diff --git a/backend/internal/config/config.go b/backend/internal/config/config.go
index 2fefe820c9..748b751463 100644
--- a/backend/internal/config/config.go
+++ b/backend/internal/config/config.go
@@ -643,6 +643,10 @@ type ProxyProbeConfig struct {
 
 type BillingConfig struct {
 	CircuitBreaker CircuitBreakerConfig `mapstructure:"circuit_breaker"`
+	// MinimumBalanceReserve is the conservative preflight floor for balance billing.
+	// Requests in balance mode are rejected when the cached balance is below this
+	// amount, even if it is still positive. Set to 0 to keep the legacy balance > 0 gate.
+	MinimumBalanceReserve float64 `mapstructure:"minimum_balance_reserve"`
 	// UserPlatformQuotaCacheTTLSeconds 用户 × 平台 quota 缓存 TTL（秒），默认 86400=1天，覆盖典型 daily 窗口。
 	// 消费点：
 	//   - billing_cache_service.cacheWriteWorker 异步累加
@@ -1615,6 +1619,7 @@ func setDefaults() {
 	viper.SetDefault("billing.circuit_breaker.failure_threshold", 5)
 	viper.SetDefault("billing.circuit_breaker.reset_timeout_seconds", 30)
 	viper.SetDefault("billing.circuit_breaker.half_open_requests", 3)
+	viper.SetDefault("billing.minimum_balance_reserve", 0.000001)
 	viper.SetDefault("billing.user_platform_quota_cache_ttl_seconds", 86400)
 	viper.SetDefault("billing.user_platform_quota_sentinel_ttl_seconds", 3600)
 
@@ -2277,6 +2282,9 @@ func (c *Config) Validate() error {
 			return fmt.Errorf("billing.circuit_breaker.half_open_requests must be positive")
 		}
 	}
+	if c.Billing.MinimumBalanceReserve < 0 {
+		return fmt.Errorf("billing.minimum_balance_reserve must be non-negative")
+	}
 	if c.Database.MaxOpenConns <= 0 {
 		return fmt.Errorf("database.max_open_conns must be positive")
 	}
diff --git a/backend/internal/config/config_test.go b/backend/internal/config/config_test.go
index ecdfaac4d1..ec0df0e3bd 100644
--- a/backend/internal/config/config_test.go
+++ b/backend/internal/config/config_test.go
@@ -1157,6 +1157,11 @@ func TestValidateConfigErrors(t *testing.T) {
 			mutate:  func(c *Config) { c.Billing.CircuitBreaker.HalfOpenRequests = 0 },
 			wantErr: "billing.circuit_breaker.half_open_requests",
 		},
+		{
+			name:    "billing minimum balance reserve",
+			mutate:  func(c *Config) { c.Billing.MinimumBalanceReserve = -0.01 },
+			wantErr: "billing.minimum_balance_reserve",
+		},
 		{
 			name:    "database max open conns",
 			mutate:  func(c *Config) { c.Database.MaxOpenConns = 0 },
diff --git a/backend/internal/repository/usage_billing_repo.go b/backend/internal/repository/usage_billing_repo.go
index 62f48b58f4..91ac536eee 100644
--- a/backend/internal/repository/usage_billing_repo.go
+++ b/backend/internal/repository/usage_billing_repo.go
@@ -113,11 +113,12 @@ func (r *usageBillingRepository) applyUsageBillingEffects(ctx context.Context, t
 	}
 
 	if cmd.BalanceCost > 0 {
-		newBalance, err := deductUsageBillingBalance(ctx, tx, cmd.UserID, cmd.BalanceCost)
+		newBalance, sufficient, err := deductUsageBillingBalance(ctx, tx, cmd.UserID, cmd.BalanceCost)
 		if err != nil {
 			return err
 		}
 		result.NewBalance = &newBalance
+		result.BalanceOverdrafted = !sufficient
 	}
 
 	if cmd.APIKeyQuotaCost > 0 {
@@ -173,9 +174,23 @@ func incrementUsageBillingSubscription(ctx context.Context, tx *sql.Tx, subscrip
 	return service.ErrSubscriptionNotFound
 }
 
