShortCooldown(t *testing.T) {
+	d := classifyGrokUpstreamFailure(http.StatusTooManyRequests,
+		[]byte(`{"error":{"message":"The model is currently at capacity due to high demand"}}`), "grok-4.6")
+	require.Equal(t, GrokFailureModelCapacity, d.Class)
+	require.Equal(t, time.Minute, d.Cooldown)
+	require.False(t, d.BlockModel)
+}
+
 func TestClassifyGrokUpstreamFailure_Billing(t *testing.T) {
 	d := classifyGrokUpstreamFailure(http.StatusForbidden, []byte(`{"code":"personal-team-blocked:spending-limit","error":"spending limit reached"}`), "")
 	require.Equal(t, GrokFailureBilling, d.Class)
@@ -61,6 +69,62 @@ func TestClassifyGrokUpstreamFailure_Billing(t *testing.T) {
 	require.True(t, d.ShouldFailover)
 }
 
+func TestClassifyGrokUpstreamFailure_GrokSubscriptionRequiredIsBilling(t *testing.T) {
+	d := classifyGrokUpstreamFailure(http.StatusPaymentRequired,
+		[]byte(`{"error":{"message":"You have run out of credits or need a Grok subscription"}}`), "grok-4.6")
+	require.Equal(t, GrokFailureBilling, d.Class)
+	require.True(t, d.ShouldFailover)
+	require.True(t, d.ShouldCooldown)
+}
+
+func TestGrokRetryableOnSameAccount_CapacityAndRateLimit(t *testing.T) {
+	account := &Account{ID: 9105, Platform: PlatformGrok, Type: AccountTypeOAuth}
+	require.True(t, grokRetryableOnSameAccount(account, http.StatusTooManyRequests,
+		[]byte(`{"error":{"message":"The model is currently at capacity due to high demand"}}`)))
+	require.False(t, grokRetryableOnSameAccount(account, http.StatusTooManyRequests,
+		[]byte(`{"error":{"message":"rate limit exceeded"}}`)))
+	require.False(t, grokRetryableOnSameAccount(account, http.StatusPaymentRequired,
+		[]byte(`{"error":{"message":"You have run out of credits or need a Grok subscription"}}`)))
+	poolAccount := &Account{ID: 9108, Platform: PlatformGrok, Type: AccountTypeOAuth,
+		Credentials: map[string]any{"pool_mode": true}}
+	require.False(t, grokRetryableOnSameAccount(poolAccount, http.StatusTooManyRequests,
+		[]byte(`{"error":{"code":"subscription:free-usage-exhausted"}}`)),
+		"pool free-usage must fail over instead of retrying the exhausted account")
+	require.False(t, grokRetryableOnSameAccount(account, http.StatusBadRequest,
+		[]byte(`{"error":{"message":"capacity field is invalid"}}`)))
+	nonGrok := &Account{ID: 9106, Platform: PlatformOpenAI, Type: AccountTypeOAuth}
+	require.False(t, grokRetryableOnSameAccount(nonGrok, http.StatusTooManyRequests,
+		[]byte(`{"error":{"message":"model at capacity"}}`)))
+}
+
+func TestShouldMarkGrokTeamModelRateLimit_ExcludesCapacity(t *testing.T) {
+	require.False(t, shouldMarkGrokTeamModelRateLimit(http.StatusTooManyRequests,
+		[]byte(`{"error":{"message":"The model is currently at capacity due to high demand"}}`)))
+	require.True(t, shouldMarkGrokTeamModelRateLimit(http.StatusTooManyRequests,
+		[]byte(`{"error":{"message":"rate limit exceeded"}}`)))
+	require.True(t, shouldMarkGrokTeamModelRateLimit(http.StatusBadRequest,
+		[]byte(`{"error":{"code":"subscription:free-usage-exhausted"}}`)))
+	require.False(t, shouldMarkGrokTeamModelRateLimit(http.StatusBadRequest,
+		[]byte(`{"error":{"message":"invalid request"}}`)))
+}
+
+func TestGrokSameAccountRetryMetadata_CapacityDeadline(t *testing.T) {
+	account := &Account{ID: 9107, Platform: PlatformGrok, Type: AccountTypeOAuth}
+	retryable, delay, deadline, retryMax := grokSameAccountRetryMetadata(account, http.StatusTooManyRequests,
+		[]byte(`{"error":{"message":"model capacity exceeded"}}`))
+	require.True(t, retryable)
+	require.Equal(t, 500*time.Millisecond, delay)
+	require.WithinDuration(t, time.Now().Add(30*time.Second), deadline, 2*time.Second)
+	require.Equal(t, 1, retryMax)
+
+	retryable, delay, deadline, retryMax = grokSameAccountRetryMetadata(account, http.StatusTooManyRequests,
+		[]byte(`{"error":{"message":"rate limit exceeded"}}`))
+	require.False(t, retryable)
+	require.Zero(t, delay)
+	require.True(t, deadline.IsZero())
+	require.Zero(t, retryMax)
+}
+
 func TestClassifyGrokUpstreamFailure_ValidationNoCool(t *testing.T) {
 	d := classifyGrokUpstreamFailure(http.StatusBadRequest, []byte(`{"error":{"message":"invalid tool schema"}}`), "")
 	require.Equal(t, GrokFailureNone, d.Class)
@@ -75,12 +139,48 @@ func TestClassifyGrokUpstreamFailure_FreeUsageWinsOver5xx(t *testing.T) {
 	require.NotEqual(t, GrokFailureServer, d.Class)
 }
 
+func TestClassifyGrokUpstreamFailure_CompatibilityDoesNotCooldown(t *testing.T) {
+	cases := []string{
+		`{"error":{"message":"Could not decode the compaction blob. Ensure it is unmodified from the compact response"}}`,
+		`{"code":"compaction_decode_error","message":"invalid response history"}`,
+	}
+	for _, body := range cases {
+		d := classifyGrokUpstreamFailure(http.StatusUnprocessableEntity, []byte(body), "grok-4.6")
+		require.Equal(t, GrokFailureCompatibility, d.Class, body)
+		require.True(t, d.ShouldFailover, body)
+		require.False(t, d.ShouldCooldown, body)
+		require.Zero(t, d.Cooldown, body)
+	}
+}
+
+func TestClassifyGrokUpstreamFailure_CompatibilityRequiresClientError(t *testing.T) {
+	body := []byte(`{"error":{"message":"upstream failed while handling the compaction blob"}}`)
+	for _, status := range []int{http.StatusBadGateway, http.StatusInternalServerError} {
+		d := classifyGrokUpstreamFailure(status, body, "grok-4.6")
+		require.NotEqual(t, GrokFailureCompatibility, d.Class)
+		require.True(t, d.ShouldCooldown)
+	}
+}
+
+func TestClassifyGrokUpstreamFailure_GenericShapeErrorDoesNotFailover(t *testing.T) {
+	d := classifyGrokUpstreamFailure(http.StatusBadRequest,
+		[]byte(`{"error":{"message":"data did not match any variant of the untagged enum content"}}`), "grok-4.6")
+	require.NotEqual(t, GrokFailureCompatibility, d.Class)
+	require.False(t, d.ShouldFailover)
+}
+
 func TestShouldFailoverGrokUpstreamError_FreeUsageBody(t *testing.T) {
 	svc := &OpenAIGatewayService{}
 	body := []byte(`{"error":{"code":"subscription:free-usage-exhausted","message":"free usage exhausted"}}`)
 	require.True(t, svc.shouldFailoverGrokUpstreamError(http.StatusBadRequest, body))
 }
 
+func TestShouldFailoverGrokUpstreamError_CompatibilityBody(t *testing.T) {
+	svc := &OpenAIGatewayService{}
+	body := []byte(`{"error":{"message":"Could not decode the compaction blob"}}`)
+	require.True(t, svc.shouldFailoverGrokUpstreamError(http.StatusUnprocessableEntity, body))
+}
+
 func TestShouldFailoverGrokUpstreamError_ContentPolicyStillNoFailover(t *testing.T) {
 	svc := &OpenAIGatewayService{}
 	body := []byte(`{"error":{"code":"new_sensitive","message":"text is sensitive"}}`)
@@ -149,6 +249,19 @@ func TestHandleGrokAccountUpstreamError_MultiAgentCapacityBlocksOnlyThatModel(t
 	require.False(t, isGrokModelQuotaBlocked(account.ID, "grok-4.5", time.Now()))
 }
 
