refactor: replace sync.Map credits state with AICredits rate limit key

Replace process-memory sync.Map + per-model runtime state with a single "AICredits" key in model_rate_limits, making credits exhaustion fully isomorphic with model-level rate limiting. Scheduler: rate-limited accounts with overages enabled + credits available are now scheduled instead of excluded. Forwarding: when model is rate-limited + credits available, inject credits proactively without waiting for a 429 round trip. Storage: credits exhaustion stored as model_rate_limits["AICredits"] with 5h duration, reusing SetModelRateLimit/isRateLimitActiveForKey. Frontend: show credits_active (yellow ⚡) when model rate-limited but credits available, credits_exhausted (red) when AICredits key active. Tests: add unit tests for shouldMarkCreditsExhausted, injectEnabledCreditTypes, clearCreditsExhausted, and update existing overages tests.
2026-04-11 10:34:46 +08:00 · 2026-03-16 04:31:22 +08:00
parent e14c87597a
commit 8a260defc2
12 changed files with 692 additions and 327 deletions
--- a/backend/internal/service/admin_service.go
+++ b/backend/internal/service/admin_service.go
@@ -143,9 +143,10 @@ type CreateGroupInput struct {
 	// 无效请求兜底分组 ID（仅 anthropic 平台使用）
 	FallbackGroupIDOnInvalidRequest *int64
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64
-	ModelRoutingEnabled bool // 是否启用模型路由
-	MCPXMLInject        *bool
+	ModelRouting             map[string][]int64
+	ModelRoutingEnabled      bool // 是否启用模型路由
+	MCPXMLInject             *bool
+	SimulateClaudeMaxEnabled *bool
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string
 	// Sora 存储配额
@@ -182,9 +183,10 @@ type UpdateGroupInput struct {
 	// 无效请求兜底分组 ID（仅 anthropic 平台使用）
 	FallbackGroupIDOnInvalidRequest *int64
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64
-	ModelRoutingEnabled *bool // 是否启用模型路由
-	MCPXMLInject        *bool
+	ModelRouting             map[string][]int64
+	ModelRoutingEnabled      *bool // 是否启用模型路由
+	MCPXMLInject             *bool
+	SimulateClaudeMaxEnabled *bool
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes *[]string
 	// Sora 存储配额
@@ -368,6 +370,10 @@ type ProxyExitInfoProber interface {
 	ProbeProxy(ctx context.Context, proxyURL string) (*ProxyExitInfo, int64, error)
 }

+type groupExistenceBatchReader interface {
+	ExistsByIDs(ctx context.Context, ids []int64) (map[int64]bool, error)
+}
+
 type proxyQualityTarget struct {
 	Target          string
 	URL             string
@@ -445,10 +451,6 @@ type userGroupRateBatchReader interface {
 	GetByUserIDs(ctx context.Context, userIDs []int64) (map[int64]map[int64]float64, error)
 }

-type groupExistenceBatchReader interface {
-	ExistsByIDs(ctx context.Context, ids []int64) (map[int64]bool, error)
-}
-
 // NewAdminService creates a new AdminService
 func NewAdminService(
 	userRepo UserRepository,
@@ -868,6 +870,13 @@ func (s *adminServiceImpl) CreateGroup(ctx context.Context, input *CreateGroupIn
 	if input.MCPXMLInject != nil {
 		mcpXMLInject = *input.MCPXMLInject
 	}
+	simulateClaudeMaxEnabled := false
+	if input.SimulateClaudeMaxEnabled != nil {
+		if platform != PlatformAnthropic && *input.SimulateClaudeMaxEnabled {
+			return nil, fmt.Errorf("simulate_claude_max_enabled only supported for anthropic groups")
+		}
+		simulateClaudeMaxEnabled = *input.SimulateClaudeMaxEnabled
+	}

 	// 如果指定了复制账号的源分组，先获取账号 ID 列表
 	var accountIDsToCopy []int64
@@ -924,6 +933,7 @@ func (s *adminServiceImpl) CreateGroup(ctx context.Context, input *CreateGroupIn
 		FallbackGroupIDOnInvalidRequest: fallbackOnInvalidRequest,
 		ModelRouting:                    input.ModelRouting,
 		MCPXMLInject:                    mcpXMLInject,
+		SimulateClaudeMaxEnabled:        simulateClaudeMaxEnabled,
 		SupportedModelScopes:            input.SupportedModelScopes,
 		SoraStorageQuotaBytes:           input.SoraStorageQuotaBytes,
 		AllowMessagesDispatch:           input.AllowMessagesDispatch,
@@ -1130,6 +1140,15 @@ func (s *adminServiceImpl) UpdateGroup(ctx context.Context, id int64, input *Upd
 	if input.MCPXMLInject != nil {
 		group.MCPXMLInject = *input.MCPXMLInject
 	}
+	if input.SimulateClaudeMaxEnabled != nil {
+		if group.Platform != PlatformAnthropic && *input.SimulateClaudeMaxEnabled {
+			return nil, fmt.Errorf("simulate_claude_max_enabled only supported for anthropic groups")
+		}
+		group.SimulateClaudeMaxEnabled = *input.SimulateClaudeMaxEnabled
+	}
+	if group.Platform != PlatformAnthropic {
+		group.SimulateClaudeMaxEnabled = false
+	}

 	// 支持的模型系列（仅 antigravity 平台使用）
 	if input.SupportedModelScopes != nil {
@@ -1530,7 +1549,7 @@ func (s *adminServiceImpl) UpdateAccount(ctx context.Context, id int64, input *U
 	if len(input.Credentials) > 0 {
 		account.Credentials = input.Credentials
 	}
-	if input.Extra != nil {
+	if len(input.Extra) > 0 {
 		// 保留配额用量字段，防止编辑账号时意外重置
 		for _, key := range []string{"quota_used", "quota_daily_used", "quota_daily_start", "quota_weekly_used", "quota_weekly_start"} {
 			if v, ok := account.Extra[key]; ok {
@@ -1539,11 +1558,15 @@ func (s *adminServiceImpl) UpdateAccount(ctx context.Context, id int64, input *U
 		}
 		account.Extra = input.Extra
 		if account.Platform == PlatformAntigravity && wasOveragesEnabled && !account.IsOveragesEnabled() {
-			delete(account.Extra, antigravityCreditsOveragesKey)
+			delete(account.Extra, "antigravity_credits_overages") // 清理旧版 overages 运行态
+			// 清除 AICredits 限流 key
+			if rawLimits, ok := account.Extra[modelRateLimitsKey].(map[string]any); ok {
+				delete(rawLimits, creditsExhaustedKey)
+			}
 		}
 		if account.Platform == PlatformAntigravity && !wasOveragesEnabled && account.IsOveragesEnabled() {
 			delete(account.Extra, modelRateLimitsKey)
-			delete(account.Extra, antigravityCreditsOveragesKey)
+			delete(account.Extra, "antigravity_credits_overages") // 清理旧版 overages 运行态
 		}
 		// 校验并预计算固定时间重置的下次重置时间
 		if err := ValidateQuotaResetConfig(account.Extra); err != nil {
@@ -1627,14 +1650,6 @@ func (s *adminServiceImpl) UpdateAccount(ctx context.Context, id int64, input *U
 	if err := s.accountRepo.Update(ctx, account); err != nil {
 		return nil, err
 	}
-	if account.Platform == PlatformAntigravity {
-		if !account.IsOveragesEnabled() && wasOveragesEnabled {
-			clearCreditsExhausted(account.ID)
-		}
-		if account.IsOveragesEnabled() && !wasOveragesEnabled {
-			clearCreditsExhausted(account.ID)
-		}
-	}