-func deductUsageBillingBalance(ctx context.Context, tx *sql.Tx, userID int64, amount float64) (float64, error) {
+func deductUsageBillingBalance(ctx context.Context, tx *sql.Tx, userID int64, amount float64) (float64, bool, error) {
 	var newBalance float64
 	err := tx.QueryRowContext(ctx, `
+		UPDATE users
+		SET balance = balance - $1,
+			updated_at = NOW()
+		WHERE id = $2 AND deleted_at IS NULL AND balance >= $1
+		RETURNING balance
+	`, amount, userID).Scan(&newBalance)
+	if err == nil {
+		return newBalance, true, nil
+	}
+	if !errors.Is(err, sql.ErrNoRows) {
+		return 0, false, err
+	}
+
+	err = tx.QueryRowContext(ctx, `
 		UPDATE users
 		SET balance = balance - $1,
 			updated_at = NOW()
@@ -183,12 +198,12 @@ func deductUsageBillingBalance(ctx context.Context, tx *sql.Tx, userID int64, am
 		RETURNING balance
 	`, amount, userID).Scan(&newBalance)
 	if errors.Is(err, sql.ErrNoRows) {
-		return 0, service.ErrUserNotFound
+		return 0, false, service.ErrUserNotFound
 	}
 	if err != nil {
-		return 0, err
+		return 0, false, err
 	}
-	return newBalance, nil
+	return newBalance, false, nil
 }
 