+func TestHandleGrokAccountUpstreamError_CapacityNeverCoolsAccount(t *testing.T) {
+	repo := &grokQuotaAccountRepo{}
+	svc := &OpenAIGatewayService{accountRepo: repo}
+	account := &Account{ID: 9121, Platform: PlatformGrok, Type: AccountTypeOAuth}
+	ctx := withGrokTeamRateLimitModel(context.Background(), "grok-4.6")
+
+	svc.handleGrokAccountUpstreamError(ctx, account, http.StatusTooManyRequests, nil,
+		[]byte(`{"error":{"message":"The model is currently at capacity due to high demand"}}`))
+
+	require.Zero(t, repo.tempUnschedCalls)
+	require.False(t, svc.isOpenAIAccountRuntimeBlocked(account))
+}
+
 func TestHandleGrokAccountUpstreamError_FreeUsageDoesNotCoolPoolMode(t *testing.T) {
 	repo := &grokQuotaAccountRepo{}
 	svc := &OpenAIGatewayService{accountRepo: repo}
diff --git a/backend/internal/service/http_upstream_profile.go b/backend/internal/service/http_upstream_profile.go
index 83bd6e70fc..b2de6002cf 100644
--- a/backend/internal/service/http_upstream_profile.go
+++ b/backend/internal/service/http_upstream_profile.go
@@ -9,6 +9,7 @@ type HTTPUpstreamProfile string
 const (
 	HTTPUpstreamProfileDefault HTTPUpstreamProfile = ""
 	HTTPUpstreamProfileOpenAI  HTTPUpstreamProfile = "openai"
+	HTTPUpstreamProfileGrok    HTTPUpstreamProfile = "grok"
 )
 
 type httpUpstreamProfileContextKey struct{}
@@ -35,7 +36,7 @@ func HTTPUpstreamProfileFromContext(ctx context.Context) HTTPUpstreamProfile {
 		return HTTPUpstreamProfileDefault
 	}
 	switch profile {
-	case HTTPUpstreamProfileOpenAI:
+	case HTTPUpstreamProfileOpenAI, HTTPUpstreamProfileGrok:
 		return profile
 	default:
 		return HTTPUpstreamProfileDefault
diff --git a/backend/internal/service/openai_gateway_chat_completions_raw.go b/backend/internal/service/openai_gateway_chat_completions_raw.go
index f51bf00c8a..6713913939 100644
--- a/backend/internal/service/openai_gateway_chat_completions_raw.go
+++ b/backend/internal/service/openai_gateway_chat_completions_raw.go
@@ -201,11 +201,16 @@ func (s *OpenAIGatewayService) forwardAsRawChatCompletions(
 			})
 			s.handleGrokAccountUpstreamError(withGrokTeamRateLimitModel(ctx, upstreamModel), account, resp.StatusCode, resp.Header, respBody)
 			if s.shouldFailoverGrokUpstreamError(resp.StatusCode, respBody) {
+				retryable, retryDelay, retryDeadline, retryMax := grokSameAccountRetryMetadata(account, resp.StatusCode, respBody)
 				return nil, &UpstreamFailoverError{
-					StatusCode:             resp.StatusCode,
-					ResponseBody:           respBody,
-					ResponseHeaders:        resp.Header.Clone(),
-					RetryableOnSameAccount: account.IsPoolMode() && account.IsPoolModeRetryableStatus(resp.StatusCode),
+					StatusCode:               resp.StatusCode,
+					ResponseBody:             respBody,
+					ResponseHeaders:          resp.Header.Clone(),
+					RetryableOnSameAccount:   retryable,
+					RequestScopedTransient:   retryable && resp.StatusCode == http.StatusTooManyRequests,
+					SameAccountRetryDelay:    retryDelay,
+					SameAccountRetryDeadline: retryDeadline,
+					SameAccountRetryMax:      retryMax,
 				}
 			}
 			return s.handleChatCompletionsErrorResponse(resp, c, account, billingModel)
diff --git a/backend/internal/service/openai_gateway_grok.go b/backend/internal/service/openai_gateway_grok.go
index 9e2c70579b..b8ab024014 100644
--- a/backend/internal/service/openai_gateway_grok.go
+++ b/backend/internal/service/openai_gateway_grok.go
@@ -119,7 +119,7 @@ func (s *OpenAIGatewayService) forwardGrokResponses(
 		// xAI can reject encrypted reasoning or a compaction blob copied from a
 		// different decoder/cache context. Retry once on the same account after
 		// preserving visible summaries and removing only opaque replay state.
-		if attempt > 0 || resp.StatusCode != http.StatusBadRequest {
+		if attempt > 0 || (resp.StatusCode != http.StatusBadRequest && resp.StatusCode != http.StatusUnprocessableEntity) {
 			break
 		}
 		respBody := s.readUpstreamErrorBody(resp)
@@ -175,17 +175,22 @@ func (s *OpenAIGatewayService) forwardGrokResponses(
 		})
 		errCtx := withGrokTeamRateLimitModel(ctx, upstreamModel)
 		s.handleGrokAccountUpstreamError(errCtx, account, resp.StatusCode, resp.Header, respBody)
-		// 429 / free-usage: stamp team+model cool so sibling accounts skip this model.
-		if resp.StatusCode == http.StatusTooManyRequests ||
-			classifyGrokUpstreamFailure(resp.StatusCode, respBody, upstreamModel).Class == GrokFailureFreeUsage {
+		// Quota/rate-limit responses stamp the team+model overlay. Capacity is
+		// request pressure and must not hide sibling accounts.
+		if shouldMarkGrokTeamModelRateLimit(resp.StatusCode, respBody) {
 			markGrokTeamModelRateLimit(account, upstreamModel, resolveGrokTeamRateLimitUntil(time.Now().Add(grokTeamRateLimitDefaultTTL), time.Now()))
 		}
 		if s.shouldFailoverGrokUpstreamError(resp.StatusCode, respBody) {
+			retryable, retryDelay, retryDeadline, retryMax := grokSameAccountRetryMetadata(account, resp.StatusCode, respBody)
 			return nil, &UpstreamFailoverError{
-				StatusCode:             resp.StatusCode,
-				ResponseBody:           respBody,
-				ResponseHeaders:        resp.Header.Clone(),
-				RetryableOnSameAccount: account.IsPoolMode() && account.IsPoolModeRetryableStatus(resp.StatusCode),
+				StatusCode:               resp.StatusCode,
+				ResponseBody:             respBody,
+				ResponseHeaders:          resp.Header.Clone(),
+				RetryableOnSameAccount:   retryable,
+				RequestScopedTransient:   retryable && resp.StatusCode == http.StatusTooManyRequests,
+				SameAccountRetryDelay:    retryDelay,
+				SameAccountRetryDeadline: retryDeadline,
+				SameAccountRetryMax:      retryMax,
 			}
 		}
 		return s.handleErrorResponse(ctx, resp, c, account, patchedBody, upstreamModel)
@@ -262,7 +267,7 @@ func (s *OpenAIGatewayService) forwardGrokResponses(
 }
 
 func isGrokInvalidEncryptedContentResponse(statusCode int, body []byte) bool {
-	if statusCode != http.StatusBadRequest {
+	if statusCode != http.StatusBadRequest && statusCode != http.StatusUnprocessableEntity {
 		return false
 	}
 
@@ -275,7 +280,7 @@ func isGrokInvalidEncryptedContentResponse(statusCode int, body []byte) bool {
 		code = strings.TrimSpace(errNode.Get("code").String())
 	}
 