 	// 绑定分组
 	if input.GroupIDs != nil {
--- a/backend/internal/service/admin_service_overages_test.go
+++ b/backend/internal/service/admin_service_overages_test.go
@@ -26,7 +26,7 @@ func (r *updateAccountOveragesRepoStub) Update(ctx context.Context, account *Acc
 	return nil
 }

-func TestUpdateAccount_DisableOveragesClearsRuntimeStateBeforePersist(t *testing.T) {
+func TestUpdateAccount_DisableOveragesClearsAICreditsKey(t *testing.T) {
 	accountID := int64(101)
 	repo := &updateAccountOveragesRepoStub{
 		account: &Account{
@@ -37,24 +37,34 @@ func TestUpdateAccount_DisableOveragesClearsRuntimeStateBeforePersist(t *testing
 			Extra: map[string]any{
 				"allow_overages":   true,
 				"mixed_scheduling": true,
-				antigravityCreditsOveragesKey: map[string]any{
+				modelRateLimitsKey: map[string]any{
 					"claude-sonnet-4-5": map[string]any{
-						"activated_at": "2026-03-15T00:00:00Z",
-						"active_until": "2099-03-15T00:00:00Z",
+						"rate_limited_at":     "2026-03-15T00:00:00Z",
+						"rate_limit_reset_at": "2099-03-15T00:00:00Z",
+					},
+					creditsExhaustedKey: map[string]any{
+						"rate_limited_at":     "2026-03-15T00:00:00Z",
+						"rate_limit_reset_at": time.Now().Add(5 * time.Hour).UTC().Format(time.RFC3339),
 					},
 				},
 			},
 		},
 	}
-	setCreditsExhausted(accountID, time.Now().Add(time.Minute))
-	t.Cleanup(func() {
-		clearCreditsExhausted(accountID)
-	})

 	svc := &adminServiceImpl{accountRepo: repo}
 	updated, err := svc.UpdateAccount(context.Background(), accountID, &UpdateAccountInput{
 		Extra: map[string]any{
 			"mixed_scheduling": true,
+			modelRateLimitsKey: map[string]any{
+				"claude-sonnet-4-5": map[string]any{
+					"rate_limited_at":     "2026-03-15T00:00:00Z",
+					"rate_limit_reset_at": "2099-03-15T00:00:00Z",
+				},
+				creditsExhaustedKey: map[string]any{
+					"rate_limited_at":     "2026-03-15T00:00:00Z",
+					"rate_limit_reset_at": time.Now().Add(5 * time.Hour).UTC().Format(time.RFC3339),
+				},
+			},
 		},
 	})

@@ -62,10 +72,17 @@ func TestUpdateAccount_DisableOveragesClearsRuntimeStateBeforePersist(t *testing
 	require.NotNil(t, updated)
 	require.Equal(t, 1, repo.updateCalls)
 	require.False(t, updated.IsOveragesEnabled())
-	require.False(t, isCreditsExhausted(accountID))

-	_, exists := repo.account.Extra[antigravityCreditsOveragesKey]
-	require.False(t, exists, "关闭 overages 时应在持久化前移除运行态")
+	// 关闭 overages 后，AICredits key 应被清除
+	rawLimits, ok := repo.account.Extra[modelRateLimitsKey].(map[string]any)
+	if ok {
+		_, exists := rawLimits[creditsExhaustedKey]
+		require.False(t, exists, "关闭 overages 时应清除 AICredits 限流 key")
+	}
+	// 普通模型限流应保留
+	require.True(t, ok)
+	_, exists := rawLimits["claude-sonnet-4-5"]
+	require.True(t, exists, "普通模型限流应保留")
 }

 func TestUpdateAccount_EnableOveragesClearsModelRateLimitsBeforePersist(t *testing.T) {
@@ -87,10 +104,6 @@ func TestUpdateAccount_EnableOveragesClearsModelRateLimitsBeforePersist(t *testi
 			},
 		},
 	}
-	setCreditsExhausted(accountID, time.Now().Add(time.Minute))
-	t.Cleanup(func() {
-		clearCreditsExhausted(accountID)
-	})

 	svc := &adminServiceImpl{accountRepo: repo}
 	updated, err := svc.UpdateAccount(context.Background(), accountID, &UpdateAccountInput{
@@ -104,7 +117,6 @@ func TestUpdateAccount_EnableOveragesClearsModelRateLimitsBeforePersist(t *testi
 	require.NotNil(t, updated)
 	require.Equal(t, 1, repo.updateCalls)
 	require.True(t, updated.IsOveragesEnabled())
-	require.False(t, isCreditsExhausted(accountID))

 	_, exists := repo.account.Extra[modelRateLimitsKey]
 	require.False(t, exists, "开启 overages 时应在持久化前清掉旧模型限流")
--- a/backend/internal/service/antigravity_credits_overages.go
+++ b/backend/internal/service/antigravity_credits_overages.go
@@ -6,14 +6,18 @@ import (
 	"io"
 	"net/http"
 	"strings"
-	"sync"
 	"time"

 	"github.com/Wei-Shaw/sub2api/internal/pkg/antigravity"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
 )

-const antigravityCreditsOveragesKey = "antigravity_credits_overages"
+const (
+	// creditsExhaustedKey 是 model_rate_limits 中标记积分耗尽的特殊 key。
+	// 与普通模型限流完全同构：通过 SetModelRateLimit / isRateLimitActiveForKey 读写。
+	creditsExhaustedKey      = "AICredits"
+	creditsExhaustedDuration = 5 * time.Hour
+)

 type antigravity429Category string

@@ -24,8 +28,6 @@ const (
 )

 var (
-	creditsExhaustedCache sync.Map
-
 	antigravityQuotaExhaustedKeywords = []string{
 		"quota_exhausted",
 		"quota exhausted",
@@ -46,28 +48,48 @@ var (
 	}
 )

