diff --git a/backend/internal/service/antigravity_credits_overages.go b/backend/internal/service/antigravity_credits_overages.go
index 3e19c563ff..e1b787415d 100644
--- a/backend/internal/service/antigravity_credits_overages.go
+++ b/backend/internal/service/antigravity_credits_overages.go
@@ -279,9 +279,10 @@ func resolveCreditsOveragesModelKey(ctx context.Context, account *Account, upstr
 
 // shouldMarkCreditsExhausted 判断一次 credits 请求失败是否应标记为 credits 耗尽。
 // 此函数在积分注入后失败时调用（预检查注入 + attemptCreditsOveragesRetry 两条路径）。
-//   - 429 + 非单模型限流：积分注入后仍 429 → 标记耗尽。
 //   - 429 + 单模型限流（"exhausted your capacity on this model"）：该模型免费配额用完，
 //     积分注入对此无效，但账号积分对其他模型可能仍可用 → 不标记积分耗尽。
+//   - 429 + 积分关键词（如 "insufficient credits"）：积分确实不足 → 标记耗尽。
+//   - 429 + 无积分关键词（节点限流、瞬时 rate limit 等）：与积分无关 → 不标记。
 //   - 403 等其他 4xx：检查 body 是否包含积分不足的关键词。
 //
 // clearCreditsExhausted 会在后续成功时自动清除。
@@ -299,9 +300,16 @@ func shouldMarkCreditsExhausted(resp *http.Response, respBody []byte, reqErr err
 		if strings.Contains(bodyLower, "exhausted your capacity on this model") {
 			return false
 		}
-		return true
+		// 仅当 body 包含积分相关关键词时才标记耗尽；
+		// 节点限流、瞬时 rate limit 等非积分 429 不应触发 5 小时熔断。
+		return containsCreditsExhaustedKeyword(bodyLower)
 	}
 	// 其他 4xx：关键词匹配（如 403 + "Insufficient credits"）
+	return containsCreditsExhaustedKeyword(bodyLower)
+}
+
+// containsCreditsExhaustedKeyword 检查响应体是否包含积分耗尽的关键词。
+func containsCreditsExhaustedKeyword(bodyLower string) bool {
 	for _, keyword := range creditsExhaustedKeywords {
 		if strings.Contains(bodyLower, keyword) {
 			return true
diff --git a/backend/internal/service/antigravity_credits_overages_test.go b/backend/internal/service/antigravity_credits_overages_test.go
index f2143f06cd..1e9106b41d 100644
--- a/backend/internal/service/antigravity_credits_overages_test.go
+++ b/backend/internal/service/antigravity_credits_overages_test.go
@@ -424,9 +424,15 @@ func TestShouldMarkCreditsExhausted(t *testing.T) {
 		require.False(t, shouldMarkCreditsExhausted(resp, body, nil))
 	})
 
-	t.Run("429 结构化限流也标记（积分注入后仍 429 即为耗尽）", func(t *testing.T) {
+	t.Run("429 结构化限流不标记（节点限流/瞬时 rate limit 与积分无关）", func(t *testing.T) {
 		resp := &http.Response{StatusCode: http.StatusTooManyRequests}
 		body := []byte(`{"error":{"status":"RESOURCE_EXHAUSTED","details":[{"@type":"type.googleapis.com/google.rpc.ErrorInfo","reason":"RATE_LIMIT_EXCEEDED"},{"@type":"type.googleapis.com/google.rpc.RetryInfo","retryDelay":"0.5s"}]}}`)
+		require.False(t, shouldMarkCreditsExhausted(resp, body, nil))
+	})
+
+	t.Run("429 含积分关键词时标记", func(t *testing.T) {
+		resp := &http.Response{StatusCode: http.StatusTooManyRequests}
+		body := []byte(`{"error":{"message":"Insufficient credits to complete this request"}}`)
 		require.True(t, shouldMarkCreditsExhausted(resp, body, nil))
 	})
 