-	if strings.EqualFold(code, "invalid_encrypted_content") {
+	if strings.EqualFold(code, "invalid_encrypted_content") || strings.EqualFold(code, "invalid_compaction") || strings.EqualFold(code, "compaction_decode_error") {
 		return true
 	}
 	// Keep the official xAI flat-code gate so unrelated 400s are not retried.
@@ -285,19 +290,22 @@ func isGrokInvalidEncryptedContentResponse(statusCode int, body []byte) bool {
 	for _, candidate := range grokStructuredErrorMessageCandidates(body) {
 		normalizedMessage := strings.ToLower(candidate)
 		// Nested OpenAI-style envelopes may omit top-level code; require decrypt text.
-		if code == "" && !strings.Contains(normalizedMessage, "decrypt") {
+		if code == "" && !strings.Contains(normalizedMessage, "decrypt") && !strings.Contains(normalizedMessage, "decode the compaction blob") {
 			continue
 		}
 		if strings.Contains(normalizedMessage, "encrypted_content") &&
 			(strings.Contains(normalizedMessage, "decrypt") || strings.Contains(normalizedMessage, "unmodified")) {
 			return true
 		}
+		if strings.Contains(normalizedMessage, "decode the compaction blob") {
+			return true
+		}
 	}
 	return false
 }
 
 func isGrokCompactionReplayDecodeError(statusCode int, body []byte) bool {
-	if statusCode != http.StatusBadRequest || len(body) == 0 {
+	if (statusCode != http.StatusBadRequest && statusCode != http.StatusUnprocessableEntity) || len(body) == 0 {
 		return false
 	}
 	for _, candidate := range grokStructuredErrorMessageCandidates(body) {
@@ -472,7 +480,8 @@ func trimGrokInvalidEncryptedContentRetryBody(body []byte) ([]byte, bool, error)
 
 	hasEncryptedReasoning := false
 	for _, item := range items {
-		if strings.TrimSpace(item.Get("type").String()) == "reasoning" && item.Get("encrypted_content").Exists() {
+		if (strings.TrimSpace(item.Get("type").String()) == "reasoning" && item.Get("encrypted_content").Exists()) ||
+			(isOpenAICompactionType(strings.TrimSpace(item.Get("type").String())) && item.Get("encrypted_content").Exists()) {
 			hasEncryptedReasoning = true
 			break
 		}
@@ -734,7 +743,7 @@ func grokSupportsXHighReasoningEffort(model string) bool {
 func grokSupportsReasoningEffort(model string) bool {
 	model = strings.ToLower(xai.StripGrokProviderPrefix(strings.TrimSpace(model)))
 	switch model {
-	case xai.DefaultTextModel, "grok-4.5-latest", "grok-4.6", "grok-4.6-latest",
+	case "grok-4.5", "grok-4.5-latest", "grok-4.6", "grok-4.6-latest",
 		"grok-4.3", "grok-4.3-latest",
 		"grok-3-mini", "grok-3-mini-fast", "grok-4.20-0309-reasoning",
 		"grok-4.20-reasoning", "grok-4.20-multi-agent-0309":
@@ -937,33 +946,98 @@ func grokResponsesToolDedupKey(tool gjson.Result) string {
 	return "json:" + normalizeCompatSeedJSON(json.RawMessage(tool.Raw))
 }
 
-// sanitizeGrokReasoningNullContent 删除 reasoning 项中的 "content": null。
-// xAI 的 untagged enum 反序列化器拒收该字段，返回 422。
+// sanitizeGrokReasoningNullContent drops explicit JSON nulls from Responses
+// input items. xAI's untagged ModelInput decoder 422s on those fields.
+// Compaction items stay unmodified per the compact contract.
 func sanitizeGrokReasoningNullContent(body []byte) ([]byte, error) {
 	input := gjson.GetBytes(body, "input")
-	if !input.Exists() || !input.IsArray() {
+	if !input.Exists() || (!input.IsArray() && !input.IsObject()) {
 		return body, nil
 	}
 
-	items := input.Array()
+	var decoded map[string]any
+	decoder := json.NewDecoder(bytes.NewReader(body))
+	decoder.UseNumber()
+	if err := decoder.Decode(&decoded); err != nil {
+		return body, nil
+	}
+	rawInput, ok := decoded["input"]
+	if !ok {
+		return body, nil
+	}
+	cleaned, changed := stripExplicitNullsFromGrokInput(rawInput)
+	if !changed {
+		return body, nil
+	}
+	decoded["input"] = cleaned
+	out, err := marshalOpenAIUpstreamJSON(decoded)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func stripExplicitNullsFromGrokInput(value any) (any, bool) {
+	switch node := value.(type) {
+	case []any:
+		changed := false
+		for i, item := range node {
+			itemMap, ok := item.(map[string]any)
+			if !ok {
+				next, childChanged := stripExplicitNullsFromGrokInput(item)
+				if childChanged {
+					node[i] = next
+					changed = true
+				}
+				continue
+			}
+			if isOpenAICompactionType(stringValue(itemMap["type"])) {
+				continue
+			}
+			next, childChanged := stripExplicitNullsFromJSONObject(itemMap)
+			if childChanged {
+				node[i] = next
+				changed = true
+			}
+		}
+		return node, changed
+	case map[string]any:
+		if isOpenAICompactionType(stringValue(node["type"])) {
+			return node, false
+		}
+		return stripExplicitNullsFromJSONObject(node)
+	default:
+		return value, false
+	}
+}
+
+func stripExplicitNullsFromJSONObject(node map[string]any) (map[string]any, bool) {
+	if node == nil {
+		return node, false
+	}
 	changed := false
-	for i := len(items) - 1; i >= 0; i-- {
-		item := items[i]
-		if strings.TrimSpace(item.Get("type").String()) != "reasoning" {
+	for key, child := range node {
+		if child == nil {
+			delete(node, key)
+			changed = true
 			continue
 		}
-		contentResult := item.Get("content")
-		if contentResult.Exists() && contentResult.Type == gjson.Null {
-			var err error
-			body, err = sjson.DeleteBytes(body, fmt.Sprintf("input.%d.content", i))
-			if err != nil {
-				return nil, err
+		switch typed := child.(type) {
+		case map[string]any:
+			next, childChanged := stripExplicitNullsFromJSONObject(typed)
+			if childChanged {
+				node[key] = next
+				changed = true
+			}
+		case []any:
+			next, childChanged := stripExplicitNullsFromGrokInput(typed)
+			if childChanged {
+				node[key] = next
+				changed = true
 			}
-			changed = true
 		}
 	}
-	_ = changed
-	return body, nil
+	return node, changed
 }
 
 var grokResponsesSupportedToolTypes = map[string]struct{}{
@@ -984,6 +1058,13 @@ func sanitizeGrokResponsesTools(body []byte) ([]byte, error) {
 		return deleteGrokOrphanToolControls(body)
 	}
 	if !tools.IsArray() {
+		// xAI rejects tool_choice when tools is null/object. Drop the malformed
+		// collection and any orphan tool controls instead of forwarding a pair
+		// the Grok Responses endpoint cannot interpret.
+		body, err := sjson.DeleteBytes(body, "tools")
+		if err != nil {
+			return nil, err
+		}
 		return deleteGrokOrphanToolControls(body)
 	}
 
@@ -1275,11 +1356,16 @@ func (s *OpenAIGatewayService) describeGrokComposerImage(
 		})
 		s.handleGrokAccountUpstreamError(withGrokTeamRateLimitModel(ctx, grokComposerImageBridgeVisionModel), account, resp.StatusCode, resp.Header, respBody)
 		if s.shouldFailoverGrokUpstreamError(resp.StatusCode, respBody) {
+			retryable, retryDelay, retryDeadline, retryMax := grokSameAccountRetryMetadata(account, resp.StatusCode, respBody)
 			return "", OpenAIUsage{}, &UpstreamFailoverError{
-				StatusCode:             resp.StatusCode,
-				ResponseBody:           respBody,
-				ResponseHeaders:        resp.Header.Clone(),
-				RetryableOnSameAccount: account.IsPoolMode() && account.IsPoolModeRetryableStatus(resp.StatusCode),
+				StatusCode:               resp.StatusCode,
+				ResponseBody:             respBody,
+				ResponseHeaders:          resp.Header.Clone(),
+				RetryableOnSameAccount:   retryable,
+				RequestScopedTransient:   retryable && resp.StatusCode == http.StatusTooManyRequests,
+				SameAccountRetryDelay:    retryDelay,
+				SameAccountRetryDeadline: retryDeadline,
+				SameAccountRetryMax:      retryMax,
 			}
 		}
 		return "", OpenAIUsage{}, fmt.Errorf("grok composer image bridge upstream error: %s", upstreamMsg)
@@ -1416,6 +1502,7 @@ func buildGrokResponsesRequest(ctx context.Context, c *gin.Context, account *Acc
 	if err != nil {
 		return nil, err
 	}
+	req = req.WithContext(WithHTTPUpstreamProfile(req.Context(), HTTPUpstreamProfileGrok))
 	req.Header.Set("Authorization", "Bearer "+token)
 	req.Header.Set("Content-Type", "application/json")
 	req.Header.Set("Accept", "application/json, text/event-stream")
@@ -1452,6 +1539,10 @@ func applyGrokCLIHeaders(headers http.Header) {
 }
 
 func (s *OpenAIGatewayService) updateGrokUsageSnapshot(ctx context.Context, account *Account, snapshot *xai.QuotaSnapshot) {
+	s.updateGrokUsageSnapshotWithRateLimit(ctx, account, snapshot, true)
+}
+
+func (s *OpenAIGatewayService) updateGrokUsageSnapshotWithRateLimit(ctx context.Context, account *Account, snapshot *xai.QuotaSnapshot, installRateLimit bool) {
 	if s == nil || account == nil || account.ID <= 0 || snapshot == nil {
 		return
 	}
@@ -1499,7 +1590,7 @@ func (s *OpenAIGatewayService) updateGrokUsageSnapshot(ctx context.Context, acco
 	// API keys retain the snapshot for observability but leave account health to
 	// the upstream pool. Other accounts install the immediate runtime and durable
 	// rate-limit state when the observed window is exhausted.
-	if hasActiveLimit && !account.IsPoolMode() {
+	if installRateLimit && hasActiveLimit && !account.IsPoolMode() {
 		s.rateLimitGrok(stateCtx, account, resetAt)
 	} else if recovery {
 		clearGrokRateLimitAfterRecovery(stateCtx, s.accountRepo, account)
@@ -1822,17 +1913,12 @@ func grokRequestedModelFromCtx(ctx context.Context) string {
 	return strings.TrimSpace(model)
 }
 