-// isCreditsExhausted 检查账号的 AI Credits 是否已被标记为耗尽。
-func isCreditsExhausted(accountID int64) bool {
-	v, ok := creditsExhaustedCache.Load(accountID)
+// isCreditsExhausted 检查账号的 AICredits 限流 key 是否生效（积分是否耗尽）。
+func (a *Account) isCreditsExhausted() bool {
+	if a == nil {
+		return false
+	}
+	return a.isRateLimitActiveForKey(creditsExhaustedKey)
+}
+
+// setCreditsExhausted 标记账号积分耗尽：写入 model_rate_limits["AICredits"] + 更新缓存。
+func (s *AntigravityGatewayService) setCreditsExhausted(ctx context.Context, account *Account) {
+	if account == nil || account.ID == 0 {
+		return
+	}
+	resetAt := time.Now().Add(creditsExhaustedDuration)
+	if err := s.accountRepo.SetModelRateLimit(ctx, account.ID, creditsExhaustedKey, resetAt); err != nil {
+		logger.LegacyPrintf("service.antigravity_gateway", "set credits exhausted failed: account=%d err=%v", account.ID, err)
+		return
+	}
+	s.updateAccountModelRateLimitInCache(ctx, account, creditsExhaustedKey, resetAt)
+	logger.LegacyPrintf("service.antigravity_gateway", "credits_exhausted_marked account=%d reset_at=%s",
+		account.ID, resetAt.UTC().Format(time.RFC3339))
+}
+
+// clearCreditsExhausted 清除账号的 AICredits 限流 key。
+func (s *AntigravityGatewayService) clearCreditsExhausted(ctx context.Context, account *Account) {
+	if account == nil || account.ID == 0 || account.Extra == nil {
+		return
+	}
+	rawLimits, ok := account.Extra[modelRateLimitsKey].(map[string]any)
 	if !ok {
-		return false
+		return
 	}
-	until, ok := v.(time.Time)
-	if !ok || time.Now().After(until) {
-		creditsExhaustedCache.Delete(accountID)
-		return false
+	if _, exists := rawLimits[creditsExhaustedKey]; !exists {
+		return
+	}
+	delete(rawLimits, creditsExhaustedKey)
+	account.Extra[modelRateLimitsKey] = rawLimits
+	if err := s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{
+		modelRateLimitsKey: rawLimits,
+	}); err != nil {
+		logger.LegacyPrintf("service.antigravity_gateway", "clear credits exhausted failed: account=%d err=%v", account.ID, err)
 	}
-	return true
-}
-
-// setCreditsExhausted 将账号标记为 AI Credits 已耗尽，直到指定时间。
-func setCreditsExhausted(accountID int64, until time.Time) {
-	creditsExhaustedCache.Store(accountID, until)
-}
-
-// clearCreditsExhausted 清除账号的 AI Credits 耗尽标记。
-func clearCreditsExhausted(accountID int64) {
-	creditsExhaustedCache.Delete(accountID)
 }

 // classifyAntigravity429 将 Antigravity 的 429 响应归类为配额耗尽、限流或未知。
@@ -117,111 +139,6 @@ func resolveCreditsOveragesModelKey(ctx context.Context, account *Account, upstr
 	return resolveAntigravityModelKey(requestedModel)
 }

-// canUseAntigravityCreditsOverages 判断当前请求是否应直接走已激活的 overages 链路。
-func canUseAntigravityCreditsOverages(ctx context.Context, account *Account, requestedModel string) bool {
-	if account == nil || account.Platform != PlatformAntigravity {
-		return false
-	}
-	if !account.IsSchedulable() {
-		return false
-	}
-	if !account.IsOveragesEnabled() || isCreditsExhausted(account.ID) {
-		return false
-	}
-	return account.getAntigravityCreditsOveragesRemainingTimeWithContext(ctx, requestedModel) > 0
-}
-
-func (a *Account) getAntigravityCreditsOveragesRemainingTimeWithContext(ctx context.Context, requestedModel string) time.Duration {
-	if a == nil || a.Extra == nil {
-		return 0
-	}
-	modelKey := resolveFinalAntigravityModelKey(ctx, a, requestedModel)
-	modelKey = strings.TrimSpace(modelKey)
-	if modelKey == "" {
-		return 0
-	}
-	rawStates, ok := a.Extra[antigravityCreditsOveragesKey].(map[string]any)
-	if !ok {
-		return 0
-	}
-	rawState, ok := rawStates[modelKey].(map[string]any)
-	if !ok {
-		return 0
-	}
-	activeUntilRaw, ok := rawState["active_until"].(string)
-	if !ok || strings.TrimSpace(activeUntilRaw) == "" {
-		return 0
-	}
-	activeUntil, err := time.Parse(time.RFC3339, activeUntilRaw)
-	if err != nil {
-		return 0
-	}
-	remaining := time.Until(activeUntil)
-	if remaining > 0 {
-		return remaining
-	}
-	return 0
-}
-
-func setAntigravityCreditsOveragesActive(ctx context.Context, repo AccountRepository, account *Account, modelKey string, activeUntil time.Time) {
-	if repo == nil || account == nil || account.ID == 0 || strings.TrimSpace(modelKey) == "" {
-		return
-	}
-	stateMap := copyAntigravityCreditsOveragesState(account)
-	stateMap[modelKey] = map[string]any{
-		"activated_at": time.Now().UTC().Format(time.RFC3339),
-		"active_until": activeUntil.UTC().Format(time.RFC3339),
-	}
-	ensureAccountExtra(account)
-	account.Extra[antigravityCreditsOveragesKey] = stateMap
-	if err := repo.UpdateExtra(ctx, account.ID, map[string]any{antigravityCreditsOveragesKey: stateMap}); err != nil {
-		logger.LegacyPrintf("service.antigravity_gateway", "set overages state failed: account=%d model=%s err=%v", account.ID, modelKey, err)
-	}
-}
-
-func clearAntigravityCreditsOveragesState(ctx context.Context, repo AccountRepository, accountID int64) error {
-	if repo == nil || accountID == 0 {
-		return nil
-	}
-	return repo.UpdateExtra(ctx, accountID, map[string]any{
-		antigravityCreditsOveragesKey: map[string]any{},
-	})
-}
-
-func clearAntigravityCreditsOveragesStateForModel(ctx context.Context, repo AccountRepository, account *Account, modelKey string) {
-	if repo == nil || account == nil || account.ID == 0 {
-		return
-	}
-	stateMap := copyAntigravityCreditsOveragesState(account)
-	delete(stateMap, modelKey)
-	ensureAccountExtra(account)
-	account.Extra[antigravityCreditsOveragesKey] = stateMap
-	if err := repo.UpdateExtra(ctx, account.ID, map[string]any{antigravityCreditsOveragesKey: stateMap}); err != nil {
-		logger.LegacyPrintf("service.antigravity_gateway", "clear overages state failed: account=%d model=%s err=%v", account.ID, modelKey, err)
-	}
-}
-
-func copyAntigravityCreditsOveragesState(account *Account) map[string]any {
-	result := make(map[string]any)
-	if account == nil || account.Extra == nil {
-		return result
-	}
-	rawState, ok := account.Extra[antigravityCreditsOveragesKey].(map[string]any)
-	if !ok {
-		return result
-	}
-	for key, value := range rawState {
-		result[key] = value
-	}
-	return result
-}
-
-func ensureAccountExtra(account *Account) {
-	if account != nil && account.Extra == nil {
-		account.Extra = make(map[string]any)
-	}
-}
-
 // shouldMarkCreditsExhausted 判断一次 credits 请求失败是否应标记为 credits 耗尽。
 func shouldMarkCreditsExhausted(resp *http.Response, respBody []byte, reqErr error) bool {
 	if reqErr != nil || resp == nil {
@@ -276,23 +193,21 @@ func (s *AntigravityGatewayService) attemptCreditsOveragesRetry(

