diff --git a/src/components/AIChat/ChatExplorer.vue b/src/components/AIChat/ChatExplorer.vue
index 45c2f89e..b698c80a 100644
--- a/src/components/AIChat/ChatExplorer.vue
+++ b/src/components/AIChat/ChatExplorer.vue
@@ -23,6 +23,7 @@ import { useSettingsStore } from '@/stores/settings'
 import { useAssistantStore } from '@/stores/assistant'
 import { useSkillStore } from '@/stores/skill'
 import { useChatScroll } from './composables/useChatScroll'
+import { useProgressiveChatHistory } from './composables/useProgressiveChatHistory'
 import { useChatModals } from './composables/useChatModals'
 import { groupMessagesToQAPairs } from './utils/chatMessages'
 import type { MentionedMemberContext } from '@/composables/useAIChat'
@@ -181,6 +182,8 @@ const assistantBackupName = computed(() => {
 
 // QA 对
 const qaPairs = computed(() => groupMessagesToQAPairs(messages.value))
+const progressiveHistory = useProgressiveChatHistory(qaPairs, currentAIChatId, chatScroll.messagesContainer)
+const { visiblePairs, hasOlderPairs, loadOlderPairs } = progressiveHistory
 
 // 检查 LLM 配置
 async function checkLLMConfig() {
@@ -515,8 +518,19 @@ watch(
                 </div>
               </div>
 
+              <div v-if="hasOlderPairs" class="flex justify-center pb-2">
+                <button
+                  type="button"
+                  class="flex h-8 items-center gap-1.5 rounded-full px-3 text-xs text-gray-500 transition-colors hover:bg-gray-100 hover:text-gray-700 dark:text-gray-400 dark:hover:bg-gray-800 dark:hover:text-gray-200"
+                  @click="loadOlderPairs"
+                >
+                  <UIcon name="i-heroicons-arrow-up" class="h-3.5 w-3.5" />
+                  <span>{{ t('ai.chat.history.loadEarlier') }}</span>
+                </button>
+              </div>
+
               <!-- QA 对渲染 -->
-              <template v-for="pair in qaPairs" :key="pair.id">
+              <template v-for="pair in visiblePairs" :key="pair.id">
                 <!-- 独立消息（summary 等非 user/assistant） -->
                 <ChatMessage
                   v-if="pair.standalone"
diff --git a/src/components/AIChat/chat/ChatMessage.vue b/src/components/AIChat/chat/ChatMessage.vue
index 6cc8ac90..67b1d015 100644
--- a/src/components/AIChat/chat/ChatMessage.vue
+++ b/src/components/AIChat/chat/ChatMessage.vue
@@ -88,9 +88,26 @@ md.renderer.rules.link_open = (tokens, idx, options, _env, self) => {
 }
 
 // 渲染 Markdown 文本
+const MAX_MARKDOWN_CACHE_ENTRIES = 32
+const markdownCache = new Map<string, string>()
+
 function renderMarkdown(text: string): string {
   if (!text) return ''
-  return md.render(text)
+
+  const cached = markdownCache.get(text)
+  if (cached !== undefined) {
+    markdownCache.delete(text)
+    markdownCache.set(text, cached)
+    return cached
+  }
+
+  const rendered = md.render(text)
+  markdownCache.set(text, rendered)
+  if (markdownCache.size > MAX_MARKDOWN_CACHE_ENTRIES) {
+    const oldest = markdownCache.keys().next().value
+    if (oldest !== undefined) markdownCache.delete(oldest)
+  }
+  return rendered
 }
 