 func incrementUsageBillingAPIKeyQuota(ctx context.Context, tx *sql.Tx, apiKeyID int64, amount float64) (bool, error) {
diff --git a/backend/internal/repository/usage_billing_repo_unit_test.go b/backend/internal/repository/usage_billing_repo_unit_test.go
new file mode 100644
index 0000000000..8ed5530a8f
--- /dev/null
+++ b/backend/internal/repository/usage_billing_repo_unit_test.go
@@ -0,0 +1,119 @@
+//go:build unit
+
+package repository
+
+import (
+	"context"
+	"database/sql"
+	"testing"
+
+	"github.com/DATA-DOG/go-sqlmock"
+	"github.com/stretchr/testify/require"
+
+	"github.com/Wei-Shaw/sub2api/internal/service"
+)
+
+const (
+	conditionalBalanceDeductSQL = `(?s)UPDATE users\s+SET balance = balance - \$1,\s+updated_at = NOW\(\)\s+WHERE id = \$2 AND deleted_at IS NULL AND balance >= \$1\s+RETURNING balance`
+	overdraftBalanceDeductSQL   = `(?s)UPDATE users\s+SET balance = balance - \$1,\s+updated_at = NOW\(\)\s+WHERE id = \$2 AND deleted_at IS NULL\s+RETURNING balance`
+)
+
+func TestDeductUsageBillingBalance_UsesSufficientBalanceGuard(t *testing.T) {
+	ctx := context.Background()
+	db, mock, err := sqlmock.New()
+	require.NoError(t, err)
+	defer func() { _ = db.Close() }()
+
+	mock.ExpectBegin()
+	tx, err := db.BeginTx(ctx, nil)
+	require.NoError(t, err)
+	mock.ExpectQuery(conditionalBalanceDeductSQL).
+		WithArgs(2.5, int64(42)).
+		WillReturnRows(sqlmock.NewRows([]string{"balance"}).AddRow(7.5))
+	mock.ExpectCommit()
+
+	newBalance, sufficient, err := deductUsageBillingBalance(ctx, tx, 42, 2.5)
+	require.NoError(t, err)
+	require.True(t, sufficient)
+	require.InDelta(t, 7.5, newBalance, 0.000001)
+	require.NoError(t, tx.Commit())
+	require.NoError(t, mock.ExpectationsWereMet())
+}
+
+func TestDeductUsageBillingBalance_RecordsOverdraftWhenGuardMisses(t *testing.T) {
+	ctx := context.Background()
+	db, mock, err := sqlmock.New()
+	require.NoError(t, err)
+	defer func() { _ = db.Close() }()
+
+	mock.ExpectBegin()
+	tx, err := db.BeginTx(ctx, nil)
+	require.NoError(t, err)
+	mock.ExpectQuery(conditionalBalanceDeductSQL).
+		WithArgs(10.0, int64(42)).
+		WillReturnError(sql.ErrNoRows)
+	mock.ExpectQuery(overdraftBalanceDeductSQL).
+		WithArgs(10.0, int64(42)).
+		WillReturnRows(sqlmock.NewRows([]string{"balance"}).AddRow(-5.0))
+	mock.ExpectCommit()
+
+	newBalance, sufficient, err := deductUsageBillingBalance(ctx, tx, 42, 10)
+	require.NoError(t, err)
+	require.False(t, sufficient)
+	require.InDelta(t, -5.0, newBalance, 0.000001)
+	require.NoError(t, tx.Commit())
+	require.NoError(t, mock.ExpectationsWereMet())
+}
+
+func TestApplyUsageBillingEffects_FlagsBalanceOverdraft(t *testing.T) {
+	ctx := context.Background()
+	db, mock, err := sqlmock.New()
+	require.NoError(t, err)
+	defer func() { _ = db.Close() }()
+
+	mock.ExpectBegin()
+	tx, err := db.BeginTx(ctx, nil)
+	require.NoError(t, err)
+	mock.ExpectQuery(conditionalBalanceDeductSQL).
+		WithArgs(10.0, int64(42)).
+		WillReturnError(sql.ErrNoRows)
+	mock.ExpectQuery(overdraftBalanceDeductSQL).
+		WithArgs(10.0, int64(42)).
+		WillReturnRows(sqlmock.NewRows([]string{"balance"}).AddRow(-5.0))
+	mock.ExpectCommit()
+
+	result := &service.UsageBillingApplyResult{Applied: true}
+	err = (&usageBillingRepository{}).applyUsageBillingEffects(ctx, tx, &service.UsageBillingCommand{
+		UserID:      42,
+		BalanceCost: 10,
+	}, result)
+	require.NoError(t, err)
+	require.NotNil(t, result.NewBalance)
+	require.InDelta(t, -5.0, *result.NewBalance, 0.000001)
+	require.True(t, result.BalanceOverdrafted)
+	require.NoError(t, tx.Commit())
+	require.NoError(t, mock.ExpectationsWereMet())
+}
+
+func TestDeductUsageBillingBalance_ReturnsUserNotFoundWhenNoUserUpdated(t *testing.T) {
+	ctx := context.Background()
+	db, mock, err := sqlmock.New()
+	require.NoError(t, err)
+	defer func() { _ = db.Close() }()
+
+	mock.ExpectBegin()
+	tx, err := db.BeginTx(ctx, nil)
+	require.NoError(t, err)
+	mock.ExpectQuery(conditionalBalanceDeductSQL).
+		WithArgs(10.0, int64(42)).
+		WillReturnError(sql.ErrNoRows)
+	mock.ExpectQuery(overdraftBalanceDeductSQL).
+		WithArgs(10.0, int64(42)).
+		WillReturnError(sql.ErrNoRows)
+	mock.ExpectRollback()
+
+	_, _, err = deductUsageBillingBalance(ctx, tx, 42, 10)
+	require.ErrorIs(t, err, service.ErrUserNotFound)
+	require.NoError(t, tx.Rollback())
+	require.NoError(t, mock.ExpectationsWereMet())
+}
diff --git a/backend/internal/repository/user_repo.go b/backend/internal/repository/user_repo.go
index 98122c0bea..3ac8dcfbf8 100644
--- a/backend/internal/repository/user_repo.go
+++ b/backend/internal/repository/user_repo.go
@@ -757,6 +757,17 @@ func (r *userRepository) UpdateBalance(ctx context.Context, id int64, amount flo
 func (r *userRepository) DeductBalance(ctx context.Context, id int64, amount float64) error {
 	client := clientFromContext(ctx, r.client)
 	n, err := client.User.Update().
+		Where(dbuser.IDEQ(id), dbuser.BalanceGTE(amount)).
+		AddBalance(-amount).
+		Save(ctx)
+	if err != nil {
+		return err
+	}
+	if n > 0 {
+		return nil
+	}
+
+	n, err = client.User.Update().
 		Where(dbuser.IDEQ(id)).
 		AddBalance(-amount).
 		Save(ctx)
diff --git a/backend/internal/service/billing_cache_service.go b/backend/internal/service/billing_cache_service.go
index b734fab13e..bfdb4c20bc 100644
--- a/backend/internal/service/billing_cache_service.go
+++ b/backend/internal/service/billing_cache_service.go
@@ -833,6 +833,21 @@ func (s *BillingCacheService) checkRPM(ctx context.Context, user *User, group *G
 	return nil
 }
 