 	creditsResp, err := p.httpUpstream.Do(creditsReq, p.proxyURL, p.account.ID, p.account.Concurrency)
 	if err == nil && creditsResp != nil && creditsResp.StatusCode < 400 {
-		clearCreditsExhausted(p.account.ID)
-		activeUntil := s.resolveCreditsOveragesActiveUntil(respBody, waitDuration)
-		setAntigravityCreditsOveragesActive(p.ctx, p.accountRepo, p.account, modelKey, activeUntil)
-		logger.LegacyPrintf("service.antigravity_gateway", "%s status=%d credit_overages_success model=%s account=%d active_until=%s",
-			p.prefix, creditsResp.StatusCode, modelKey, p.account.ID, activeUntil.UTC().Format(time.RFC3339))
+		s.clearCreditsExhausted(p.ctx, p.account)
+		logger.LegacyPrintf("service.antigravity_gateway", "%s status=%d credit_overages_success model=%s account=%d",
+			p.prefix, creditsResp.StatusCode, modelKey, p.account.ID)
 		return &creditsOveragesRetryResult{handled: true, resp: creditsResp}
 	}

-	s.handleCreditsRetryFailure(p.prefix, modelKey, p.account, waitDuration, creditsResp, err)
+	s.handleCreditsRetryFailure(p.ctx, p.prefix, modelKey, p.account, creditsResp, err)
 	return &creditsOveragesRetryResult{handled: true}
 }

 func (s *AntigravityGatewayService) handleCreditsRetryFailure(
+	ctx context.Context,
 	prefix string,
 	modelKey string,
 	account *Account,
-	waitDuration time.Duration,
 	creditsResp *http.Response,
 	reqErr error,
 ) {
@@ -307,11 +222,9 @@ func (s *AntigravityGatewayService) handleCreditsRetryFailure(
 	}

 	if shouldMarkCreditsExhausted(creditsResp, creditsRespBody, reqErr) && account != nil {
-		exhaustedUntil := s.resolveCreditsOveragesActiveUntil(creditsRespBody, waitDuration)
-		setCreditsExhausted(account.ID, exhaustedUntil)
-		clearAntigravityCreditsOveragesStateForModel(context.Background(), s.accountRepo, account, modelKey)
-		logger.LegacyPrintf("service.antigravity_gateway", "%s credit_overages_failed model=%s account=%d marked_exhausted=true status=%d exhausted_until=%v body=%s",
-			prefix, modelKey, account.ID, creditsStatusCode, exhaustedUntil, truncateForLog(creditsRespBody, 200))
+		s.setCreditsExhausted(ctx, account)
+		logger.LegacyPrintf("service.antigravity_gateway", "%s credit_overages_failed model=%s account=%d marked_exhausted=true status=%d body=%s",
+			prefix, modelKey, account.ID, creditsStatusCode, truncateForLog(creditsRespBody, 200))
 		return
 	}
 	if account != nil {
@@ -320,11 +233,3 @@ func (s *AntigravityGatewayService) handleCreditsRetryFailure(
 	}
 }

-func (s *AntigravityGatewayService) resolveCreditsOveragesActiveUntil(respBody []byte, waitDuration time.Duration) time.Time {
-	resetAt := ParseGeminiRateLimitResetTime(respBody)
-	defaultDur := waitDuration
-	if defaultDur <= 0 {
-		defaultDur = s.getDefaultRateLimitDuration()
-	}
-	return s.resolveResetTime(resetAt, defaultDur)
-}
--- a/backend/internal/service/antigravity_credits_overages_test.go
+++ b/backend/internal/service/antigravity_credits_overages_test.go
@@ -40,58 +40,50 @@ func TestClassifyAntigravity429(t *testing.T) {
 	})
 }

-func TestCanUseAntigravityCreditsOverages(t *testing.T) {
-	activeUntil := time.Now().Add(10 * time.Minute).UTC().Format(time.RFC3339)
-
-	t.Run("必须有运行态才可直接走 overages", func(t *testing.T) {
+func TestIsCreditsExhausted_UsesAICreditsKey(t *testing.T) {
+	t.Run("无 AICredits key 则积分可用", func(t *testing.T) {
 		account := &Account{
-			ID:          1,
-			Platform:    PlatformAntigravity,
-			Status:      StatusActive,
-			Schedulable: true,
+			ID:       1,
+			Platform: PlatformAntigravity,
 			Extra: map[string]any{
 				"allow_overages": true,
 			},
 		}
-		require.False(t, canUseAntigravityCreditsOverages(context.Background(), account, "claude-sonnet-4-5"))
+		require.False(t, account.isCreditsExhausted())
 	})

-	t.Run("运行态有效时允许使用 overages", func(t *testing.T) {
+	t.Run("AICredits key 生效则积分耗尽", func(t *testing.T) {
 		account := &Account{
-			ID:          2,
-			Platform:    PlatformAntigravity,
-			Status:      StatusActive,
-			Schedulable: true,
+			ID:       2,
+			Platform: PlatformAntigravity,
 			Extra: map[string]any{
 				"allow_overages": true,
-				antigravityCreditsOveragesKey: map[string]any{
-					"claude-sonnet-4-5": map[string]any{
-						"active_until": activeUntil,
+				modelRateLimitsKey: map[string]any{
+					creditsExhaustedKey: map[string]any{
+						"rate_limited_at":     time.Now().UTC().Format(time.RFC3339),
+						"rate_limit_reset_at": time.Now().Add(5 * time.Hour).UTC().Format(time.RFC3339),
 					},
 				},
 			},
 		}
-		require.True(t, canUseAntigravityCreditsOverages(context.Background(), account, "claude-sonnet-4-5"))
+		require.True(t, account.isCreditsExhausted())
 	})