 // 思考标签名称映射
diff --git a/src/components/AIChat/composables/useChatScroll.test.ts b/src/components/AIChat/composables/useChatScroll.test.ts
new file mode 100644
index 00000000..0dd6aa12
--- /dev/null
+++ b/src/components/AIChat/composables/useChatScroll.test.ts
@@ -0,0 +1,40 @@
+import assert from 'node:assert/strict'
+import test from 'node:test'
+import { createCoalescedScrollScheduler } from './useChatScroll'
+
+test('coalesces repeated scroll requests into one pending task', (t) => {
+  t.mock.timers.enable({ apis: ['setTimeout'] })
+  const calls: boolean[] = []
+  const scheduler = createCoalescedScrollScheduler((force) => calls.push(force))
+
+  scheduler.schedule()
+  scheduler.schedule()
+  scheduler.schedule()
+  t.mock.timers.tick(100)
+
+  assert.deepEqual(calls, [false])
+})
+
+test('preserves a force request when upgrading a pending scroll', (t) => {
+  t.mock.timers.enable({ apis: ['setTimeout'] })
+  const calls: boolean[] = []
+  const scheduler = createCoalescedScrollScheduler((force) => calls.push(force))
+
+  scheduler.schedule()
+  scheduler.schedule(true)
+  t.mock.timers.tick(100)
+
+  assert.deepEqual(calls, [true])
+})
+
+test('cancels the pending scroll task', (t) => {
+  t.mock.timers.enable({ apis: ['setTimeout'] })
+  const calls: boolean[] = []
+  const scheduler = createCoalescedScrollScheduler((force) => calls.push(force))
+
+  scheduler.schedule(true)
+  scheduler.cancel()
+  t.mock.timers.tick(100)
+
+  assert.deepEqual(calls, [])
+})
diff --git a/src/components/AIChat/composables/useChatScroll.ts b/src/components/AIChat/composables/useChatScroll.ts
index 0bb509b0..a952e1b2 100644
--- a/src/components/AIChat/composables/useChatScroll.ts
+++ b/src/components/AIChat/composables/useChatScroll.ts
@@ -10,21 +10,57 @@ import type { Ref } from 'vue'
 import type { ChatMessage } from '@/stores/aiChat'
 
 const RESTICK_THRESHOLD = 30
+const SCROLL_DELAY_MS = 100
+
+export function createCoalescedScrollScheduler(
+  callback: (force: boolean) => void,
+  delayMs = SCROLL_DELAY_MS
+): {
+  schedule: (force?: boolean) => void
+  cancel: () => void
+} {
+  let pendingTimer: ReturnType<typeof setTimeout> | null = null
+  let pendingForce = false
+
+  function schedule(force = false) {
+    pendingForce ||= force
+    if (pendingTimer !== null) return
+
+    pendingTimer = setTimeout(() => {
+      pendingTimer = null
+      const forceCurrentRun = pendingForce
+      pendingForce = false
+      callback(forceCurrentRun)
+    }, delayMs)
+  }
+
+  function cancel() {
+    if (pendingTimer !== null) {
+      clearTimeout(pendingTimer)
+      pendingTimer = null
+    }
+    pendingForce = false
+  }
+
+  return { schedule, cancel }
+}
 
 export function useChatScroll(messages: Ref<ChatMessage[]>, isAIThinking: Ref<boolean>) {
   const messagesContainer = ref<HTMLElement | null>(null)
   const isStickToBottom = ref(true)
   const showScrollToBottom = ref(false)
 
+  const scrollScheduler = createCoalescedScrollScheduler((force) => {
+    if (!messagesContainer.value) return
+    if (force || isStickToBottom.value) {
+      messagesContainer.value.scrollTop = messagesContainer.value.scrollHeight
+      isStickToBottom.value = true
+      showScrollToBottom.value = false
+    }
+  })
+
   function scrollToBottom(force = false) {
-    setTimeout(() => {
-      if (!messagesContainer.value) return
-      if (force || isStickToBottom.value) {
-        messagesContainer.value.scrollTop = messagesContainer.value.scrollHeight
-        isStickToBottom.value = true
-        showScrollToBottom.value = false
-      }
-    }, 100)
+    scrollScheduler.schedule(force)
   }
 
   function handleWheel(event: WheelEvent) {
@@ -67,6 +103,7 @@ export function useChatScroll(messages: Ref<ChatMessage[]>, isAIThinking: Ref<bo
   })
 
