diff --git a/common/constants.go b/common/constants.go
index 274c514f9146..69a746e88a5c 100644
--- a/common/constants.go
+++ b/common/constants.go
@@ -173,7 +173,7 @@ var (
 
 	CriticalRateLimitEnable   bool
 	CriticalRateLimitNum            = 20
-	CriticalRateLimitDuration int64 = 20 * 60
+	CriticalRateLimitDuration int64 = 60
 
 	UploadRateLimitNum            = 10
 	UploadRateLimitDuration int64 = 60
diff --git a/common/init.go b/common/init.go
index 35b4c6be17ee..107301832678 100644
--- a/common/init.go
+++ b/common/init.go
@@ -120,7 +120,7 @@ func InitEnv() {
 
 	CriticalRateLimitEnable = GetEnvOrDefaultBool("CRITICAL_RATE_LIMIT_ENABLE", true)
 	CriticalRateLimitNum = GetEnvOrDefault("CRITICAL_RATE_LIMIT", 20)
-	CriticalRateLimitDuration = int64(GetEnvOrDefault("CRITICAL_RATE_LIMIT_DURATION", 20*60))
+	CriticalRateLimitDuration = int64(GetEnvOrDefault("CRITICAL_RATE_LIMIT_DURATION", 60))
 
 	SearchRateLimitEnable = GetEnvOrDefaultBool("SEARCH_RATE_LIMIT_ENABLE", true)
 	SearchRateLimitNum = GetEnvOrDefault("SEARCH_RATE_LIMIT", 10)
diff --git a/middleware/rate-limit.go b/middleware/rate-limit.go
index d8dd15d9c5d7..9c6ab6728eab 100644
--- a/middleware/rate-limit.go
+++ b/middleware/rate-limit.go
@@ -18,6 +18,14 @@ var defNext = func(c *gin.Context) {
 	c.Next()
 }
 
+func abortRateLimited(c *gin.Context) {
+	c.JSON(http.StatusTooManyRequests, gin.H{
+		"success": false,
+		"message": "请求过于频繁，请稍后再试",
+	})
+	c.Abort()
+}
+
 func redisRateLimiter(c *gin.Context, maxRequestNum int, duration int64, mark string) {
 	ctx := context.Background()
 	rdb := common.RDB
@@ -53,8 +61,7 @@ func redisRateLimiter(c *gin.Context, maxRequestNum int, duration int64, mark st
 		// See: https://stackoverflow.com/questions/50970900/why-is-time-since-returning-negative-durations-on-windows
 		if int64(nowTime.Sub(oldTime).Seconds()) < duration {
 			rdb.Expire(ctx, key, common.RateLimitKeyExpirationDuration)
-			c.Status(http.StatusTooManyRequests)
-			c.Abort()
+			abortRateLimited(c)
 			return
 		} else {
 			rdb.LPush(ctx, key, time.Now().Format(timeFormat))
@@ -67,8 +74,7 @@ func redisRateLimiter(c *gin.Context, maxRequestNum int, duration int64, mark st
 func memoryRateLimiter(c *gin.Context, maxRequestNum int, duration int64, mark string) {
 	key := mark + c.ClientIP()
 	if !inMemoryRateLimiter.Request(key, maxRequestNum, duration) {
-		c.Status(http.StatusTooManyRequests)
-		c.Abort()
+		abortRateLimited(c)
 		return
 	}
 }
@@ -143,8 +149,7 @@ func userRateLimitFactory(maxRequestNum int, duration int64, mark string) func(c
 		}
 		key := fmt.Sprintf("%s:user:%d", mark, userId)
 		if !inMemoryRateLimiter.Request(key, maxRequestNum, duration) {
-			c.Status(http.StatusTooManyRequests)
-			c.Abort()
+			abortRateLimited(c)
 			return
 		}
 	}
@@ -184,8 +189,7 @@ func userRedisRateLimiter(c *gin.Context, maxRequestNum int, duration int64, key
 		}
 		if int64(nowTime.Sub(oldTime).Seconds()) < duration {
 			rdb.Expire(ctx, key, common.RateLimitKeyExpirationDuration)
-			c.Status(http.StatusTooManyRequests)
-			c.Abort()
+			abortRateLimited(c)
 			return
 		} else {
 			rdb.LPush(ctx, key, time.Now().Format(timeFormat))
diff --git a/model/log.go b/model/log.go
index 3cc9b74b24b5..43ac690f24b2 100644
--- a/model/log.go
+++ b/model/log.go
@@ -37,7 +37,8 @@ type Log struct {
 	Ip               string `json:"ip" gorm:"index;default:''"`
 	RequestId        string `json:"request_id,omitempty" gorm:"type:varchar(64);index:idx_logs_request_id;default:''"`
 	Other            string `json:"other"`
-	Balance          int    `json:"balance" gorm:"-"`
+	Balance          int    `json:"balance" gorm:"default:0"`
+	BalanceValid     bool   `json:"-" gorm:"default:false"`
 }
 