-	t.Run("credits 耗尽后不可继续使用 overages", func(t *testing.T) {
+	t.Run("AICredits key 过期则积分可用", func(t *testing.T) {
 		account := &Account{
-			ID:          3,
-			Platform:    PlatformAntigravity,
-			Status:      StatusActive,
-			Schedulable: true,
+			ID:       3,
+			Platform: PlatformAntigravity,
 			Extra: map[string]any{
 				"allow_overages": true,
-				antigravityCreditsOveragesKey: map[string]any{
-					"claude-sonnet-4-5": map[string]any{
-						"active_until": activeUntil,
+				modelRateLimitsKey: map[string]any{
+					creditsExhaustedKey: map[string]any{
+						"rate_limited_at":     time.Now().Add(-6 * time.Hour).UTC().Format(time.RFC3339),
+						"rate_limit_reset_at": time.Now().Add(-1 * time.Hour).UTC().Format(time.RFC3339),
 					},
 				},
 			},
 		}
-		setCreditsExhausted(account.ID, time.Now().Add(time.Minute))
-		t.Cleanup(func() { clearCreditsExhausted(account.ID) })
-		require.False(t, canUseAntigravityCreditsOverages(context.Background(), account, "claude-sonnet-4-5"))
+		require.False(t, account.isCreditsExhausted())
 	})
 }

@@ -152,12 +144,6 @@ func TestHandleSmartRetry_QuotaExhausted_UsesCreditsAndStoresIndependentState(t
 	require.Len(t, upstream.requestBodies, 1)
 	require.Contains(t, string(upstream.requestBodies[0]), "enabledCreditTypes")
 	require.Empty(t, repo.modelRateLimitCalls, "overages 成功后不应写入普通 model_rate_limits")
-	require.Len(t, repo.extraUpdateCalls, 1)
-
-	state, ok := account.Extra[antigravityCreditsOveragesKey].(map[string]any)
-	require.True(t, ok)
-	_, exists := state["claude-sonnet-4-5"]
-	require.True(t, exists, "应使用最终映射模型写入独立 overages 运行态")
 }

 func TestHandleSmartRetry_RateLimited_DoesNotUseCredits(t *testing.T) {
@@ -221,7 +207,7 @@ func TestHandleSmartRetry_RateLimited_DoesNotUseCredits(t *testing.T) {
 	require.Empty(t, repo.modelRateLimitCalls)
 }

-func TestAntigravityRetryLoop_ActiveOverages_InjectsCreditsBody(t *testing.T) {
+func TestAntigravityRetryLoop_ModelRateLimited_InjectsCredits(t *testing.T) {
 	oldBaseURLs := append([]string(nil), antigravity.BaseURLs...)
 	oldAvailability := antigravity.DefaultURLAvailability
 	defer func() {
@@ -232,7 +218,6 @@ func TestAntigravityRetryLoop_ActiveOverages_InjectsCreditsBody(t *testing.T) {
 	antigravity.BaseURLs = []string{"https://ag-1.test"}
 	antigravity.DefaultURLAvailability = antigravity.NewURLAvailability(time.Minute)

-	activeUntil := time.Now().Add(10 * time.Minute).UTC().Format(time.RFC3339)
 	upstream := &queuedHTTPUpstreamStub{
 		responses: []*http.Response{
 			{
@@ -243,6 +228,7 @@ func TestAntigravityRetryLoop_ActiveOverages_InjectsCreditsBody(t *testing.T) {
 		},
 		errors: []error{nil},
 	}
+	// 模型已限流 + overages 启用 + 无 AICredits key → 应直接注入积分
 	account := &Account{
 		ID:          103,
 		Name:        "acc-103",
@@ -252,9 +238,10 @@ func TestAntigravityRetryLoop_ActiveOverages_InjectsCreditsBody(t *testing.T) {
 		Schedulable: true,
 		Extra: map[string]any{
 			"allow_overages": true,
-			antigravityCreditsOveragesKey: map[string]any{
+			modelRateLimitsKey: map[string]any{
 				"claude-sonnet-4-5": map[string]any{
-					"active_until": activeUntil,
+					"rate_limited_at":     time.Now().UTC().Format(time.RFC3339),
+					"rate_limit_reset_at": time.Now().Add(30 * time.Minute).UTC().Format(time.RFC3339),
 				},
 			},
 		},
@@ -281,7 +268,61 @@ func TestAntigravityRetryLoop_ActiveOverages_InjectsCreditsBody(t *testing.T) {
 	require.Contains(t, string(upstream.requestBodies[0]), "enabledCreditTypes")
 }

-func TestAntigravityRetryLoop_ActiveOverages_ExplicitCreditErrorMarksExhausted(t *testing.T) {
+func TestAntigravityRetryLoop_CreditsExhausted_DoesNotInject(t *testing.T) {
+	oldBaseURLs := append([]string(nil), antigravity.BaseURLs...)
+	oldAvailability := antigravity.DefaultURLAvailability
+	defer func() {
+		antigravity.BaseURLs = oldBaseURLs
+		antigravity.DefaultURLAvailability = oldAvailability
+	}()
+
+	antigravity.BaseURLs = []string{"https://ag-1.test"}
+	antigravity.DefaultURLAvailability = antigravity.NewURLAvailability(time.Minute)
+
+	// 模型限流 + overages 启用 + AICredits key 生效 → 不应注入积分，应切号
+	account := &Account{
+		ID:          104,
+		Name:        "acc-104",
+		Type:        AccountTypeOAuth,
+		Platform:    PlatformAntigravity,
+		Status:      StatusActive,
+		Schedulable: true,
+		Extra: map[string]any{
+			"allow_overages": true,
+			modelRateLimitsKey: map[string]any{
+				"claude-sonnet-4-5": map[string]any{
+					"rate_limited_at":     time.Now().UTC().Format(time.RFC3339),
+					"rate_limit_reset_at": time.Now().Add(30 * time.Minute).UTC().Format(time.RFC3339),
+				},
+				creditsExhaustedKey: map[string]any{
+					"rate_limited_at":     time.Now().UTC().Format(time.RFC3339),
+					"rate_limit_reset_at": time.Now().Add(5 * time.Hour).UTC().Format(time.RFC3339),
+				},
+			},
+		},
+	}
+
+	svc := &AntigravityGatewayService{}
+	_, err := svc.antigravityRetryLoop(antigravityRetryLoopParams{
+		ctx:            context.Background(),
+		prefix:         "[test]",
+		account:        account,
+		accessToken:    "token",
+		action:         "generateContent",
+		body:           []byte(`{"model":"claude-sonnet-4-5","request":{}}`),
+		requestedModel: "claude-sonnet-4-5",
+		handleError: func(ctx context.Context, prefix string, account *Account, statusCode int, headers http.Header, body []byte, requestedModel string, groupID int64, sessionHash string, isStickySession bool) *handleModelRateLimitResult {
+			return nil
+		},
+	})
+
+	// 模型限流 + 积分耗尽 → 应触发切号错误
+	require.Error(t, err)
+	var switchErr *AntigravityAccountSwitchError
+	require.ErrorAs(t, err, &switchErr)
+}
+
+func TestAntigravityRetryLoop_CreditErrorMarksExhausted(t *testing.T) {
 	oldBaseURLs := append([]string(nil), antigravity.BaseURLs...)
 	oldAvailability := antigravity.DefaultURLAvailability
 	defer func() {
@@ -292,8 +333,6 @@ func TestAntigravityRetryLoop_ActiveOverages_ExplicitCreditErrorMarksExhausted(t
 	antigravity.BaseURLs = []string{"https://ag-1.test"}
 	antigravity.DefaultURLAvailability = antigravity.NewURLAvailability(time.Minute)