   onBeforeUnmount(() => {
+    scrollScheduler.cancel()
     if (messagesContainer.value) {
       messagesContainer.value.removeEventListener('scroll', checkScrollPosition)
       messagesContainer.value.removeEventListener('wheel', handleWheel)
diff --git a/src/components/AIChat/composables/useProgressiveChatHistory.test.ts b/src/components/AIChat/composables/useProgressiveChatHistory.test.ts
new file mode 100644
index 00000000..828fc831
--- /dev/null
+++ b/src/components/AIChat/composables/useProgressiveChatHistory.test.ts
@@ -0,0 +1,72 @@
+import assert from 'node:assert/strict'
+import test from 'node:test'
+import { effectScope, nextTick, ref } from 'vue'
+import { useProgressiveChatHistory } from './useProgressiveChatHistory'
+import type { QAPair } from '../utils/chatMessages'
+
+function createPairs(count: number): QAPair[] {
+  return Array.from({ length: count }, (_, index) => ({
+    id: `pair-${index}`,
+    user: null,
+    assistant: null,
+    standalone: null,
+  }))
+}
+
+test('starts with the latest 20 pairs and reveals older history in batches', async () => {
+  const scope = effectScope()
+  const pairs = ref(createPairs(45))
+  const conversationId = ref<string | null>('conversation-a')
+  const container = ref<HTMLElement | null>(null)
+  const history = scope.run(() => useProgressiveChatHistory(pairs, conversationId, container))!
+
+  assert.deepEqual(
+    history.visiblePairs.value.map((pair) => pair.id),
+    createPairs(45)
+      .slice(-20)
+      .map((pair) => pair.id)
+  )
+  await history.loadOlderPairs()
+  assert.equal(history.visiblePairs.value.length, 40)
+  await history.loadOlderPairs()
+  assert.equal(history.visiblePairs.value.length, 45)
+  assert.equal(history.hasOlderPairs.value, false)
+
+  scope.stop()
+})
+
+test('preserves the visible anchor when older pairs are inserted above', async () => {
+  const scope = effectScope()
+  const pairs = ref(createPairs(45))
+  const conversationId = ref<string | null>('conversation-a')
+  const container = ref<HTMLElement | null>(null)
+  const history = scope.run(() => useProgressiveChatHistory(pairs, conversationId, container))!
+  const element = {
+    scrollTop: 120,
+    get scrollHeight() {
+      return history.visiblePairs.value.length * 100
+    },
+  } as HTMLElement
+  container.value = element
+
+  await history.loadOlderPairs()
+
+  assert.equal(element.scrollTop, 2_120)
+  scope.stop()
+})
+
+test('resets the window when switching conversations', async () => {
+  const scope = effectScope()
+  const pairs = ref(createPairs(10))
+  const conversationId = ref<string | null>('conversation-a')
+  const container = ref<HTMLElement | null>(null)
+  const history = scope.run(() => useProgressiveChatHistory(pairs, conversationId, container))!
+
+  assert.equal(history.visiblePairs.value.length, 10)
+  conversationId.value = 'conversation-b'
+  pairs.value = createPairs(45)
+  await nextTick()
+
+  assert.equal(history.visiblePairs.value.length, 20)
+  scope.stop()
+})
diff --git a/src/components/AIChat/composables/useProgressiveChatHistory.ts b/src/components/AIChat/composables/useProgressiveChatHistory.ts
new file mode 100644
index 00000000..321bf9c2
--- /dev/null
+++ b/src/components/AIChat/composables/useProgressiveChatHistory.ts
@@ -0,0 +1,49 @@
+import { computed, nextTick, ref, watch, type Ref } from 'vue'
+import type { QAPair } from '../utils/chatMessages'
+
+export const CHAT_HISTORY_BATCH_SIZE = 20
+
+export function useProgressiveChatHistory(
+  pairs: Ref<QAPair[]>,
+  conversationId: Ref<string | null>,
+  scrollContainer: Ref<HTMLElement | null>
+) {
+  const visibleCount = ref(CHAT_HISTORY_BATCH_SIZE)