diff --git a/backend/internal/service/gateway_service.go b/backend/internal/service/gateway_service.go
index e01b3b7469..f18954ca7e 100644
--- a/backend/internal/service/gateway_service.go
+++ b/backend/internal/service/gateway_service.go
@@ -3006,7 +3006,7 @@ func (s *GatewayService) selectAccountForModelWithPlatform(ctx context.Context,
 						if clearSticky {
 							_ = s.cache.DeleteSessionAccountID(ctx, derefGroupID(groupID), sessionHash)
 						}
-						if !clearSticky && s.isAccountInGroup(account, groupID) && account.Platform == platform && (requestedModel == "" || s.isModelSupportedByAccountWithContext(ctx, account, requestedModel)) && s.isAccountSchedulableForModelSelection(ctx, account, requestedModel) && s.isAccountSchedulableForQuota(account) && s.isAccountSchedulableForWindowCost(ctx, account, true) && s.isAccountSchedulableForRPM(ctx, account, true) {
+						if !clearSticky && s.isAccountInGroup(account, groupID) && account.Platform == platform && (requestedModel == "" || s.isModelSupportedByAccountWithContext(ctx, account, requestedModel)) && s.isAccountSchedulableForModelSelection(ctx, account, requestedModel) && s.isAccountSchedulableForQuota(account) && s.isAccountSchedulableForWindowCost(ctx, account, true) && s.isAccountSchedulableForRPM(ctx, account, true) && !s.isStickyAccountUpstreamRestricted(ctx, groupID, account, requestedModel) {
 							if s.debugModelRoutingEnabled() {
 								logger.LegacyPrintf("service.gateway", "[ModelRoutingDebug] legacy routed sticky hit: group_id=%v model=%s session=%s account=%d", derefGroupID(groupID), requestedModel, shortSessionHash(sessionHash), accountID)
 							}
@@ -3361,7 +3361,7 @@ func (s *GatewayService) selectAccountWithMixedScheduling(ctx context.Context, g
 					if clearSticky {
 						_ = s.cache.DeleteSessionAccountID(ctx, derefGroupID(groupID), sessionHash)
 					}
-					if !clearSticky && s.isAccountInGroup(account, groupID) && (requestedModel == "" || s.isModelSupportedByAccountWithContext(ctx, account, requestedModel)) && s.isAccountSchedulableForModelSelection(ctx, account, requestedModel) && s.isAccountSchedulableForQuota(account) && s.isAccountSchedulableForWindowCost(ctx, account, true) && s.isAccountSchedulableForRPM(ctx, account, true) {
+					if !clearSticky && s.isAccountInGroup(account, groupID) && (requestedModel == "" || s.isModelSupportedByAccountWithContext(ctx, account, requestedModel)) && s.isAccountSchedulableForModelSelection(ctx, account, requestedModel) && s.isAccountSchedulableForQuota(account) && s.isAccountSchedulableForWindowCost(ctx, account, true) && s.isAccountSchedulableForRPM(ctx, account, true) && !s.isStickyAccountUpstreamRestricted(ctx, groupID, account, requestedModel) {
 						if account.Platform == nativePlatform || (account.Platform == PlatformAntigravity && account.IsMixedSchedulingEnabled()) {
 							return account, nil
 						}
@@ -3385,7 +3385,6 @@ func (s *GatewayService) selectAccountWithMixedScheduling(ctx context.Context, g
 	ctx = s.withRPMPrefetch(ctx, accounts)
 
 	// 3. 按优先级+最久未用选择（考虑模型支持和混合调度）
-	// needsUpstreamCheck 仅在主选择循环中使用；粘性会话命中时跳过此检查。
 	needsUpstreamCheck := s.needsUpstreamChannelRestrictionCheck(ctx, groupID)
 	var selected *Account
 	for i := range accounts {
@@ -8455,6 +8454,19 @@ func (s *GatewayService) needsUpstreamChannelRestrictionCheck(ctx context.Contex
 	return ch.BillingModelSource == BillingModelSourceUpstream
 }
 
+// isStickyAccountUpstreamRestricted 检查粘性会话命中的账号是否受 upstream 渠道限制。
+// 合并 needsUpstreamChannelRestrictionCheck + isUpstreamModelRestrictedByChannel 两步调用，
+// 供 sticky session 条件链使用，避免内联多个函数调用导致行过长。
+func (s *GatewayService) isStickyAccountUpstreamRestricted(ctx context.Context, groupID *int64, account *Account, requestedModel string) bool {
+	if groupID == nil {
+		return false
+	}
+	if !s.needsUpstreamChannelRestrictionCheck(ctx, groupID) {
+		return false
+	}
+	return s.isUpstreamModelRestrictedByChannel(ctx, *groupID, account, requestedModel)
+}
+
 // ForwardCountTokens 转发 count_tokens 请求到上游 API
 // 特点：不记录使用量、仅支持非流式响应
 func (s *GatewayService) ForwardCountTokens(ctx context.Context, c *gin.Context, account *Account, parsed *ParsedRequest) error {