 // don't use iota, avoid change log type value
@@ -87,7 +88,7 @@ func AttachLogBalances(logs []*Log) {
 	}
 	balances := make(map[int]int)
 	for _, log := range logs {
-		if log == nil || log.UserId == 0 {
+		if log == nil || log.UserId == 0 || log.BalanceValid {
 			continue
 		}
 		if _, ok := balances[log.UserId]; ok {
@@ -105,7 +106,7 @@ func AttachLogBalances(logs []*Log) {
 		balances[log.UserId] = quota - newerDelta
 	}
 	for _, log := range logs {
-		if log == nil || log.UserId == 0 {
+		if log == nil || log.UserId == 0 || log.BalanceValid {
 			continue
 		}
 		balance, ok := balances[log.UserId]
@@ -117,6 +118,20 @@ func AttachLogBalances(logs []*Log) {
 	}
 }
 
+func attachCurrentBalanceSnapshot(log *Log) {
+	if log == nil || log.UserId == 0 {
+		return
+	}
+	quota, err := GetUserQuota(log.UserId, true)
+	if err != nil {
+		common.SysLog(fmt.Sprintf("failed to attach log balance snapshot for user %d: %v", log.UserId, err))
+		return
+	}
+	quota += pendingBatchUpdateValue(BatchUpdateTypeUserQuota, log.UserId)
+	log.Balance = quota
+	log.BalanceValid = true
+}
+
 func GetLogByTokenId(tokenId int) (logs []*Log, err error) {
 	err = LOG_DB.Model(&Log{}).Where("token_id = ?", tokenId).Order("id desc").Limit(common.MaxRecentItems).Find(&logs).Error
 	formatUserLogs(logs, 0)
@@ -135,6 +150,7 @@ func RecordLog(userId int, logType int, content string) {
 		Type:      logType,
 		Content:   content,
 	}
+	attachCurrentBalanceSnapshot(log)
 	err := LOG_DB.Create(log).Error
 	if err != nil {
 		common.SysLog("failed to record log: " + err.Error())
@@ -154,6 +170,7 @@ func RecordLogWithAdminInfo(userId int, logType int, content string, adminInfo m
 		Type:      logType,
 		Content:   content,
 	}
+	attachCurrentBalanceSnapshot(log)
 	if len(adminInfo) > 0 {
 		other := map[string]interface{}{
 			"admin_info": adminInfo,
@@ -187,6 +204,7 @@ func RecordTopupLog(userId int, content string, callerIp string, paymentMethod s
 		Ip:        callerIp,
 		Other:     common.MapToJsonStr(other),
 	}
+	attachCurrentBalanceSnapshot(log)
 	err := LOG_DB.Create(log).Error
 	if err != nil {
 		common.SysLog("failed to record topup log: " + err.Error())
@@ -231,6 +249,7 @@ func RecordErrorLog(c *gin.Context, userId int, channelId int, modelName string,
 		RequestId: requestId,
 		Other:     otherStr,
 	}
+	attachCurrentBalanceSnapshot(log)
 	err := LOG_DB.Create(log).Error
 	if err != nil {
 		logger.LogError(c, "failed to record log: "+err.Error())
@@ -292,6 +311,7 @@ func RecordConsumeLog(c *gin.Context, userId int, params RecordConsumeLogParams)
 		RequestId: requestId,
 		Other:     otherStr,
 	}
+	attachCurrentBalanceSnapshot(log)
 	err := LOG_DB.Create(log).Error
 	if err != nil {
 		logger.LogError(c, "failed to record log: "+err.Error())
@@ -340,6 +360,7 @@ func RecordTaskBillingLog(params RecordTaskBillingLogParams) {
 		Group:     params.Group,
 		Other:     common.MapToJsonStr(params.Other),
 	}
+	attachCurrentBalanceSnapshot(log)
 	err := LOG_DB.Create(log).Error
 	if err != nil {
 		common.SysLog("failed to record task billing log: " + err.Error())
diff --git a/model/utils.go b/model/utils.go
index adfd8e139a05..ca53f47f4ed6 100644
--- a/model/utils.go
+++ b/model/utils.go
@@ -49,6 +49,15 @@ func addNewRecord(type_ int, id int, value int) {
 	}
 }
 