+  const visiblePairs = computed(() => pairs.value.slice(-visibleCount.value))
+  const hasOlderPairs = computed(() => visibleCount.value < pairs.value.length)
+
+  watch(
+    [conversationId, () => pairs.value.length] as const,
+    ([currentConversationId, length], [previousConversationId, previousLength]) => {
+      if (currentConversationId !== previousConversationId) {
+        visibleCount.value = CHAT_HISTORY_BATCH_SIZE
+        return
+      }
+      if (length < previousLength) {
+        visibleCount.value = Math.max(CHAT_HISTORY_BATCH_SIZE, Math.min(visibleCount.value, length))
+      } else if (previousLength > 0 && visibleCount.value >= previousLength) {
+        visibleCount.value = length
+      }
+    }
+  )
+
+  async function loadOlderPairs(): Promise<void> {
+    if (!hasOlderPairs.value) return
+
+    const container = scrollContainer.value
+    const previousScrollHeight = container?.scrollHeight ?? 0
+    const previousScrollTop = container?.scrollTop ?? 0
+    visibleCount.value = Math.min(pairs.value.length, visibleCount.value + CHAT_HISTORY_BATCH_SIZE)
+    await nextTick()
+
+    if (container && scrollContainer.value === container) {
+      container.scrollTop = previousScrollTop + (container.scrollHeight - previousScrollHeight)
+    }
+  }
+
+  return {
+    visiblePairs,
+    hasOlderPairs,
+    loadOlderPairs,
+  }
+}
diff --git a/src/i18n/locales/en-US/ai.json b/src/i18n/locales/en-US/ai.json
index 817c26b1..81403731 100644
--- a/src/i18n/locales/en-US/ai.json
+++ b/src/i18n/locales/en-US/ai.json
@@ -5,6 +5,9 @@
   },
   "chat": {
     "scrollToBottom": "Back to Bottom",
+    "history": {
+      "loadEarlier": "Load earlier messages"
+    },
     "input": {
       "placeholderWithSlash": "Enter your question, or type / for skills and {'@'} for members...",
       "placeholderWithActiveSkill": "/{name} selected. Enter your specific question...",
diff --git a/src/i18n/locales/ja-JP/ai.json b/src/i18n/locales/ja-JP/ai.json
index f8e4d6af..4604ae1c 100644
--- a/src/i18n/locales/ja-JP/ai.json
+++ b/src/i18n/locales/ja-JP/ai.json
@@ -5,6 +5,9 @@
   },
   "chat": {
     "scrollToBottom": "下へ移動",
+    "history": {
+      "loadEarlier": "以前の会話を読み込む"
+    },
     "input": {
       "placeholderWithSlash": "質問を入力、または / でスキル、{'@'} でメンバーを選択...",
       "placeholderWithActiveSkill": "/{name} を選択済みです。具体的な質問を入力してください...",
diff --git a/src/i18n/locales/zh-CN/ai.json b/src/i18n/locales/zh-CN/ai.json
index 8e818767..83219e65 100644
--- a/src/i18n/locales/zh-CN/ai.json
+++ b/src/i18n/locales/zh-CN/ai.json
@@ -5,6 +5,9 @@
   },
   "chat": {
     "scrollToBottom": "返回底部",
+    "history": {
+      "loadEarlier": "加载更早对话"
+    },
     "input": {
       "placeholderWithSlash": "输入你的问题，或输入 / 选择技能、输入 {'@'} 选择成员...",
       "placeholderWithActiveSkill": "已选择 /{name}，请输入具体问题...",
diff --git a/src/i18n/locales/zh-TW/ai.json b/src/i18n/locales/zh-TW/ai.json
index 88e1bc6d..0fbb695b 100644
--- a/src/i18n/locales/zh-TW/ai.json
+++ b/src/i18n/locales/zh-TW/ai.json
@@ -5,6 +5,9 @@
   },
   "chat": {
     "scrollToBottom": "回到底部",
+    "history": {
+      "loadEarlier": "載入更早對話"
+    },
     "input": {
       "placeholderWithSlash": "輸入你的問題，或輸入 / 選擇技能、輸入 {'@'} 選擇成員...",
       "placeholderWithActiveSkill": "已選擇 /{name}，請輸入具體問題...",
diff --git a/src/stores/aiChat.ts b/src/stores/aiChat.ts
index f0e7a329..c8370987 100644
--- a/src/stores/aiChat.ts
+++ b/src/stores/aiChat.ts
@@ -44,6 +44,7 @@ import {
   toSerializableContentBlocks,
 } from './aiChatContentBlocks'
 import { createToolLifecycleTracker, type ToolStatus } from './aiToolLifecycle'
+import { createAIStreamTextBatcher } from './aiChatStreamBatcher'
 