+func (s *BillingCacheService) minimumBalanceReserve() float64 {
+	if s == nil || s.cfg == nil || s.cfg.Billing.MinimumBalanceReserve <= 0 {
+		return 0
+	}
+	return s.cfg.Billing.MinimumBalanceReserve
+}
+
+func (s *BillingCacheService) balanceBelowEligibilityThreshold(balance float64) bool {
+	if balance <= 0 {
+		return true
+	}
+	minimumReserve := s.minimumBalanceReserve()
+	return minimumReserve > 0 && balance < minimumReserve
+}
+
 // checkBalanceEligibility 检查余额模式资格
 func (s *BillingCacheService) checkBalanceEligibility(ctx context.Context, userID int64) error {
 	balance, err := s.GetUserBalance(ctx, userID)
@@ -847,7 +862,7 @@ func (s *BillingCacheService) checkBalanceEligibility(ctx context.Context, userI
 		s.circuitBreaker.OnSuccess()
 	}
 
-	if balance <= 0 {
+	if s.balanceBelowEligibilityThreshold(balance) {
 		return ErrInsufficientBalance
 	}
 
diff --git a/backend/internal/service/billing_cache_service_balance_test.go b/backend/internal/service/billing_cache_service_balance_test.go
new file mode 100644
index 0000000000..06deb7010c
--- /dev/null
+++ b/backend/internal/service/billing_cache_service_balance_test.go
@@ -0,0 +1,128 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"errors"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/stretchr/testify/require"
+)
+
+type balanceEligibilityCacheStub struct {
+	billingCacheWorkerStub
+
+	balance                  float64
+	cacheMissAfterInvalidate bool
+	invalidated              atomic.Bool
+	deductCalls              atomic.Int64
+	invalidateCalls          atomic.Int64
+}
+
+func (s *balanceEligibilityCacheStub) GetUserBalance(context.Context, int64) (float64, error) {
+	if s.cacheMissAfterInvalidate && s.invalidated.Load() {
+		return 0, errors.New("cache miss")
+	}
+	return s.balance, nil
+}
+
+func (s *balanceEligibilityCacheStub) DeductUserBalance(context.Context, int64, float64) error {
+	s.deductCalls.Add(1)
+	return nil
+}
+
+func (s *balanceEligibilityCacheStub) InvalidateUserBalance(context.Context, int64) error {
+	s.invalidateCalls.Add(1)
+	s.invalidated.Store(true)
+	return nil
+}
+
+func TestCheckBillingEligibility_RejectsBalanceBelowMinimumReserve(t *testing.T) {
+	cache := &balanceEligibilityCacheStub{balance: 0.005}
+	cfg := &config.Config{}
+	cfg.Billing.MinimumBalanceReserve = 0.01
+	svc := NewBillingCacheService(cache, nil, nil, nil, nil, nil, cfg, nil)
+	t.Cleanup(svc.Stop)
+
+	err := svc.CheckBillingEligibility(context.Background(), &User{ID: 1}, nil, nil, nil, "")
+	require.ErrorIs(t, err, ErrInsufficientBalance)
+}
+
+func TestCheckBillingEligibility_AllowsBalanceAtMinimumReserve(t *testing.T) {
+	cache := &balanceEligibilityCacheStub{balance: 0.01}
+	cfg := &config.Config{}
+	cfg.Billing.MinimumBalanceReserve = 0.01
+	svc := NewBillingCacheService(cache, nil, nil, nil, nil, nil, cfg, nil)
+	t.Cleanup(svc.Stop)
+
+	err := svc.CheckBillingEligibility(context.Background(), &User{ID: 1}, nil, nil, nil, "")
+	require.NoError(t, err)
+}
+
+func TestSyncBalanceCacheAfterDeduction_InvalidatesExhaustedBalance(t *testing.T) {
+	cache := &balanceEligibilityCacheStub{
+		balance:                  0.50,
+		cacheMissAfterInvalidate: true,
+	}
+	userRepo := &balanceLoadUserRepoStub{balance: -0.25}
+	cfg := &config.Config{}
+	cfg.Billing.MinimumBalanceReserve = 0.01
+	svc := NewBillingCacheService(cache, userRepo, nil, nil, nil, nil, cfg, nil)
