diff --git a/backend/internal/service/openai_ws_forwarder_ingress_session_test.go b/backend/internal/service/openai_ws_forwarder_ingress_session_test.go
index c527f2eb..6bf9a9ff 100644
--- a/backend/internal/service/openai_ws_forwarder_ingress_session_test.go
+++ b/backend/internal/service/openai_ws_forwarder_ingress_session_test.go
@@ -413,7 +413,12 @@ func TestOpenAIGatewayService_ProxyResponsesWebSocketFromClient_PassthroughModeR
 
 	select {
 	case serverErr := <-serverErrCh:
-		require.NoError(t, serverErr)
+		// After normal client close, the server goroutine may receive the close frame
+		// as an error — this is expected behavior, not a test failure.
+		if serverErr != nil {
+			require.Contains(t, serverErr.Error(), "StatusNormalClosure",
+				"server error should only be a normal close frame, got: %v", serverErr)
+		}
 	case <-time.After(5 * time.Second):
 		t.Fatal("等待 passthrough websocket 结束超时")
 	}
diff --git a/backend/internal/service/ratelimit_service_401_test.go b/backend/internal/service/ratelimit_service_401_test.go
index 67b22e52..9e5e2b0e 100644
--- a/backend/internal/service/ratelimit_service_401_test.go
+++ b/backend/internal/service/ratelimit_service_401_test.go
@@ -102,6 +102,8 @@ func TestRateLimitService_HandleUpstreamError_OAuth401SetsTempUnschedulable(t *t
 	})
 }
 
+// TestRateLimitService_HandleUpstreamError_OAuth401InvalidatorError
+// OpenAI OAuth 401 缓存失效出错时仍走 temp_unschedulable
 func TestRateLimitService_HandleUpstreamError_OAuth401InvalidatorError(t *testing.T) {
 	repo := &rateLimitAccountRepoStub{}
 	invalidator := &tokenCacheInvalidatorRecorder{err: errors.New("boom")}
@@ -109,7 +111,7 @@ func TestRateLimitService_HandleUpstreamError_OAuth401InvalidatorError(t *testin
 	service.SetTokenCacheInvalidator(invalidator)
 	account := &Account{
 		ID:       101,
-		Platform: PlatformGemini,
+		Platform: PlatformOpenAI,
 		Type:     AccountTypeOAuth,
 	}
 
diff --git a/frontend/src/components/account/AccountUsageCell.vue b/frontend/src/components/account/AccountUsageCell.vue
index 2e4eea0c..1c023fb3 100644
--- a/frontend/src/components/account/AccountUsageCell.vue
+++ b/frontend/src/components/account/AccountUsageCell.vue
@@ -439,15 +439,20 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted, onUnmounted, watch } from 'vue'
+import { ref, computed, onMounted, onBeforeUnmount, onUnmounted, watch } from 'vue'
 import { useI18n } from 'vue-i18n'
 import { adminAPI } from '@/api/admin'
 import type { Account, AccountUsageInfo, GeminiCredentials, WindowStats } from '@/types'
 import { buildOpenAIUsageRefreshKey } from '@/utils/accountUsageRefresh'
+import { enqueueUsageRequest } from '@/utils/usageLoadQueue'
 import { formatCompactNumber } from '@/utils/format'
 import UsageProgressBar from './UsageProgressBar.vue'
 import AccountQuotaInfo from './AccountQuotaInfo.vue'
 
+// Module-level cache shared across all AccountUsageCell instances
+const _usageCache = new Map<number, { data: AccountUsageInfo; ts: number }>()
+const USAGE_CACHE_TTL = 5 * 60 * 1000 // 5 minutes
+
 const props = withDefaults(
   defineProps<{
     account: Account
@@ -465,6 +470,9 @@ const props = withDefaults(
 const { t } = useI18n()
 const desktopViewportQuery = '(min-width: 768px)'
 
+const unmounted = ref(false)
+onBeforeUnmount(() => { unmounted.value = true })
+
 const loading = ref(false)
 const activeQueryLoading = ref(false)
 const error = ref<string | null>(null)
@@ -941,19 +949,36 @@ const isAnthropicOAuthOrSetupToken = computed(() => {
   return props.account.platform === 'anthropic' && (props.account.type === 'oauth' || props.account.type === 'setup-token')
 })
 
-const loadUsage = async (source?: 'passive' | 'active') => {
+const loadUsage = async (options?: { source?: 'passive' | 'active'; bypassCache?: boolean }) => {
   if (!shouldFetchUsage.value) return
 