-func isGrokHeavyTransientModel(requestedModel string) bool {
-	model := strings.ToLower(strings.TrimSpace(xai.ResolveGrokTextResponsesModelID(requestedModel)))
-	return strings.Contains(model, "multi-agent")
-}
-
 func persistGrokTransientModelCooldown(account *Account, decision GrokUpstreamFailureDecision) bool {
 	if account == nil {
 		return false
 	}
 	model := strings.TrimSpace(decision.Model)
-	if model == "" || !isGrokHeavyTransientModel(model) {
+	if model == "" {
 		return false
 	}
 	cooldown := decision.Cooldown
@@ -1851,14 +1937,17 @@ func (s *OpenAIGatewayService) handleGrokAccountUpstreamError(ctx context.Contex
 		return
 	}
 	now := time.Now()
+	decision := classifyGrokUpstreamFailure(statusCode, responseBody, grokRequestedModelFromCtx(ctx))
 	snapshot := parseGrokQuotaSnapshot(headers, statusCode, now)
 	stampGrokQuotaSnapshotForPlan(account, snapshot, grokRequestedModelFromCtx(ctx))
-	s.updateGrokUsageSnapshot(ctx, account, snapshot)
+	// Capacity 429 is model pressure, not account quota exhaustion. Keep the
+	// snapshot for observability but do not install account-level rate limiting;
+	// the failover decision below applies a bounded model-scoped block instead.
+	s.updateGrokUsageSnapshotWithRateLimit(ctx, account, snapshot, decision.Class != GrokFailureModelCapacity)
 
 	// Body-first free-usage / empty / billing / capacity must run before the
 	// status switch so non-429 free-usage bodies still cool the account.
 	// Pool-mode still skips durable mutation unless an explicit temp rule matches.
-	decision := classifyGrokUpstreamFailure(statusCode, responseBody, grokRequestedModelFromCtx(ctx))
 	if decision.ShouldCooldown && decision.Class != GrokFailureNone && decision.Class != GrokFailureRateLimit {
 		if account.IsPoolMode() {
 			// Allow configured temp rules (403) below; skip default body cools.
diff --git a/backend/internal/service/openai_gateway_grok_chat_bridge.go b/backend/internal/service/openai_gateway_grok_chat_bridge.go
index e2b8f9a477..f04ab5ae43 100644
--- a/backend/internal/service/openai_gateway_grok_chat_bridge.go
+++ b/backend/internal/service/openai_gateway_grok_chat_bridge.go
@@ -633,11 +633,16 @@ func (s *OpenAIGatewayService) forwardGrokChatCompletionsViaResponses(
 		})
 		s.handleGrokAccountUpstreamError(withGrokTeamRateLimitModel(ctx, upstreamModel), account, resp.StatusCode, resp.Header, respBody)
 		if s.shouldFailoverGrokUpstreamError(resp.StatusCode, respBody) {
+			retryable, retryDelay, retryDeadline, retryMax := grokSameAccountRetryMetadata(account, resp.StatusCode, respBody)
 			return nil, &UpstreamFailoverError{
-				StatusCode:             resp.StatusCode,
-				ResponseBody:           respBody,
-				ResponseHeaders:        resp.Header.Clone(),
-				RetryableOnSameAccount: account.IsPoolMode() && account.IsPoolModeRetryableStatus(resp.StatusCode),
+				StatusCode:               resp.StatusCode,
+				ResponseBody:             respBody,
+				ResponseHeaders:          resp.Header.Clone(),
+				RetryableOnSameAccount:   retryable,
+				RequestScopedTransient:   retryable && resp.StatusCode == http.StatusTooManyRequests,
+				SameAccountRetryDelay:    retryDelay,
+				SameAccountRetryDeadline: retryDeadline,
+				SameAccountRetryMax:      retryMax,
 			}
 		}
 		return s.handleChatCompletionsErrorResponse(resp, c, account, billingModel)
diff --git a/backend/internal/service/openai_gateway_grok_chat_bridge_test.go b/backend/internal/service/openai_gateway_grok_chat_bridge_test.go
index 2df7bb1f8b..31178122ca 100644
--- a/backend/internal/service/openai_gateway_grok_chat_bridge_test.go
+++ b/backend/internal/service/openai_gateway_grok_chat_bridge_test.go
@@ -251,7 +251,7 @@ func TestForwardGrokChatViaResponsesNonStreamingCachesAndReturnsChat(t *testing.
 	require.NotNil(t, result)
 	require.Equal(t, xai.DefaultCLIBaseURL+"/responses", upstream.lastReq.URL.String())
 	require.Equal(t, grokChatResponsesEndpoint, result.UpstreamEndpoint)
-	require.Equal(t, "grok-4.5", result.UpstreamModel)
+	require.Equal(t, "grok-4.6", result.UpstreamModel)
 	require.Equal(t, 9908, result.Usage.InputTokens)
 	require.Equal(t, 12, result.Usage.OutputTokens)
 	require.Equal(t, 9856, result.Usage.CacheReadInputTokens)
@@ -352,7 +352,7 @@ func TestForwardGrokChatViaResponsesCodeBuddyUsesStableConversationHeader(t *tes
 			c.Request.Header.Set("X-Request-ID", "generic-"+tt.requestID)
 			c.Set("api_key", &APIKey{ID: 7111})
 
-			identity := resolveGrokCacheIdentity(c, tt.body, "", "grok-4.5")
+			identity := resolveGrokCacheIdentity(c, tt.body, "", "grok-4.6")
 			require.NotEmpty(t, identity)
 			if index == 0 {
 				stableIdentity = identity
@@ -406,8 +406,8 @@ func TestForwardGrokChatViaResponsesTraeToolHistoryKeepsCacheRoute(t *testing.T)
 		accountRepo:       repo,
 	}
 
-	firstTurnIdentity := resolveGrokCacheIdentity(c, firstTurnBody, "", "grok-4.5")
-	extendedTurnIdentity := resolveGrokCacheIdentity(c, body, "", "grok-4.5")
+	firstTurnIdentity := resolveGrokCacheIdentity(c, firstTurnBody, "", "grok-4.6")
+	extendedTurnIdentity := resolveGrokCacheIdentity(c, body, "", "grok-4.6")
 	require.NotEmpty(t, firstTurnIdentity)
 	require.Equal(t, firstTurnIdentity, extendedTurnIdentity)
 
@@ -463,8 +463,8 @@ func TestForwardGrokChatViaResponsesTraeCompatibilityFieldsKeepCacheRoute(t *tes
 		accountRepo:       repo,
 	}
 
-	firstTurnIdentity := resolveGrokCacheIdentity(c, firstTurnBody, "", "grok-4.5")
-	extendedTurnIdentity := resolveGrokCacheIdentity(c, body, "", "grok-4.5")
+	firstTurnIdentity := resolveGrokCacheIdentity(c, firstTurnBody, "", "grok-4.6")
+	extendedTurnIdentity := resolveGrokCacheIdentity(c, body, "", "grok-4.6")
 	require.NotEmpty(t, firstTurnIdentity)
 	require.Equal(t, firstTurnIdentity, extendedTurnIdentity)
 