+	t.Cleanup(svc.Stop)
+
+	newBalance := -0.25
+	syncBalanceCacheAfterDeduction(context.Background(), &postUsageBillingParams{
+		Cost: &CostBreakdown{ActualCost: 0.75},
+		User: &User{ID: 1},
+	}, &billingDeps{billingCacheService: svc}, &UsageBillingApplyResult{
+		NewBalance:         &newBalance,
+		BalanceOverdrafted: true,
+	})
+
+	require.Equal(t, int64(1), cache.invalidateCalls.Load())
+	require.Equal(t, int64(0), cache.deductCalls.Load())
+
+	err := svc.CheckBillingEligibility(context.Background(), &User{ID: 1}, nil, nil, nil, "")
+	require.ErrorIs(t, err, ErrInsufficientBalance)
+	require.Equal(t, int64(1), userRepo.calls.Load())
+}
+
+func TestSyncBalanceCacheAfterDeduction_InvalidatesWhenBalanceFallsBelowReserve(t *testing.T) {
+	cache := &balanceEligibilityCacheStub{balance: 0.50}
+	cfg := &config.Config{}
+	cfg.Billing.MinimumBalanceReserve = 0.01
+	svc := NewBillingCacheService(cache, nil, nil, nil, nil, nil, cfg, nil)
+	t.Cleanup(svc.Stop)
+
+	newBalance := 0.005
+	syncBalanceCacheAfterDeduction(context.Background(), &postUsageBillingParams{
+		Cost: &CostBreakdown{ActualCost: 0.495},
+		User: &User{ID: 1},
+	}, &billingDeps{billingCacheService: svc}, &UsageBillingApplyResult{NewBalance: &newBalance})
+
+	require.Equal(t, int64(1), cache.invalidateCalls.Load())
+	require.Equal(t, int64(0), cache.deductCalls.Load())
+}
+
+func TestSyncBalanceCacheAfterDeduction_QueuesDeductWhenBalanceStillEligible(t *testing.T) {
+	cache := &balanceEligibilityCacheStub{balance: 1}
+	cfg := &config.Config{}
+	cfg.Billing.MinimumBalanceReserve = 0.01
+	svc := NewBillingCacheService(cache, nil, nil, nil, nil, nil, cfg, nil)
+	t.Cleanup(svc.Stop)
+
+	newBalance := 0.75
+	syncBalanceCacheAfterDeduction(context.Background(), &postUsageBillingParams{
+		Cost: &CostBreakdown{ActualCost: 0.25},
+		User: &User{ID: 1},
+	}, &billingDeps{billingCacheService: svc}, &UsageBillingApplyResult{NewBalance: &newBalance})
+
+	require.Equal(t, int64(0), cache.invalidateCalls.Load())
+	require.Eventually(t, func() bool {
+		return cache.deductCalls.Load() == 1
+	}, 2*time.Second, 10*time.Millisecond)
+}
diff --git a/backend/internal/service/gateway_service.go b/backend/internal/service/gateway_service.go
index cfbc5dec48..d9f3ef93b6 100644
--- a/backend/internal/service/gateway_service.go
+++ b/backend/internal/service/gateway_service.go
@@ -8913,6 +8913,10 @@ func postUsageBilling(ctx context.Context, p *postUsageBillingParams, deps *bill
 		if cost.ActualCost > 0 {
 			if err := deps.userRepo.DeductBalance(billingCtx, p.User.ID, cost.ActualCost); err != nil {
 				slog.Error("deduct balance failed", "user_id", p.User.ID, "error", err)
+			} else if deps.billingCacheService != nil {
+				if err := deps.billingCacheService.InvalidateUserBalance(billingCtx, p.User.ID); err != nil {
+					slog.Warn("invalidate balance cache after legacy deduction failed", "user_id", p.User.ID, "error", err)
+				}
 			}
 		}
 	}
@@ -9093,7 +9097,7 @@ func finalizePostUsageBilling(ctx context.Context, p *postUsageBillingParams, de
 			deps.billingCacheService.QueueUpdateSubscriptionUsage(p.User.ID, *p.APIKey.GroupID, p.Cost.ActualCost)
 		}
 	} else if p.Cost.ActualCost > 0 && p.User != nil {
-		deps.billingCacheService.QueueDeductBalance(p.User.ID, p.Cost.ActualCost)
+		syncBalanceCacheAfterDeduction(ctx, p, deps, result)
 	}
 