-	accountID := int64(104)
-	activeUntil := time.Now().Add(10 * time.Minute).UTC().Format(time.RFC3339)
 	repo := &stubAntigravityAccountRepo{}
 	upstream := &queuedHTTPUpstreamStub{
 		responses: []*http.Response{
@@ -305,24 +344,24 @@ func TestAntigravityRetryLoop_ActiveOverages_ExplicitCreditErrorMarksExhausted(t
 		},
 		errors: []error{nil},
 	}
+	// 模型限流 + overages 启用 + 积分可用 → 注入积分但上游返回积分不足
 	account := &Account{
-		ID:          accountID,
-		Name:        "acc-104",
+		ID:          105,
+		Name:        "acc-105",
 		Type:        AccountTypeOAuth,
 		Platform:    PlatformAntigravity,
 		Status:      StatusActive,
 		Schedulable: true,
 		Extra: map[string]any{
 			"allow_overages": true,
-			antigravityCreditsOveragesKey: map[string]any{
+			modelRateLimitsKey: map[string]any{
 				"claude-sonnet-4-5": map[string]any{
-					"active_until": activeUntil,
+					"rate_limited_at":     time.Now().UTC().Format(time.RFC3339),
+					"rate_limit_reset_at": time.Now().Add(30 * time.Minute).UTC().Format(time.RFC3339),
 				},
 			},
 		},
 	}
-	clearCreditsExhausted(accountID)
-	t.Cleanup(func() { clearCreditsExhausted(accountID) })

 	svc := &AntigravityGatewayService{accountRepo: repo}
 	result, err := svc.antigravityRetryLoop(antigravityRetryLoopParams{
@@ -333,6 +372,7 @@ func TestAntigravityRetryLoop_ActiveOverages_ExplicitCreditErrorMarksExhausted(t
 		action:         "generateContent",
 		body:           []byte(`{"model":"claude-sonnet-4-5","request":{}}`),
 		httpUpstream:   upstream,
+		accountRepo:    repo,
 		requestedModel: "claude-sonnet-4-5",
 		handleError: func(ctx context.Context, prefix string, account *Account, statusCode int, headers http.Header, body []byte, requestedModel string, groupID int64, sessionHash string, isStickySession bool) *handleModelRateLimitResult {
 			return nil
@@ -341,6 +381,158 @@ func TestAntigravityRetryLoop_ActiveOverages_ExplicitCreditErrorMarksExhausted(t

 	require.NoError(t, err)
 	require.NotNil(t, result)
-	require.True(t, isCreditsExhausted(accountID))
-	require.Len(t, repo.extraUpdateCalls, 1, "应清理对应模型的 overages 运行态")
+	// 验证 AICredits key 已通过 SetModelRateLimit 写入数据库
+	require.Len(t, repo.modelRateLimitCalls, 1, "应通过 SetModelRateLimit 写入 AICredits key")
+	require.Equal(t, creditsExhaustedKey, repo.modelRateLimitCalls[0].modelKey)
+}
+
+func TestShouldMarkCreditsExhausted(t *testing.T) {
+	t.Run("reqErr 不为 nil 时不标记", func(t *testing.T) {
+		resp := &http.Response{StatusCode: http.StatusForbidden}
+		require.False(t, shouldMarkCreditsExhausted(resp, []byte(`{"error":"Insufficient credits"}`), io.ErrUnexpectedEOF))
+	})
+
+	t.Run("resp 为 nil 时不标记", func(t *testing.T) {
+		require.False(t, shouldMarkCreditsExhausted(nil, []byte(`{"error":"Insufficient credits"}`), nil))
+	})
+
+	t.Run("5xx 响应不标记", func(t *testing.T) {
+		resp := &http.Response{StatusCode: http.StatusInternalServerError}
+		require.False(t, shouldMarkCreditsExhausted(resp, []byte(`{"error":"Insufficient credits"}`), nil))
+	})
+
+	t.Run("408 RequestTimeout 不标记", func(t *testing.T) {
+		resp := &http.Response{StatusCode: http.StatusRequestTimeout}
+		require.False(t, shouldMarkCreditsExhausted(resp, []byte(`{"error":"Insufficient credits"}`), nil))
+	})
+
+	t.Run("URL 级限流不标记", func(t *testing.T) {
+		resp := &http.Response{StatusCode: http.StatusTooManyRequests}
+		body := []byte(`{"error":{"message":"Resource has been exhausted"}}`)
+		require.False(t, shouldMarkCreditsExhausted(resp, body, nil))
+	})
+
+	t.Run("结构化限流不标记", 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("含 credits 关键词时标记", func(t *testing.T) {
+		resp := &http.Response{StatusCode: http.StatusForbidden}
+		for _, keyword := range []string{
+			"Insufficient GOOGLE_ONE_AI credits",
+			"insufficient credit balance",
+			"not enough credits for this request",
+			"Credits exhausted",
+			"minimumCreditAmountForUsage requirement not met",
+		} {
+			body := []byte(`{"error":{"message":"` + keyword + `"}}`)
+			require.True(t, shouldMarkCreditsExhausted(resp, body, nil), "should mark for keyword: %s", keyword)
+		}
+	})
+
+	t.Run("无 credits 关键词时不标记", func(t *testing.T) {
+		resp := &http.Response{StatusCode: http.StatusForbidden}
+		body := []byte(`{"error":{"message":"permission denied"}}`)
+		require.False(t, shouldMarkCreditsExhausted(resp, body, nil))
+	})
+}
+
+func TestInjectEnabledCreditTypes(t *testing.T) {
+	t.Run("正常 JSON 注入成功", func(t *testing.T) {
+		body := []byte(`{"model":"claude-sonnet-4-5","request":{}}`)
+		result := injectEnabledCreditTypes(body)
+		require.NotNil(t, result)
+		require.Contains(t, string(result), `"enabledCreditTypes"`)
+		require.Contains(t, string(result), `GOOGLE_ONE_AI`)
+	})
+
+	t.Run("非法 JSON 返回 nil", func(t *testing.T) {
+		require.Nil(t, injectEnabledCreditTypes([]byte(`not json`)))
+	})
+
+	t.Run("空 body 返回 nil", func(t *testing.T) {
+		require.Nil(t, injectEnabledCreditTypes([]byte{}))
+	})
+
+	t.Run("已有 enabledCreditTypes 会被覆盖", func(t *testing.T) {
+		body := []byte(`{"enabledCreditTypes":["OLD"],"model":"test"}`)
+		result := injectEnabledCreditTypes(body)
+		require.NotNil(t, result)
+		require.Contains(t, string(result), `GOOGLE_ONE_AI`)
+		require.NotContains(t, string(result), `OLD`)
+	})
+}
+
+func TestClearCreditsExhausted(t *testing.T) {
+	t.Run("account 为 nil 不操作", func(t *testing.T) {
+		repo := &stubAntigravityAccountRepo{}
+		svc := &AntigravityGatewayService{accountRepo: repo}
+		svc.clearCreditsExhausted(context.Background(), nil)
+		require.Empty(t, repo.extraUpdateCalls)
+	})
+
+	t.Run("Extra 为 nil 不操作", func(t *testing.T) {
+		repo := &stubAntigravityAccountRepo{}
+		svc := &AntigravityGatewayService{accountRepo: repo}
+		svc.clearCreditsExhausted(context.Background(), &Account{ID: 1})
+		require.Empty(t, repo.extraUpdateCalls)
+	})
+
+	t.Run("无 modelRateLimitsKey 不操作", func(t *testing.T) {
+		repo := &stubAntigravityAccountRepo{}
+		svc := &AntigravityGatewayService{accountRepo: repo}
+		svc.clearCreditsExhausted(context.Background(), &Account{
+			ID:    1,
+			Extra: map[string]any{"some_key": "value"},
+		})
+		require.Empty(t, repo.extraUpdateCalls)
+	})
+
+	t.Run("无 AICredits key 不操作", func(t *testing.T) {
+		repo := &stubAntigravityAccountRepo{}
+		svc := &AntigravityGatewayService{accountRepo: repo}
+		svc.clearCreditsExhausted(context.Background(), &Account{
+			ID: 1,
+			Extra: map[string]any{
+				modelRateLimitsKey: map[string]any{
+					"claude-sonnet-4-5": map[string]any{
+						"rate_limited_at":     "2026-03-15T00:00:00Z",
+						"rate_limit_reset_at": "2099-03-15T00:00:00Z",
+					},
+				},
+			},
+		})
+		require.Empty(t, repo.extraUpdateCalls)
+	})
+
+	t.Run("有 AICredits key 时删除并调用 UpdateExtra", func(t *testing.T) {
+		repo := &stubAntigravityAccountRepo{}
+		svc := &AntigravityGatewayService{accountRepo: repo}
+		account := &Account{
+			ID: 1,
+			Extra: map[string]any{
+				modelRateLimitsKey: map[string]any{
+					"claude-sonnet-4-5": map[string]any{
+						"rate_limited_at":     "2026-03-15T00:00:00Z",
+						"rate_limit_reset_at": "2099-03-15T00:00:00Z",
+					},
+					creditsExhaustedKey: map[string]any{
+						"rate_limited_at":     "2026-03-15T00:00:00Z",
+						"rate_limit_reset_at": time.Now().Add(5 * time.Hour).UTC().Format(time.RFC3339),
+					},
+				},
+			},
+		}
+		svc.clearCreditsExhausted(context.Background(), account)
+		require.Len(t, repo.extraUpdateCalls, 1)
+		// AICredits key 应被删除
+		rawLimits := account.Extra[modelRateLimitsKey].(map[string]any)
+		_, exists := rawLimits[creditsExhaustedKey]
+		require.False(t, exists, "AICredits key 应被删除")
+		// 普通模型限流应保留
+		_, exists = rawLimits["claude-sonnet-4-5"]
+		require.True(t, exists, "普通模型限流应保留")
+	})
 }
--- a/backend/internal/service/antigravity_gateway_service.go
+++ b/backend/internal/service/antigravity_gateway_service.go
@@ -201,7 +201,7 @@ func (s *AntigravityGatewayService) handleSmartRetry(p antigravityRetryLoopParam
 	if resp.StatusCode == http.StatusTooManyRequests &&
 		category == antigravity429QuotaExhausted &&
 		p.account.IsOveragesEnabled() &&
-		!isCreditsExhausted(p.account.ID) {
+		!p.account.isCreditsExhausted() {
 		result := s.attemptCreditsOveragesRetry(p, baseURL, modelName, waitDuration, resp.StatusCode, respBody)
 		if result.handled && result.resp != nil {
 			return &smartRetryResult{
@@ -552,13 +552,15 @@ func (s *AntigravityGatewayService) handleSingleAccountRetryInPlace(