@@ -534,7 +534,7 @@ func TestForwardGrokChatRuntimeGateFallsBackToRaw(t *testing.T) {
 		mappedModel  string
 		wantUpstream string
 	}{
-		{name: "missing cache identity", wantUpstream: "grok-4.5"},
+		{name: "missing cache identity", wantUpstream: "grok-4.6"},
 		{name: "non cache capable mapped model", setAPIKey: true, mappedModel: "grok-4.3", wantUpstream: "grok-4.3"},
 	}
 
diff --git a/backend/internal/service/openai_gateway_grok_test.go b/backend/internal/service/openai_gateway_grok_test.go
index f248cf1951..27c19866c0 100644
--- a/backend/internal/service/openai_gateway_grok_test.go
+++ b/backend/internal/service/openai_gateway_grok_test.go
@@ -418,9 +418,10 @@ func TestSanitizeGrokResponsesToolsKeepsToolChoiceOnlyWithSupportedTools(t *test
 			wantToolChoice: true,
 		},
 		{
-			name:      "malformed non-array tools drop orphan controls",
-			body:      `{"input":"hello","tools":{"type":"function","name":"lookup"},"tool_choice":"auto"}`,
-			wantTools: true,
+			name:           "malformed non-array tools are removed",
+			body:           `{"input":"hello","tools":{"type":"function","name":"lookup"},"tool_choice":"auto"}`,
+			wantTools:      false,
+			wantToolChoice: false,
 		},
 	}
 
@@ -1188,7 +1189,7 @@ func TestForwardGrokMediaAppliesAccountModelMappingAfterEndpointNormalization(t
 			modelMapping:     map[string]any{"grok-imagine-image-quality": "vendor-image-model"},
 			wantRequestModel: "grok-imagine-image-quality",
 			wantUpstream:     "vendor-image-model",
-			wantBody:         `{"model":"vendor-image-model","prompt":"draw"}`,
+			wantBody:         `{"model":"vendor-image-model","prompt":"draw","resolution":"1k","aspect_ratio":"1:1"}`,
 			responseBody:     `{"data":[{"url":"https://images.test/mapped.png"}]}`,
 		},
 		{
@@ -1311,7 +1312,8 @@ func TestForwardGrokMediaImagesGenerationStripsUnsupportedSize(t *testing.T) {
 
 	result, err := svc.ForwardGrokMedia(context.Background(), c, account, GrokMediaEndpointImagesGenerations, "", body, "application/json")
 	require.NoError(t, err)
-	require.JSONEq(t, `{"model":"grok-imagine-image","prompt":"draw a cat"}`, string(upstream.lastBody))
+	require.JSONEq(t, `{"model":"grok-imagine-image","prompt":"draw a cat","resolution":"1k","aspect_ratio":"1:1"}`, string(upstream.lastBody))
+	require.False(t, gjson.GetBytes(upstream.lastBody, "size").Exists())
 	require.Equal(t, ImageBillingSize1K, result.ImageSize)
 	require.Equal(t, "1024x1024", result.ImageInputSize)
 }
@@ -1372,6 +1374,58 @@ func TestForwardGrokMediaImagesEditMultipartConvertsToJSON(t *testing.T) {
 	require.Equal(t, "vendor-image-edit", result.UpstreamModel)
 }
 
+func TestForwardGrokMediaImagesEditMultipartPreservesExplicitGeometry(t *testing.T) {
+	t.Setenv(xai.EnvAllowUnsafeURLOverrides, "true")
+	gin.SetMode(gin.TestMode)
+
+	var buf bytes.Buffer
+	writer := multipart.NewWriter(&buf)
+	require.NoError(t, writer.WriteField("model", "grok-imagine-edit"))
+	require.NoError(t, writer.WriteField("prompt", "edit this private image"))
+	require.NoError(t, writer.WriteField("size", "1024x1024"))
+	require.NoError(t, writer.WriteField("resolution", "2k"))
+	require.NoError(t, writer.WriteField("aspect_ratio", "16:9"))
+	partHeader := textproto.MIMEHeader{}
+	partHeader.Set("Content-Disposition", `form-data; name="image"; filename="input.png"`)
+	partHeader.Set("Content-Type", "image/png")
+	part, err := writer.CreatePart(partHeader)
+	require.NoError(t, err)
+	_, err = part.Write([]byte{0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a})
+	require.NoError(t, err)
+	require.NoError(t, writer.Close())
+
+	recorder := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(recorder)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/images/edits", bytes.NewReader(buf.Bytes()))
+	c.Request.Header.Set("Content-Type", writer.FormDataContentType())
+
+	account := &Account{
+		ID:          67,
+		Name:        "grok",
+		Platform:    PlatformGrok,
+		Type:        AccountTypeAPIKey,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"api_key":  "api-key",
+			"base_url": "https://xai.test/v1",
+		},
+	}
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"application/json"}},
+		Body:       io.NopCloser(strings.NewReader(`{"data":[{"url":"https://images.test/edited.png"}]}`)),
+	}}
+	svc := &OpenAIGatewayService{httpUpstream: upstream}
+
+	result, err := svc.ForwardGrokMedia(context.Background(), c, account, GrokMediaEndpointImagesEdits, "", buf.Bytes(), writer.FormDataContentType())
+	require.NoError(t, err)
+	require.False(t, gjson.GetBytes(upstream.lastBody, "size").Exists())
+	require.Equal(t, "2k", gjson.GetBytes(upstream.lastBody, "resolution").String())
+	require.Equal(t, "16:9", gjson.GetBytes(upstream.lastBody, "aspect_ratio").String())
+	require.Equal(t, ImageBillingSize1K, result.ImageSize)
+	require.Equal(t, "1024x1024", result.ImageInputSize)
+}
+
 func TestForwardGrokMediaVideoGenerationReturnsUsageAndResponseID(t *testing.T) {
 	t.Setenv(xai.EnvAllowUnsafeURLOverrides, "true")
 	gin.SetMode(gin.TestMode)
@@ -1802,10 +1856,10 @@ func TestForwardAsChatCompletionsForGrokStopFallsBackToXAIChatCompletions(t *tes
 	require.Equal(t, "Bearer access-token", upstream.lastReq.Header.Get("Authorization"))
 	require.NotEmpty(t, upstream.lastReq.Header.Get(grokConversationIDHeader))
 	require.NotEqual(t, "raw-client-cache-key", upstream.lastReq.Header.Get(grokConversationIDHeader))
-	require.Equal(t, "grok-4.5", gjson.GetBytes(upstream.lastBody, "model").String())
+	require.Equal(t, "grok-4.6", gjson.GetBytes(upstream.lastBody, "model").String())
 	require.False(t, gjson.GetBytes(upstream.lastBody, "prompt_cache_key").Exists())
 	require.Equal(t, "grok", result.Model)
-	require.Equal(t, "grok-4.5", result.UpstreamModel)
+	require.Equal(t, "grok-4.6", result.UpstreamModel)
 	require.Equal(t, 1, result.Usage.InputTokens)
 	require.Equal(t, 2, result.Usage.OutputTokens)
 	require.Equal(t, 1, result.Usage.CacheReadInputTokens)
@@ -1919,7 +1973,7 @@ func TestForwardGrokResponsesAPIKeyUsesXAIResponses(t *testing.T) {
 	require.Equal(t, "Bearer xai-test-key", upstream.lastReq.Header.Get("Authorization"))
 	require.Empty(t, upstream.lastReq.Header.Get("X-Grok-Client-Version"))
 	require.NotEqual(t, grokUpstreamUserAgent, upstream.lastReq.Header.Get("User-Agent"))
-	require.Equal(t, "grok-4.5", gjson.GetBytes(upstream.lastBody, "model").String())
+	require.Equal(t, "grok-4.6", gjson.GetBytes(upstream.lastBody, "model").String())
 	require.Equal(t, "resp_grok_api_key", result.ResponseID)
 	require.Equal(t, 2, result.Usage.InputTokens)
 	require.Equal(t, 1, result.Usage.OutputTokens)
@@ -2067,6 +2121,16 @@ func TestForwardGrokResponsesInvalidEncryptedContentRecoveryDoesNotOvermatch(t *
 	}
 }
 