+  // Check cache
+  if (!options?.bypassCache) {
+    const cached = _usageCache.get(props.account.id)
+    if (cached && Date.now() - cached.ts < USAGE_CACHE_TTL) {
+      usageInfo.value = cached.data
+      loading.value = false
+      return
+    }
+  }
+
   loading.value = true
   error.value = null
 
   try {
-    usageInfo.value = await adminAPI.accounts.getUsage(props.account.id, source)
+    const fetchFn = () => adminAPI.accounts.getUsage(props.account.id, options?.source)
+    const result = await enqueueUsageRequest(props.account, fetchFn)
+    if (!unmounted.value) {
+      usageInfo.value = result
+      _usageCache.set(props.account.id, { data: result, ts: Date.now() })
+    }
   } catch (e: any) {
-    error.value = t('common.error')
-    console.error('Failed to load usage:', e)
+    if (!unmounted.value) {
+      error.value = t('common.error')
+      console.error('Failed to load usage:', e)
+    }
   } finally {
-    loading.value = false
+    if (!unmounted.value) loading.value = false
   }
 }
 
@@ -962,7 +987,7 @@ const flushPendingAutoLoad = () => {
   const source = pendingAutoLoadSource.value
   pendingAutoLoad.value = false
   pendingAutoLoadSource.value = undefined
-  loadUsage(source).catch((e) => {
+  loadUsage({ source }).catch((e) => {
     console.error('Failed to load deferred usage:', e)
   })
 }
@@ -974,7 +999,7 @@ const requestAutoLoad = (source?: 'passive' | 'active') => {
     pendingAutoLoadSource.value = source
     return
   }
-  loadUsage(source).catch((e) => {
+  loadUsage({ source }).catch((e) => {
     console.error('Failed to auto load usage:', e)
   })
 }
@@ -1138,7 +1163,10 @@ watch(
     if (!shouldFetchUsage.value) return
 
     const source = isAnthropicOAuthOrSetupToken.value ? 'passive' : undefined
-    requestAutoLoad(source)
+    _usageCache.delete(props.account.id)
+    loadUsage({ source, bypassCache: true }).catch((e) => {
+      console.error('Failed to refresh usage after manual refresh:', e)
+    })
   }
 )
 