 // antigravityRetryLoop 执行带 URL fallback 的重试循环
 func (s *AntigravityGatewayService) antigravityRetryLoop(p antigravityRetryLoopParams) (*antigravityRetryLoopResult, error) {
-	// 预检查：如果模型已进入 overages 运行态，则直接注入 AI Credits。
-	overagesActive := false
-	if p.requestedModel != "" && canUseAntigravityCreditsOverages(p.ctx, p.account, p.requestedModel) {
+	// 预检查：模型限流 + overages 启用 + 积分未耗尽 → 直接注入 AI Credits
+	overagesInjected := false
+	if p.requestedModel != "" && p.account.Platform == PlatformAntigravity &&
+		p.account.IsOveragesEnabled() && !p.account.isCreditsExhausted() &&
+		p.account.isModelRateLimitedWithContext(p.ctx, p.requestedModel) {
 		if creditsBody := injectEnabledCreditTypes(p.body); creditsBody != nil {
 			p.body = creditsBody
-			overagesActive = true
-			logger.LegacyPrintf("service.antigravity_gateway", "%s pre_check: credit_overages_active model=%s account=%d (injecting enabledCreditTypes)",
+			overagesInjected = true
+			logger.LegacyPrintf("service.antigravity_gateway", "%s pre_check: model_rate_limited_credits_inject model=%s account=%d (injecting enabledCreditTypes)",
 				p.prefix, p.requestedModel, p.account.ID)
 		}
 	}
@@ -566,9 +568,9 @@ func (s *AntigravityGatewayService) antigravityRetryLoop(p antigravityRetryLoopP
 	// 预检查：如果账号已限流，直接返回切换信号
 	if p.requestedModel != "" {
 		if remaining := p.account.GetRateLimitRemainingTimeWithContext(p.ctx, p.requestedModel); remaining > 0 {
-			// 进入 overages 运行态的模型不再受普通模型限流预检查阻断。
-			if overagesActive {
-				logger.LegacyPrintf("service.antigravity_gateway", "%s pre_check: credit_overages_ignore_rate_limit remaining=%v model=%s account=%d",
+			// 已注入积分的请求不再受普通模型限流预检查阻断。
+			if overagesInjected {
+				logger.LegacyPrintf("service.antigravity_gateway", "%s pre_check: credits_injected_ignore_rate_limit remaining=%v model=%s account=%d",
 					p.prefix, remaining.Truncate(time.Millisecond), p.requestedModel, p.account.ID)
 			} else if isSingleAccountRetry(p.ctx) {
 				// 单账号 503 退避重试模式：跳过限流预检查，直接发请求。
@@ -666,9 +668,9 @@ urlFallbackLoop:
 				respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
 				_ = resp.Body.Close()