+func TestGrokCompactionBlobRecoveryStripsCompactionItem(t *testing.T) {
+	body := []byte(`{"model":"grok","input":[{"type":"compaction","id":"cmp_1","encrypted_content":"blob"},{"type":"message","role":"user","content":"hi"}]}`)
+	require.True(t, isGrokInvalidEncryptedContentResponse(http.StatusUnprocessableEntity, []byte(`{"code":"invalid_compaction","error":"could not decode the compaction blob"}`)))
+	retry, changed, err := trimGrokInvalidEncryptedContentRetryBody(body)
+	require.NoError(t, err)
+	require.True(t, changed)
+	require.Equal(t, "message", gjson.GetBytes(retry, "input.0.type").String())
+	require.False(t, gjson.GetBytes(retry, "input.#(type==\"compaction\")").Exists())
+}
+
 func TestForwardGrokResponsesInvalidEncryptedContentRecoveryNestedErrorShape(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 
@@ -2371,7 +2435,7 @@ func TestForwardAsChatCompletionsForGrokStreamingUsesRawXAIChatCompletions(t *te
 	require.Equal(t, "Bearer access-token", upstream.lastReq.Header.Get("Authorization"))
 	require.Equal(t, "text/event-stream", upstream.lastReq.Header.Get("Accept"))
 	require.Equal(t, xai.CLIUserAgent(xai.CLIClientVersion), upstream.lastReq.Header.Get("User-Agent"))
-	require.Equal(t, "grok-4.5", gjson.GetBytes(upstream.lastBody, "model").String())
+	require.Equal(t, "grok-4.6", gjson.GetBytes(upstream.lastBody, "model").String())
 	require.True(t, gjson.GetBytes(upstream.lastBody, "stream_options.include_usage").Bool())
 	require.True(t, result.Stream)
 	require.Equal(t, 6, result.Usage.InputTokens)
@@ -2424,7 +2488,7 @@ func TestForwardGrokResponsesNonStreamingUsesCacheIdentityAndCachedUsage(t *test
 	require.Equal(t, 7, result.Usage.InputTokens)
 	require.Equal(t, 2, result.Usage.OutputTokens)
 	require.Equal(t, 4, result.Usage.CacheReadInputTokens)
-	require.Equal(t, "grok-4.5", gjson.GetBytes(upstream.lastBody, "model").String())
+	require.Equal(t, "grok-4.6", gjson.GetBytes(upstream.lastBody, "model").String())
 	identity := gjson.GetBytes(upstream.lastBody, "prompt_cache_key").String()
 	require.NotEmpty(t, identity)
 	require.Equal(t, identity, upstream.lastReq.Header.Get(grokConversationIDHeader))
@@ -2547,7 +2611,7 @@ func TestForwardAsChatCompletionsForGrokStreamingStopFallsBackToRawXAIChatComple
 	require.Equal(t, grokCLIVersion, upstream.lastReq.Header.Get("X-Grok-Client-Version"))
 	require.NotEmpty(t, upstream.lastReq.Header.Get(grokConversationIDHeader))
 	require.NotEqual(t, "native-client-conversation", upstream.lastReq.Header.Get(grokConversationIDHeader))
-	require.Equal(t, "grok-4.5", gjson.GetBytes(upstream.lastBody, "model").String())
+	require.Equal(t, "grok-4.6", gjson.GetBytes(upstream.lastBody, "model").String())
 	require.True(t, gjson.GetBytes(upstream.lastBody, "stream_options.include_usage").Bool())
 	require.True(t, result.Stream)
 	require.Equal(t, 6, result.Usage.InputTokens)
@@ -2653,7 +2717,7 @@ func TestForwardAsAnthropicForGrokUsesXAIResponses(t *testing.T) {
 	require.Equal(t, "grok-experimental", upstream.lastReq.Header.Get("OpenAI-Beta"))
 	require.Empty(t, upstream.lastReq.Header.Get("originator"))
 	require.Empty(t, upstream.lastReq.Header.Get("version"))
-	require.Equal(t, "grok-4.5", gjson.GetBytes(upstream.lastBody, "model").String())
+	require.Equal(t, "grok-4.6", gjson.GetBytes(upstream.lastBody, "model").String())
 	require.NotEmpty(t, gjson.GetBytes(upstream.lastBody, "prompt_cache_key").String())
 	require.Equal(t, gjson.GetBytes(upstream.lastBody, "prompt_cache_key").String(), upstream.lastReq.Header.Get(grokConversationIDHeader))
 	require.Equal(t, "web_search", gjson.GetBytes(upstream.lastBody, "tools.0.type").String())
@@ -2663,7 +2727,7 @@ func TestForwardAsAnthropicForGrokUsesXAIResponses(t *testing.T) {
 	require.True(t, gjson.GetBytes(upstream.lastBody, "stream").Bool())
 	require.NotContains(t, string(upstream.lastBody), "chatgpt.com")
 	require.Equal(t, "grok", result.Model)
-	require.Equal(t, "grok-4.5", result.UpstreamModel)
+	require.Equal(t, "grok-4.6", result.UpstreamModel)
 	require.Equal(t, 5, result.Usage.InputTokens)
 	require.Equal(t, 2, result.Usage.OutputTokens)
 	require.Equal(t, 3, result.Usage.CacheReadInputTokens)
@@ -3451,6 +3515,43 @@ func TestPatchGrokResponsesBody_StripsReasoningContentNull(t *testing.T) {
 	require.False(t, reasoning.Get("content").Exists(), "content: null should be stripped")
 }
 
+func TestPatchGrokResponsesBody_StripsNestedInputNulls(t *testing.T) {
+	t.Parallel()
+
+	body := []byte(`{
+		"model": "grok-4.6",
+		"input": [
+			{"type":"message","role":"user","content":[{"type":"input_text","text":"hi","logprobs":null}]}
+		]
+	}`)
+	patched, err := patchGrokResponsesBody(body, "grok-4.6")
+	require.NoError(t, err)
+	require.False(t, gjson.GetBytes(patched, "input.0.content.0.logprobs").Exists())
+	require.Equal(t, "hi", gjson.GetBytes(patched, "input.0.content.0.text").String())
+}
+
+func TestStripExplicitNullsFromGrokInputSkipsCompaction(t *testing.T) {
+	t.Parallel()
+
+	input := []any{
+		map[string]any{"type": "message", "role": "user", "content": []any{map[string]any{"type": "input_text", "text": "hi", "logprobs": nil}}},
+		map[string]any{"type": "compaction", "id": "cmp_1", "encrypted_content": "blob", "status": nil},
+	}
+	out, changed := stripExplicitNullsFromGrokInput(input)
+	require.True(t, changed)
+	items, ok := out.([]any)
+	require.True(t, ok)
+	require.Len(t, items, 2)
+	msg := items[0].(map[string]any)
+	content := msg["content"].([]any)[0].(map[string]any)
+	_, hasLogprobs := content["logprobs"]
+	require.False(t, hasLogprobs)
+	compaction := items[1].(map[string]any)
+	require.Equal(t, "blob", compaction["encrypted_content"])
+	_, hasStatus := compaction["status"]
+	require.True(t, hasStatus, "compaction items must stay unmodified")
+}
+
 func TestPatchGrokResponsesBody_KeepsReasoningContentNonNull(t *testing.T) {
 	t.Parallel()
 
diff --git a/backend/internal/service/openai_gateway_response_handling.go b/backend/internal/service/openai_gateway_response_handling.go
index 779ade3cee..45d08e86f3 100644
--- a/backend/internal/service/openai_gateway_response_handling.go
+++ b/backend/internal/service/openai_gateway_response_handling.go
@@ -15,6 +15,7 @@ import (
 