diff --git a/frontend/src/utils/__tests__/usageLoadQueue.spec.ts b/frontend/src/utils/__tests__/usageLoadQueue.spec.ts
new file mode 100644
index 00000000..e5509261
--- /dev/null
+++ b/frontend/src/utils/__tests__/usageLoadQueue.spec.ts
@@ -0,0 +1,205 @@
+import { describe, expect, it } from 'vitest'
+import { enqueueUsageRequest } from '../usageLoadQueue'
+import type { Account } from '@/types'
+
+/** Helper to create a minimal Account with proxy info */
+function makeAccount(
+  platform: string,
+  type: string = 'oauth',
+  proxy?: { host: string; port: number; username?: string | null } | null
+): Account {
+  return {
+    id: Math.floor(Math.random() * 10000),
+    platform,
+    type,
+    name: 'test',
+    status: 'active',
+    proxy_id: proxy ? 1 : null,
+    proxy: proxy
+      ? { id: 1, name: 'p', protocol: 'http', host: proxy.host, port: proxy.port, username: proxy.username ?? null, status: 'active', created_at: '', updated_at: '' }
+      : undefined,
+    credentials: {},
+    created_at: '',
+    updated_at: ''
+  } as unknown as Account
+}
+
+describe('usageLoadQueue', () => {
+  // ─── Anthropic 账号：按代理出口排队 ───
+
+  it('Anthropic 同代理出口串行执行，间隔 >= 1s', async () => {
+    const timestamps: number[] = []
+    const makeFn = () => async () => {
+      timestamps.push(Date.now())
+      return 'ok'
+    }
+
+    const acc = makeAccount('anthropic', 'oauth', { host: '1.2.3.4', port: 8080, username: 'u1' })
+
+    const p1 = enqueueUsageRequest(acc, makeFn())
+    const p2 = enqueueUsageRequest(acc, makeFn())
+    const p3 = enqueueUsageRequest(acc, makeFn())
+
+    await Promise.all([p1, p2, p3])
+
+    expect(timestamps).toHaveLength(3)
+    expect(timestamps[1] - timestamps[0]).toBeGreaterThanOrEqual(950)
+    expect(timestamps[1] - timestamps[0]).toBeLessThan(2100)
+    expect(timestamps[2] - timestamps[1]).toBeGreaterThanOrEqual(950)
+    expect(timestamps[2] - timestamps[1]).toBeLessThan(2100)
+  })
+
+  it('Anthropic 不同代理出口并行执行', async () => {
+    const timestamps: Record<string, number> = {}
+    const makeTracked = (key: string) => async () => {
+      timestamps[key] = Date.now()
+      return key
+    }
+
+    const acc1 = makeAccount('anthropic', 'oauth', { host: '1.2.3.4', port: 8080, username: 'u1' })
+    const acc2 = makeAccount('anthropic', 'oauth', { host: '5.6.7.8', port: 3128, username: 'u2' })
+
+    const p1 = enqueueUsageRequest(acc1, makeTracked('proxy1'))
+    const p2 = enqueueUsageRequest(acc2, makeTracked('proxy2'))
+
+    await Promise.all([p1, p2])
+
+    const spread = Math.abs(timestamps['proxy1'] - timestamps['proxy2'])
+    expect(spread).toBeLessThan(50)
+  })
+
+  it('Anthropic 相同代理连接信息的不同账号归为同一队列', async () => {
+    const timestamps: number[] = []
+    const makeFn = () => async () => {
+      timestamps.push(Date.now())
+      return 'ok'
+    }
+
+    const acc1 = makeAccount('anthropic', 'oauth', { host: '10.0.0.1', port: 3128, username: 'admin' })
+    const acc2 = makeAccount('anthropic', 'setup-token', { host: '10.0.0.1', port: 3128, username: 'admin' })
+
+    const p1 = enqueueUsageRequest(acc1, makeFn())
+    const p2 = enqueueUsageRequest(acc2, makeFn())
+
+    await Promise.all([p1, p2])
+
+    expect(timestamps).toHaveLength(2)
+    expect(timestamps[1] - timestamps[0]).toBeGreaterThanOrEqual(950)
+  })
+
+  it('Anthropic 直连（无代理）的账号归为同一队列', async () => {
+    const order: number[] = []
+    const makeFn = (n: number) => async () => {
+      order.push(n)
+      return n
+    }
+
+    const acc1 = makeAccount('anthropic', 'oauth')
+    const acc2 = makeAccount('anthropic', 'setup-token')
+
+    const p1 = enqueueUsageRequest(acc1, makeFn(1))
+    const p2 = enqueueUsageRequest(acc2, makeFn(2))
+
+    await Promise.all([p1, p2])
+
+    expect(order).toEqual([1, 2])
+  })
+
+  it('Anthropic 请求失败时 reject，后续任务继续执行', async () => {
+    const results: string[] = []
+    const acc = makeAccount('anthropic', 'oauth', { host: '99.99.99.99', port: 1234 })
+
+    const p1 = enqueueUsageRequest(acc, async () => {
+      throw new Error('fail')
+    })
+    const p2 = enqueueUsageRequest(acc, async () => {
+      results.push('second')
+      return 'ok'
+    })
+
+    await expect(p1).rejects.toThrow('fail')
+    await p2
+    expect(results).toEqual(['second'])
+  })
+
+  // ─── 非 Anthropic 平台：直接执行，不排队 ───
+
+  it('非 Anthropic 平台直接执行，不排队', async () => {
+    const timestamps: number[] = []
+    const makeFn = () => async () => {
+      timestamps.push(Date.now())
+      return 'ok'
+    }
+
+    // 同一代理的 Gemini 账号 — 应当并行，不排队
+    const acc1 = makeAccount('gemini', 'oauth', { host: '1.2.3.4', port: 8080 })
+    const acc2 = makeAccount('gemini', 'oauth', { host: '1.2.3.4', port: 8080 })
+
+    const p1 = enqueueUsageRequest(acc1, makeFn())