+func pendingBatchUpdateValue(type_ int, id int) int {
+	if !common.BatchUpdateEnabled {
+		return 0
+	}
+	batchUpdateLocks[type_].Lock()
+	defer batchUpdateLocks[type_].Unlock()
+	return batchUpdateStores[type_][id]
+}
+
 func batchUpdate() {
 	// check if there's any data to update
 	hasData := false
diff --git a/service/billing_session.go b/service/billing_session.go
index 7ca158bd5255..4a5119299d22 100644
--- a/service/billing_session.go
+++ b/service/billing_session.go
@@ -119,10 +119,8 @@ func (s *BillingSession) Refund(c *gin.Context) {
 				common.SysLog("error refunding token quota: " + err.Error())
 			}
 		}
-		// 3) 退还用户已用额度（tokenConsumed已经是折扣后的值）
-		if tokenConsumed > 0 {
-			model.UpdateUserUsedQuotaAndRequestCount(s.relayInfo.UserId, -tokenConsumed)
-		}
+		// 提交阶段失败不会写消费日志，也不会增加 users.used_quota。
+		// 这里只回滚预扣的资金和令牌额度；异步任务已生成消费日志后的失败退款由 RefundTaskQuota 回退 used_quota。
 	})
 }
 
diff --git a/service/task_billing.go b/service/task_billing.go
index 54a1f3fc4266..aa3c54c6e301 100644
--- a/service/task_billing.go
+++ b/service/task_billing.go
@@ -165,9 +165,9 @@ func RefundTaskQuota(ctx context.Context, task *model.Task, reason string) {
 	// 2. 退还令牌额度
 	taskAdjustTokenQuota(ctx, task, -quota)
 
-	//3. 更新用户和渠道的使用额度
-	//model.UpdateUserUsedQuotaAndRequestCount(task.UserId, -quota)
-	//model.UpdateChannelUsedQuota(task.ChannelId, -quota)
+	// 3. 更新用户和渠道的使用额度
+	model.UpdateUserUsedQuotaAndRequestCount(task.UserId, -quota)
+	model.UpdateChannelUsedQuota(task.ChannelId, -quota)
 
 	// 4. 记录日志
 	other := taskBillingOther(task)
@@ -255,8 +255,8 @@ func RecalculateTaskQuota(ctx context.Context, task *model.Task, actualQuota int
 	} else {
 		logType = model.LogTypeRefund
 		logQuota = -quotaDelta
-		//model.UpdateUserUsedQuotaAndRequestCount(task.UserId, -quotaDelta)
-		//model.UpdateChannelUsedQuota(task.ChannelId, -quotaDelta)
+		model.UpdateUserUsedQuotaAndRequestCount(task.UserId, quotaDelta)
+		model.UpdateChannelUsedQuota(task.ChannelId, quotaDelta)
 	}
 	other := taskBillingOther(task)
 	other["task_id"] = task.TaskID