 	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/xai"
 	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
 	"github.com/gin-gonic/gin"
 	"github.com/google/uuid"
@@ -1396,6 +1397,17 @@ func openAIUsageFromGJSON(value gjson.Result) (OpenAIUsage, bool) {
 	if outputTokens == 0 {
 		outputTokens = value.Get("completion_tokens").Int()
 	}
+	// xAI reports visible output separately from reasoning_tokens; OpenAI
+	// folds reasoning into completion/output. Use total_tokens to tell them apart.
+	reasoningTokens := max(int(firstPositiveGJSONInt(
+		value.Get("completion_tokens_details.reasoning_tokens"),
+		value.Get("output_tokens_details.reasoning_tokens"),
+	)), 0)
+	if reasoningTokens > 0 {
+		outputTokens = xai.IncludeIndependentReasoningTokens(
+			inputTokens, outputTokens, value.Get("total_tokens").Int(), int64(reasoningTokens),
+		)
+	}
 	cacheReadTokens := openAICacheReadTokensFromUsage(value)
 	cacheCreationTokens := openAICacheCreationTokensFromUsage(value)
 	imageOutputTokens := value.Get("output_tokens_details.image_tokens").Int()
diff --git a/backend/internal/service/openai_gateway_service_test.go b/backend/internal/service/openai_gateway_service_test.go
index 8c1bdc293a..122a542964 100644
--- a/backend/internal/service/openai_gateway_service_test.go
+++ b/backend/internal/service/openai_gateway_service_test.go
@@ -3509,6 +3509,30 @@ func TestExtractOpenAIUsageFromJSONBytes_AcceptsResponseAndChatUsageShapes(t *te
 	usage, ok = extractOpenAIUsageFromJSONBytes([]byte(`{"usage":{"input_tokens":20,"output_tokens":2,"cache_read_input_tokens":19,"input_tokens_details":{"cached_tokens":0}}}`))
 	require.True(t, ok)
 	require.Zero(t, usage.CacheReadInputTokens, "官方嵌套缓存读取字段显式为零时仍应优先于兼容顶层别名")
+
+	// xAI reports reasoning_tokens outside visible output_tokens. Only the
+	// arithmetic-consistent shape is independent; OpenAI's canonical shape
+	// already includes reasoning in completion/output_tokens.
+	usage, ok = extractOpenAIUsageFromJSONBytes([]byte(`{"usage":{"input_tokens":10000,"output_tokens":500,"total_tokens":10800,"output_tokens_details":{"reasoning_tokens":300}}}`))
+	require.True(t, ok)
+	require.Equal(t, 800, usage.OutputTokens)
+	usage, ok = extractOpenAIUsageFromJSONBytes([]byte(`{"usage":{"input_tokens":10000,"output_tokens":500,"total_tokens":10500,"output_tokens_details":{"reasoning_tokens":300}}}`))
+	require.True(t, ok)
+	require.Equal(t, 500, usage.OutputTokens)
+}
+
+func TestExtractOpenAIUsageFromJSONBytes_IncludesGrokReasoningTokens(t *testing.T) {
+	usage, ok := extractOpenAIUsageFromJSONBytes([]byte(`{"usage":{"prompt_tokens":32,"completion_tokens":9,"total_tokens":135,"completion_tokens_details":{"reasoning_tokens":94}}}`))
+	require.True(t, ok)
+	require.Equal(t, 103, usage.OutputTokens, "Grok Chat usage bills visible completion plus reasoning tokens")
+
+	usage, ok = extractOpenAIUsageFromJSONBytes([]byte(`{"usage":{"input_tokens":32,"output_tokens":103,"total_tokens":135,"output_tokens_details":{"reasoning_tokens":94}}}`))
+	require.True(t, ok)
+	require.Equal(t, 103, usage.OutputTokens, "Responses output_tokens already includes reasoning when total confirms it")
+
+	usage, ok = extractOpenAIUsageFromJSONBytes([]byte(`{"usage":{"input_tokens":32,"output_tokens":9,"total_tokens":135,"output_tokens_details":{"reasoning_tokens":94}}}`))
+	require.True(t, ok)
+	require.Equal(t, 103, usage.OutputTokens, "Responses detail-only shape is normalized when total exposes the full output")
 }
 
 func TestExtractCodexFinalResponse_SampleReplay(t *testing.T) {
diff --git a/backend/internal/service/openai_ws_http_bridge_test.go b/backend/internal/service/openai_ws_http_bridge_test.go
index 5825d180ff..a8c10cf893 100644
--- a/backend/internal/service/openai_ws_http_bridge_test.go
+++ b/backend/internal/service/openai_ws_http_bridge_test.go
@@ -1298,7 +1298,7 @@ func TestProxyResponsesWebSocketFromClientForGrokUsesXAIHTTPBridgeAndPreservesMa
 	require.Equal(t, "Bearer access-token", upstream.lastReq.Header.Get("Authorization"))
 	require.Equal(t, xai.CLIUserAgent(xai.CLIClientVersion), upstream.lastReq.Header.Get("User-Agent"))
 	require.Equal(t, grokCLIVersion, upstream.lastReq.Header.Get("X-Grok-Client-Version"))
-	require.Equal(t, "grok-4.5", gjson.GetBytes(upstream.bodies[0], "model").String())
+	require.Equal(t, "grok-4.6", gjson.GetBytes(upstream.bodies[0], "model").String())
 	require.Equal(t, "grok-4.3", gjson.GetBytes(upstream.bodies[1], "model").String())
 	require.Equal(t, "grok-4.3", gjson.GetBytes(upstream.bodies[2], "model").String())
 	require.NotEmpty(t, gjson.GetBytes(upstream.lastBody, "prompt_cache_key").String())
diff --git a/backend/internal/service/openai_ws_v2/passthrough_relay.go b/backend/internal/service/openai_ws_v2/passthrough_relay.go
index 992966bee5..e5e2633644 100644
--- a/backend/internal/service/openai_ws_v2/passthrough_relay.go
+++ b/backend/internal/service/openai_ws_v2/passthrough_relay.go
@@ -14,6 +14,8 @@ import (
 
 	coderws "github.com/coder/websocket"
 	"github.com/tidwall/gjson"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/xai"
 )
 
 type FrameConn interface {
@@ -1040,6 +1042,15 @@ func parseUsageAndAccumulate(
 		// 解析失败时不做部分字段累加，避免计费 usage 出现“半有效”状态。
 		return Usage{}
 	}
+	reasoningTokens := usageResult.Get("output_tokens_details.reasoning_tokens").Int()
+	if reasoningTokens == 0 {
+		reasoningTokens = usageResult.Get("completion_tokens_details.reasoning_tokens").Int()
+	}
+	if reasoningTokens > 0 {
+		outputTokens = int(xai.IncludeIndependentReasoningTokens(
+			int64(inputTokens), int64(outputTokens), usageResult.Get("total_tokens").Int(), reasoningTokens,
+		))
+	}
 	parsedUsage := Usage{
 		InputTokens:              inputTokens,
 		OutputTokens:             outputTokens,
diff --git a/backend/internal/service/openai_ws_v2/passthrough_relay_internal_test.go b/backend/internal/service/openai_ws_v2/passthrough_relay_internal_test.go
index 547af3fc62..cd02c3858f 100644
--- a/backend/internal/service/openai_ws_v2/passthrough_relay_internal_test.go
+++ b/backend/internal/service/openai_ws_v2/passthrough_relay_internal_test.go
@@ -325,6 +325,29 @@ func TestParseUsageAndEnrichCoverage(t *testing.T) {
 	enrichResult(nil, state, 0)
 }
 
+func TestParseUsageAndAccumulateIncludesIndependentReasoningTokens(t *testing.T) {
+	t.Parallel()
+
+	state := &relayState{}
+	got := parseUsageAndAccumulate(
+		state,
+		[]byte(`{"type":"response.completed","response":{"usage":{"input_tokens":32,"output_tokens":9,"total_tokens":151,"output_tokens_details":{"reasoning_tokens":110}}}}`),
+		"response.completed",
+		nil,
+	)
+	require.Equal(t, 32, got.InputTokens)
+	require.Equal(t, 119, got.OutputTokens)
+
+	state = &relayState{}
+	got = parseUsageAndAccumulate(
+		state,
+		[]byte(`{"type":"response.completed","response":{"usage":{"input_tokens":32,"output_tokens":119,"total_tokens":151,"output_tokens_details":{"reasoning_tokens":110}}}}`),
+		"response.completed",
+		nil,
+	)
+	require.Equal(t, 119, got.OutputTokens, "inclusive Responses output must not double-count reasoning")
+}
+
 func TestParseUsageAndAccumulateAcceptsChatUsageAliases(t *testing.T) {
 	t.Parallel()
 
diff --git a/backend/internal/service/setting_gateway_runtime.go b/backend/internal/service/setting_gateway_runtime.go
index 2f13278e3b..e7eda7c30e 100644
--- a/backend/internal/service/setting_gateway_runtime.go
+++ b/backend/internal/service/setting_gateway_runtime.go
@@ -457,6 +457,37 @@ func (s *SettingService) MigrateOpenAIAllowClaudeCodeCodexPluginSetting(ctx cont
 	return nil
 }
 
+// MigrateGrokDefaultTextModel upgrades the pre-4.6 built-in default for
+// existing installations. Explicit operator choices are left untouched.
+func (s *SettingService) MigrateGrokDefaultTextModel(ctx context.Context) error {
+	if s == nil || s.settingRepo == nil {
+		return nil
+	}
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	dbCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), codexRestrictionPolicyDBTimeout)
+	defer cancel()
+
+	value, err := s.settingRepo.GetValue(dbCtx, SettingKeyGrokDefaultTextModel)
+	if err != nil {
+		if errors.Is(err, ErrSettingNotFound) {
+			return nil
+		}
+		return fmt.Errorf("get %s setting: %w", SettingKeyGrokDefaultTextModel, err)
+	}
+	// Only migrate the value that was previously shipped as the built-in
+	// default. Any other value is an explicit operator choice or a future
+	// default and must remain unchanged.
+	if strings.TrimSpace(value) != "grok-4.5" {
+		return nil
+	}
+	if err := s.settingRepo.Set(dbCtx, SettingKeyGrokDefaultTextModel, "grok-4.6"); err != nil {
+		return fmt.Errorf("set %s setting: %w", SettingKeyGrokDefaultTextModel, err)
+	}
+	return nil
+}
+
 // MigrateCodexBodyFingerprintToSignals 把已废弃的 codex_cli_only_allow_body_engine_fingerprint
 // 开关并入引擎指纹信号列表。幂等:信号键已存在(非空)则不动;缺失时写默认种子,
 // 并把 body 路径行的 Required 设为旧 body 开关的值(旧 true ⇒ 勾上 body 行)。
diff --git a/backend/internal/service/setting_parse.go b/backend/internal/service/setting_parse.go
index 9f3dfbddf2..7787e8810f 100644
--- a/backend/internal/service/setting_parse.go
+++ b/backend/internal/service/setting_parse.go
@@ -193,7 +193,7 @@ func (s *SettingService) InitializeDefaultSettings(ctx context.Context) error {
 		SettingKeyChannelMonitorShowQuota:              "false",
 
