diff --git a/frontend/src/features/usage/components/HorizontalRequestTimeline.vue b/frontend/src/features/usage/components/HorizontalRequestTimeline.vue
index 56310a3..cd77a76 100644
--- a/frontend/src/features/usage/components/HorizontalRequestTimeline.vue
+++ b/frontend/src/features/usage/components/HorizontalRequestTimeline.vue
@@ -479,10 +479,25 @@ const groupedTimeline = computed<NodeGroup[]>(() => {
   return groups
 })
 
-// 计算链路总耗时（从第一个节点开始到最后一个节点结束）
+// 计算链路总耗时（使用成功候选的 latency_ms 字段）
+// 优先使用 latency_ms，因为它与 Usage.response_time_ms 使用相同的时间基准
+// 避免 finished_at - started_at 带来的额外延迟（数据库操作时间）
 const totalTraceLatency = computed(() => {
   if (!timeline.value || timeline.value.length === 0) return 0
 
+  // 查找成功的候选，使用其 latency_ms
+  const successCandidate = timeline.value.find(c => c.status === 'success')
+  if (successCandidate?.latency_ms != null) {
+    return successCandidate.latency_ms
+  }
+
+  // 如果没有成功的候选，查找失败但有 latency_ms 的候选
+  const failedWithLatency = timeline.value.find(c => c.status === 'failed' && c.latency_ms != null)
+  if (failedWithLatency?.latency_ms != null) {
+    return failedWithLatency.latency_ms
+  }
+
+  // 回退：使用 finished_at - started_at 计算
   let earliestStart: number | null = null
   let latestEnd: number | null = null
 
diff --git a/frontend/src/features/usage/components/UsageRecordsTable.vue b/frontend/src/features/usage/components/UsageRecordsTable.vue
index a0ff06f..0e184be 100644
--- a/frontend/src/features/usage/components/UsageRecordsTable.vue
+++ b/frontend/src/features/usage/components/UsageRecordsTable.vue
@@ -177,8 +177,9 @@
             费用
           </TableHead>
           <TableHead class="h-12 font-semibold w-[70px] text-right">
-            <div class="inline-block max-w-[2rem] leading-tight">
-              响应时间
+            <div class="flex flex-col items-end text-xs gap-0.5">
+              <span>首字</span>
+              <span class="text-muted-foreground font-normal">总耗时</span>
             </div>
           </TableHead>
         </TableRow>
@@ -356,15 +357,28 @@
             </div>
           </TableCell>
           <TableCell class="text-right py-4 w-[70px]">
-            <span
+            <div
               v-if="record.status === 'pending' || record.status === 'streaming'"
-              class="text-primary tabular-nums"
+              class="flex flex-col items-end text-xs gap-0.5"
             >
-              {{ getElapsedTime(record) }}
-            </span>
-            <span v-else-if="record.response_time_ms">
-              {{ (record.response_time_ms / 1000).toFixed(2) }}s
-            </span>
+              <span class="text-primary tabular-nums">
+                {{ getElapsedTime(record) }}
+              </span>
+            </div>
+            <div
+              v-else-if="record.response_time_ms != null"
+              class="flex flex-col items-end text-xs gap-0.5"
+            >
+              <span
+                v-if="record.first_byte_time_ms != null"
+                class="tabular-nums"
+              >{{ (record.first_byte_time_ms / 1000).toFixed(2) }}s</span>
+              <span
+                v-else
+                class="text-muted-foreground"
+              >-</span>
+              <span class="text-muted-foreground tabular-nums">{{ (record.response_time_ms / 1000).toFixed(2) }}s</span>
+            </div>
             <span
               v-else
               class="text-muted-foreground"
diff --git a/frontend/src/features/usage/types.ts b/frontend/src/features/usage/types.ts
index 62475c2..3561e42 100644
--- a/frontend/src/features/usage/types.ts
+++ b/frontend/src/features/usage/types.ts
@@ -78,6 +78,7 @@ export interface UsageRecord {
   cost: number
   actual_cost?: number
   response_time_ms?: number
+  first_byte_time_ms?: number  // 首字时间 (TTFB)
   is_stream: boolean
   status_code?: number
   error_message?: string