 	if p.Cost.ActualCost > 0 && p.APIKey != nil && p.APIKey.HasRateLimits() {
@@ -9142,6 +9146,24 @@ func finalizePostUsageBilling(ctx context.Context, p *postUsageBillingParams, de
 	go notifyAccountQuota(p, deps, result)
 }
 
+func syncBalanceCacheAfterDeduction(ctx context.Context, p *postUsageBillingParams, deps *billingDeps, result *UsageBillingApplyResult) {
+	if p == nil || p.Cost == nil || p.User == nil || deps == nil || deps.billingCacheService == nil {
+		return
+	}
+	if result != nil && result.NewBalance != nil && deps.billingCacheService.balanceBelowEligibilityThreshold(*result.NewBalance) {
+		if err := deps.billingCacheService.InvalidateUserBalance(ctx, p.User.ID); err != nil {
+			slog.Warn("invalidate balance cache after exhausted deduction failed",
+				"user_id", p.User.ID,
+				"new_balance", *result.NewBalance,
+				"balance_overdrafted", result.BalanceOverdrafted,
+				"error", err,
+			)
+		}
+		return
+	}
+	deps.billingCacheService.QueueDeductBalance(p.User.ID, p.Cost.ActualCost)
+}
+
 // notifyBalanceLow sends balance low notification after deduction.
 // When result.NewBalance is available (from DB transaction RETURNING), it is used directly
 // to reconstruct oldBalance, avoiding stale Redis reads and concurrent-deduction races.
diff --git a/backend/internal/service/usage_billing.go b/backend/internal/service/usage_billing.go
index 30495624b5..accc7cb2cb 100644
--- a/backend/internal/service/usage_billing.go
+++ b/backend/internal/service/usage_billing.go
@@ -115,6 +115,7 @@ type UsageBillingApplyResult struct {
 	Applied              bool
 	APIKeyQuotaExhausted bool
 	NewBalance           *float64           // post-deduction balance (nil = no balance deduction)
+	BalanceOverdrafted   bool               // true when the sufficient-balance guard missed and debt was still recorded
 	QuotaState           *AccountQuotaState // post-increment quota state (nil = no quota increment)
 }
 
diff --git a/deploy/config.example.yaml b/deploy/config.example.yaml
index 2cb65d8fec..67cff357e7 100644
--- a/deploy/config.example.yaml
+++ b/deploy/config.example.yaml
@@ -1027,6 +1027,10 @@ billing:
     # Number of requests to allow in half-open state
     # 半开状态允许通过的请求数
     half_open_requests: 3
+  # Conservative minimum balance required before forwarding balance-billed requests.
+  # Set to 0 to only require balance > 0.
+  # 余额计费请求转发前要求的保守最小余额；设为 0 则仅要求余额 > 0。
+  minimum_balance_reserve: 0.000001
   # Cache TTL (seconds) for per-user × per-platform quota records
   # 用户 × 平台 quota 缓存 TTL（秒），默认 86400=1天，覆盖典型 daily 窗口
   user_platform_quota_cache_ttl_seconds: 86400