 		// Grok: safe defaults — no cross-vendor model rewrite unless operators enable it.
-		SettingKeyGrokDefaultTextModel:           "grok-4.5",
+		SettingKeyGrokDefaultTextModel:           "grok-4.6",
 		SettingKeyGrokCrossClientModelMapEnabled: "true",
 		SettingKeyGrokDefaultBaseURLMode:         GrokDefaultBaseURLModeCLI,
 
@@ -806,7 +806,7 @@ func (s *SettingService) parseSettings(settings map[string]string) *SystemSettin
 	// Grok default mapping policy
 	result.GrokDefaultTextModel = strings.TrimSpace(settings[SettingKeyGrokDefaultTextModel])
 	if result.GrokDefaultTextModel == "" {
-		result.GrokDefaultTextModel = "grok-4.5"
+		result.GrokDefaultTextModel = "grok-4.6"
 	}
 	// Default true (missing/empty → enabled) so Claude/Codex→Grok mapping keeps working.
 	// Operators can set false to disable silent cross-client rewrite.
diff --git a/backend/internal/service/setting_service_codex_policy_test.go b/backend/internal/service/setting_service_codex_policy_test.go
index 1c17b91ef8..e426dea2c5 100644
--- a/backend/internal/service/setting_service_codex_policy_test.go
+++ b/backend/internal/service/setting_service_codex_policy_test.go
@@ -202,3 +202,34 @@ func TestMigrateCodexBodyFingerprintToSignals(t *testing.T) {
 		require.Equal(t, openai.DefaultEngineFingerprintSignalsJSON(), repo.values[SettingKeyCodexCLIOnlyEngineFingerprintSignals])
 	})
 }
+
+func TestMigrateGrokDefaultTextModel(t *testing.T) {
+	t.Run("upgrades legacy built-in default", func(t *testing.T) {
+		repo := &codexPolicyMigrationRepoStub{values: map[string]string{
+			SettingKeyGrokDefaultTextModel: "grok-4.5",
+		}}
+		svc := NewSettingService(repo, &config.Config{})
+		require.NoError(t, svc.MigrateGrokDefaultTextModel(context.Background()))
+		require.Equal(t, "grok-4.6", repo.values[SettingKeyGrokDefaultTextModel])
+		require.Equal(t, "grok-4.6", repo.sets[SettingKeyGrokDefaultTextModel])
+	})
+
+	t.Run("does not overwrite an explicit model", func(t *testing.T) {
+		repo := &codexPolicyMigrationRepoStub{values: map[string]string{
+			SettingKeyGrokDefaultTextModel: "grok-4.3",
+		}}
+		svc := NewSettingService(repo, &config.Config{})
+		require.NoError(t, svc.MigrateGrokDefaultTextModel(context.Background()))
+		require.Equal(t, "grok-4.3", repo.values[SettingKeyGrokDefaultTextModel])
+		_, wrote := repo.sets[SettingKeyGrokDefaultTextModel]
+		require.False(t, wrote)
+	})
+
+	t.Run("missing setting is left for normal defaults", func(t *testing.T) {
+		repo := &codexPolicyMigrationRepoStub{values: map[string]string{}}
+		svc := NewSettingService(repo, &config.Config{})
+		require.NoError(t, svc.MigrateGrokDefaultTextModel(context.Background()))
+		_, wrote := repo.sets[SettingKeyGrokDefaultTextModel]
+		require.False(t, wrote)
+	})
+}
diff --git a/backend/internal/service/setting_update.go b/backend/internal/service/setting_update.go
index a4fadded0c..65e28ff6ee 100644
--- a/backend/internal/service/setting_update.go
+++ b/backend/internal/service/setting_update.go
@@ -423,7 +423,7 @@ func (s *SettingService) buildSystemSettingsUpdates(ctx context.Context, setting
 	if v := strings.TrimSpace(settings.GrokDefaultTextModel); v != "" {
 		updates[SettingKeyGrokDefaultTextModel] = v
 	} else {
-		updates[SettingKeyGrokDefaultTextModel] = "grok-4.5"
+		updates[SettingKeyGrokDefaultTextModel] = "grok-4.6"
 	}
 	updates[SettingKeyGrokCrossClientModelMapEnabled] = strconv.FormatBool(settings.GrokCrossClientModelMapEnabled)
 	updates[SettingKeyGrokDefaultBaseURLMode] = normalizeGrokDefaultBaseURLMode(settings.GrokDefaultBaseURLMode)
diff --git a/backend/internal/service/wire.go b/backend/internal/service/wire.go
index 10467f944c..4c34a62e30 100644
--- a/backend/internal/service/wire.go
+++ b/backend/internal/service/wire.go
@@ -748,6 +748,9 @@ func ProvideSettingService(settingRepo SettingRepository, groupRepo GroupReposit
 	if err := svc.MigrateCodexBodyFingerprintToSignals(context.Background()); err != nil {
 		logger.LegacyPrintf("service.setting", "Warning: migrate codex body fingerprint to signals failed: %v", err)
 	}
+	if err := svc.MigrateGrokDefaultTextModel(context.Background()); err != nil {
+		logger.LegacyPrintf("service.setting", "Warning: migrate Grok default text model failed: %v", err)
+	}
 	antigravity.SetUserAgentVersionResolver(svc.GetAntigravityUserAgentVersion)
 	// enforceCodexIdentityHeaders 是所有 Codex 出站路径共用的纯函数收口点，拿不到 ctx，
 	// 故注入无参解析器；解析器内部自带 60s TTL 缓存，热路径不触库。
diff --git a/deploy/config.example.yaml b/deploy/config.example.yaml
index 8f134dd609..38a50d0607 100644
--- a/deploy/config.example.yaml
+++ b/deploy/config.example.yaml
@@ -214,6 +214,9 @@ gateway:
   # OpenAI/Codex upstream response header timeout (seconds, 0=disabled)
   # OpenAI/Codex 等待上游响应头超时时间（秒，0=禁用本地响应头超时）
   openai_response_header_timeout: 0
+  # Grok text Responses first-byte timeout (seconds, 0=disabled; default 120)
+  # Grok 文本 Responses 首字节超时时间（秒，0=禁用；默认 120）
+  grok_response_header_timeout: 120
   # Native OpenAI HTTP Responses first semantic output timeout (seconds, 0=disabled)
   # Includes response-header wait; does not apply to passthrough or WebSocket transports.
   # A timed-out request may already have incurred upstream usage; account failover can therefore duplicate upstream billing.
diff --git a/frontend/src/composables/__tests__/useModelWhitelist.spec.ts b/frontend/src/composables/__tests__/useModelWhitelist.spec.ts
index 9fd93e6573..af6cc61820 100644
--- a/frontend/src/composables/__tests__/useModelWhitelist.spec.ts
+++ b/frontend/src/composables/__tests__/useModelWhitelist.spec.ts
@@ -51,7 +51,8 @@ describe('useModelWhitelist', () => {
     expect(models).toContain('grok-4.5')
     expect(models).toContain('grok-4.5-latest')
     expect(models).toContain('grok-build-latest')
-    expect(models).toContain('grok-imagine-video-1.5-preview')
+    expect(models).toContain('grok-imagine-image-2.0')
+    expect(models).toContain('grok-imagine-video-1.5')
   })
 
   it('combined 模式支持 Grok 4.5 官方别名映射', () => {
diff --git a/frontend/src/composables/useModelWhitelist.ts b/frontend/src/composables/useModelWhitelist.ts
index 2ca860c673..4b16d44b02 100644
--- a/frontend/src/composables/useModelWhitelist.ts
+++ b/frontend/src/composables/useModelWhitelist.ts
@@ -157,8 +157,8 @@ const xaiModels = [
   'grok-imagine',
   'grok-imagine-image-quality',
   'grok-imagine-image',
+  'grok-imagine-image-2.0',
   'grok-imagine-video',
-  'grok-imagine-video-1.5-preview',
   'grok-imagine-video-1.5'
 ]