-				if overagesActive && shouldMarkCreditsExhausted(resp, respBody, nil) {
+				if overagesInjected && shouldMarkCreditsExhausted(resp, respBody, nil) {
 					modelKey := resolveCreditsOveragesModelKey(p.ctx, p.account, "", p.requestedModel)
-					s.handleCreditsRetryFailure(p.prefix, modelKey, p.account, 0, &http.Response{
+					s.handleCreditsRetryFailure(p.ctx, p.prefix, modelKey, p.account, &http.Response{
 						StatusCode: resp.StatusCode,
 						Header:     resp.Header.Clone(),
 						Body:       io.NopCloser(bytes.NewReader(respBody)),
@@ -1717,7 +1719,7 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 	var clientDisconnect bool
 	if claudeReq.Stream {
 		// 客户端要求流式，直接透传转换
-		streamRes, err := s.handleClaudeStreamingResponse(c, resp, startTime, originalModel)
+		streamRes, err := s.handleClaudeStreamingResponse(c, resp, startTime, originalModel, account.ID)
 		if err != nil {
 			logger.LegacyPrintf("service.antigravity_gateway", "%s status=stream_error error=%v", prefix, err)
 			return nil, err
@@ -1727,7 +1729,7 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 		clientDisconnect = streamRes.clientDisconnect
 	} else {
 		// 客户端要求非流式，收集流式响应后转换返回
-		streamRes, err := s.handleClaudeStreamToNonStreaming(c, resp, startTime, originalModel)
+		streamRes, err := s.handleClaudeStreamToNonStreaming(c, resp, startTime, originalModel, account.ID)
 		if err != nil {
 			logger.LegacyPrintf("service.antigravity_gateway", "%s status=stream_collect_error error=%v", prefix, err)
 			return nil, err
@@ -1736,6 +1738,9 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 		firstTokenMs = streamRes.firstTokenMs
 	}

+	// Claude Max cache billing: 同步 ForwardResult.Usage 与客户端响应体一致
+	applyClaudeMaxCacheBillingPolicyToUsage(usage, parsedRequestFromGinContext(c), claudeMaxGroupFromGinContext(c), originalModel, account.ID)
+
 	return &ForwardResult{
 		RequestID:        requestID,
 		Usage:            *usage,
@@ -3639,7 +3644,7 @@ func (s *AntigravityGatewayService) writeGoogleError(c *gin.Context, status int,

 // handleClaudeStreamToNonStreaming 收集上游流式响应，转换为 Claude 非流式格式返回
 // 用于处理客户端非流式请求但上游只支持流式的情况
-func (s *AntigravityGatewayService) handleClaudeStreamToNonStreaming(c *gin.Context, resp *http.Response, startTime time.Time, originalModel string) (*antigravityStreamResult, error) {
+func (s *AntigravityGatewayService) handleClaudeStreamToNonStreaming(c *gin.Context, resp *http.Response, startTime time.Time, originalModel string, accountID int64) (*antigravityStreamResult, error) {
 	scanner := bufio.NewScanner(resp.Body)
 	maxLineSize := defaultMaxLineSize
 	if s.settingService.cfg != nil && s.settingService.cfg.Gateway.MaxLineSize > 0 {
@@ -3797,6 +3802,9 @@ returnResponse:
 		return nil, s.writeClaudeError(c, http.StatusBadGateway, "upstream_error", "Failed to parse upstream response")
 	}

+	// Claude Max cache billing simulation (non-streaming)
+	claudeResp = applyClaudeMaxNonStreamingRewrite(c, claudeResp, agUsage, originalModel, accountID)
+
 	c.Data(http.StatusOK, "application/json", claudeResp)

 	// 转换为 service.ClaudeUsage
@@ -3811,7 +3819,7 @@ returnResponse:
 }

 // handleClaudeStreamingResponse 处理 Claude 流式响应（Gemini SSE → Claude SSE 转换）
-func (s *AntigravityGatewayService) handleClaudeStreamingResponse(c *gin.Context, resp *http.Response, startTime time.Time, originalModel string) (*antigravityStreamResult, error) {
+func (s *AntigravityGatewayService) handleClaudeStreamingResponse(c *gin.Context, resp *http.Response, startTime time.Time, originalModel string, accountID int64) (*antigravityStreamResult, error) {
 	c.Header("Content-Type", "text/event-stream")
 	c.Header("Cache-Control", "no-cache")
 	c.Header("Connection", "keep-alive")
@@ -3824,6 +3832,8 @@ func (s *AntigravityGatewayService) handleClaudeStreamingResponse(c *gin.Context
 	}

 	processor := antigravity.NewStreamingProcessor(originalModel)
+	setupClaudeMaxStreamingHook(c, processor, originalModel, accountID)
+
 	var firstTokenMs *int
 	// 使用 Scanner 并限制单行大小，避免 ReadString 无上限导致 OOM
 	scanner := bufio.NewScanner(resp.Body)
--- a/backend/internal/service/antigravity_quota_scope.go
+++ b/backend/internal/service/antigravity_quota_scope.go
@@ -32,6 +32,10 @@ func (a *Account) IsSchedulableForModelWithContext(ctx context.Context, requeste
 		return false
 	}
 	if a.isModelRateLimitedWithContext(ctx, requestedModel) {
+		// Antigravity + overages 启用 + 积分未耗尽 → 放行（有积分可用）
+		if a.Platform == PlatformAntigravity && a.IsOveragesEnabled() && !a.isCreditsExhausted() {
+			return true
+		}
 		return false
 	}
 	return true
--- a/backend/internal/service/ratelimit_service.go
+++ b/backend/internal/service/ratelimit_service.go
@@ -1100,9 +1100,6 @@ func (s *RateLimitService) ClearRateLimit(ctx context.Context, accountID int64)
 	if err := s.accountRepo.ClearModelRateLimits(ctx, accountID); err != nil {
 		return err
 	}
-	if err := clearAntigravityCreditsOveragesState(ctx, s.accountRepo, accountID); err != nil {
-		return err
-	}
 	// 清除限流时一并清理临时不可调度状态，避免周限/窗口重置后仍被本地临时状态阻断。
 	if err := s.accountRepo.ClearTempUnschedulable(ctx, accountID); err != nil {
 		return err
@@ -1112,7 +1109,6 @@ func (s *RateLimitService) ClearRateLimit(ctx context.Context, accountID int64)
 			slog.Warn("temp_unsched_cache_delete_failed", "account_id", accountID, "error", err)
 		}
 	}
-	clearCreditsExhausted(accountID)
 	return nil
 }

@@ -1179,8 +1175,7 @@ func hasRecoverableRuntimeState(account *Account) bool {
 		return false
 	}
 	return hasNonEmptyMapValue(account.Extra, "model_rate_limits") ||
-		hasNonEmptyMapValue(account.Extra, "antigravity_quota_scopes") ||
-		hasNonEmptyMapValue(account.Extra, antigravityCreditsOveragesKey)
+		hasNonEmptyMapValue(account.Extra, "antigravity_quota_scopes")
 }

 func hasNonEmptyMapValue(extra map[string]any, key string) bool {