+    const p2 = enqueueUsageRequest(acc2, makeFn())
+
+    await Promise.all([p1, p2])
+
+    expect(timestamps).toHaveLength(2)
+    // 并行执行，几乎同时完成
+    expect(Math.abs(timestamps[1] - timestamps[0])).toBeLessThan(50)
+  })
+
+  it('OpenAI 平台直接执行，不排队', async () => {
+    const timestamps: number[] = []
+    const makeFn = () => async () => {
+      timestamps.push(Date.now())
+      return 'ok'
+    }
+
+    const acc1 = makeAccount('openai', 'oauth', { host: '1.2.3.4', port: 8080 })
+    const acc2 = makeAccount('openai', 'oauth', { host: '1.2.3.4', port: 8080 })
+
+    const p1 = enqueueUsageRequest(acc1, makeFn())
+    const p2 = enqueueUsageRequest(acc2, makeFn())
+
+    await Promise.all([p1, p2])
+
+    expect(timestamps).toHaveLength(2)
+    expect(Math.abs(timestamps[1] - timestamps[0])).toBeLessThan(50)
+  })
+
+  // ─── Anthropic apikey 类型不排队 ───
+
+  it('Anthropic apikey 类型直接执行，不排队', async () => {
+    const timestamps: number[] = []
+    const makeFn = () => async () => {
+      timestamps.push(Date.now())
+      return 'ok'
+    }
+
+    const acc1 = makeAccount('anthropic', 'apikey', { host: '1.2.3.4', port: 8080 })
+    const acc2 = makeAccount('anthropic', 'apikey', { host: '1.2.3.4', port: 8080 })
+
+    const p1 = enqueueUsageRequest(acc1, makeFn())
+    const p2 = enqueueUsageRequest(acc2, makeFn())
+
+    await Promise.all([p1, p2])
+
+    expect(timestamps).toHaveLength(2)
+    expect(Math.abs(timestamps[1] - timestamps[0])).toBeLessThan(50)
+  })
+
+  // ─── 返回值透传 ───
+
+  it('返回值正确透传', async () => {
+    const acc = makeAccount('anthropic', 'oauth')
+    const result = await enqueueUsageRequest(acc, async () => {
+      return { usage: 42 }
+    })
+    expect(result).toEqual({ usage: 42 })
+  })
+
+  it('非 Anthropic 返回值正确透传', async () => {
+    const acc = makeAccount('gemini', 'oauth')
+    const result = await enqueueUsageRequest(acc, async () => {
+      return { quota: 100 }
+    })
+    expect(result).toEqual({ quota: 100 })
+  })
+})
diff --git a/frontend/src/utils/usageLoadQueue.ts b/frontend/src/utils/usageLoadQueue.ts
new file mode 100644
index 00000000..7bea5679
--- /dev/null
+++ b/frontend/src/utils/usageLoadQueue.ts
@@ -0,0 +1,93 @@
+/**
+ * Usage request scheduler — throttles Anthropic API calls by proxy exit.
+ *
+ * Anthropic OAuth/setup-token accounts sharing the same proxy exit are placed
+ * into a serial queue with a random 1–2s delay between requests, preventing
+ * upstream 429 rate-limit errors.
+ *
+ * Proxy identity = host:port:username — two proxy records pointing to the
+ * same exit share a single queue. Accounts without a proxy go into a
+ * "direct" queue.
+ *
+ * All other platforms bypass the queue and execute immediately.
+ */
+
+import type { Account } from '@/types'
+
+const GROUP_DELAY_MIN_MS = 1000
+const GROUP_DELAY_MAX_MS = 2000
+
+type Task<T> = {
+  fn: () => Promise<T>
+  resolve: (value: T) => void
+  reject: (reason: unknown) => void
+}
+
+const queues = new Map<string, Task<unknown>[]>()
+const running = new Set<string>()
+
+/** Whether this account needs throttled queuing. */
+function needsThrottle(account: Account): boolean {
+  return (
+    account.platform === 'anthropic' &&
+    (account.type === 'oauth' || account.type === 'setup-token')
+  )
+}
+
+/** Build a queue key from proxy connection details. */
+function buildGroupKey(account: Account): string {
+  const proxy = account.proxy
+  const proxyIdentity = proxy
+    ? `${proxy.host}:${proxy.port}:${proxy.username || ''}`
+    : 'direct'
+  return `anthropic:${proxyIdentity}`
+}
+
+async function drain(groupKey: string) {
+  if (running.has(groupKey)) return
+  running.add(groupKey)
+
+  const queue = queues.get(groupKey)
+  while (queue && queue.length > 0) {
+    const task = queue.shift()!
+    try {
+      const result = await task.fn()
+      task.resolve(result)
+    } catch (err) {
+      task.reject(err)
+    }
+    if (queue.length > 0) {
+      const jitter = GROUP_DELAY_MIN_MS + Math.random() * (GROUP_DELAY_MAX_MS - GROUP_DELAY_MIN_MS)
+      await new Promise((r) => setTimeout(r, jitter))
+    }
+  }
+
+  running.delete(groupKey)
+  queues.delete(groupKey)
+}
+
+/**
+ * Schedule a usage fetch. Anthropic accounts are queued by proxy exit;
+ * all other platforms execute immediately.
+ */
+export function enqueueUsageRequest<T>(
+  account: Account,
+  fn: () => Promise<T>
+): Promise<T> {
+  // Non-Anthropic → fire immediately, no queuing
+  if (!needsThrottle(account)) {
+    return fn()
+  }
+
+  const key = buildGroupKey(account)
+
+  return new Promise<T>((resolve, reject) => {
+    let queue = queues.get(key)
+    if (!queue) {
+      queue = []
+      queues.set(key, queue)
+    }
+    queue.push({ fn, resolve, reject } as Task<unknown>)
+    drain(key)
+  })
+}