 export type { ToolStatus } from './aiToolLifecycle'
 
@@ -664,61 +665,18 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
     ) => void
     completeTurn: (hadToolCalls: boolean) => void
     settleProcessDuration: () => void
+    flushPendingText: () => void
   }
 
   function createStreamBlockHelpers(targetBuffer: AIChatBuffer, getAiMessageIndex: () => number): StreamBlockHelpers {
     let currentTurnFirstContentAt: number | undefined
 
-    const updateAIMessage = (updates: Partial<ChatMessage>) => {
+    const applyAIMessageUpdates = (updates: Partial<ChatMessage>) => {
       const idx = getAiMessageIndex()
       targetBuffer.messages[idx] = { ...targetBuffer.messages[idx], ...updates }
     }
 
-    const settleProcessDuration = () => {
-      const idx = getAiMessageIndex()
-      const message = targetBuffer.messages[idx]
-      if (!message || message.processDurationMs !== undefined) return
-      const durationMs = Math.max(0, Date.now() - message.timestamp)
-      updateAIMessage({
-        processDurationMs: durationMs,
-        contentBlocks: persistProcessDurationMs(message.contentBlocks, durationMs),
-      })
-    }
-
-    const appendTextToBlocks = (text: string) => {
-      if (!text) return
-      const idx = getAiMessageIndex()
-      const blocks = targetBuffer.messages[idx].contentBlocks || []
-      const lastBlock = blocks[blocks.length - 1]
-      if (text.trim().length === 0 && (!lastBlock || lastBlock.type !== 'text')) return
-      if (text.trim().length > 0 && currentTurnFirstContentAt === undefined) {
-        currentTurnFirstContentAt = Date.now()
-      }
-      if (lastBlock && lastBlock.type === 'text') {
-        lastBlock.text += text
-      } else {
-        blocks.push({ type: 'text', text })
-      }
-      updateAIMessage({ contentBlocks: [...blocks], content: targetBuffer.messages[idx].content + text })
-    }
-
-    const completeTurn = (hadToolCalls: boolean) => {
-      const firstContentAt = currentTurnFirstContentAt
-      currentTurnFirstContentAt = undefined
-      if (hadToolCalls || firstContentAt === undefined) return
-
-      const idx = getAiMessageIndex()
-      const message = targetBuffer.messages[idx]
-      if (!message || message.processDurationMs !== undefined) return
-      const durationMs = Math.max(0, firstContentAt - message.timestamp)
-      updateAIMessage({
-        processDurationMs: durationMs,
-        contentBlocks: persistProcessDurationMs(message.contentBlocks, durationMs),
-      })
-    }
-
-    const appendThinkToBlocks = (text: string, tag?: string, durationMs?: number) => {
-      if (!text && durationMs === undefined) return
+    const appendThinkImmediately = (text: string, tag?: string, durationMs?: number) => {
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       const thinkTag = tag || 'think'
@@ -741,11 +699,95 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
       if (durationMs !== undefined && targetBlock && targetBlock.type === 'think') {
         targetBlock.durationMs = durationMs
       }
-      updateAIMessage({ contentBlocks: [...blocks] })
+      applyAIMessageUpdates({ contentBlocks: [...blocks] })
+    }
+
+    const streamTextBatcher = createAIStreamTextBatcher((deltas) => {
+      const idx = getAiMessageIndex()
+      const message = targetBuffer.messages[idx]
+      const blocks = message.contentBlocks || []
+      let content = message.content
+      let changed = false
+
+      for (const delta of deltas) {
+        const lastBlock = blocks[blocks.length - 1]
+        if (delta.type === 'content') {
+          if (delta.content.trim().length === 0 && (!lastBlock || lastBlock.type !== 'text')) continue
+          if (lastBlock?.type === 'text') lastBlock.text += delta.content
+          else blocks.push({ type: 'text', text: delta.content })
+          content += delta.content
+          changed = true
+          continue
+        }
+
+        const thinkTag = delta.thinkTag || 'think'
+        if (lastBlock?.type === 'think' && lastBlock.tag === thinkTag) {
+          lastBlock.text += delta.content
+          changed = true
+        } else if (delta.content.trim().length > 0) {
+          blocks.push({ type: 'think', tag: thinkTag, text: delta.content })
+          changed = true
+        }
+      }
+
+      if (changed) applyAIMessageUpdates({ contentBlocks: [...blocks], content })
+    })
+
+    const flushPendingText = () => streamTextBatcher.flush()
+
+    const updateAIMessage = (updates: Partial<ChatMessage>) => {
+      flushPendingText()
+      applyAIMessageUpdates(updates)
+    }
+
+    const settleProcessDuration = () => {
+      flushPendingText()
+      const idx = getAiMessageIndex()
+      const message = targetBuffer.messages[idx]
+      if (!message || message.processDurationMs !== undefined) return
+      const durationMs = Math.max(0, Date.now() - message.timestamp)
+      applyAIMessageUpdates({
+        processDurationMs: durationMs,
+        contentBlocks: persistProcessDurationMs(message.contentBlocks, durationMs),
+      })
+    }
+
+    const appendTextToBlocks = (text: string) => {
+      if (!text) return
+      if (text.trim().length > 0 && currentTurnFirstContentAt === undefined) {
+        currentTurnFirstContentAt = Date.now()
+      }
+      streamTextBatcher.push({ type: 'content', content: text })
+    }
+
+    const completeTurn = (hadToolCalls: boolean) => {
+      flushPendingText()
+      const firstContentAt = currentTurnFirstContentAt
+      currentTurnFirstContentAt = undefined
+      if (hadToolCalls || firstContentAt === undefined) return
+
+      const idx = getAiMessageIndex()
+      const message = targetBuffer.messages[idx]
+      if (!message || message.processDurationMs !== undefined) return
+      const durationMs = Math.max(0, firstContentAt - message.timestamp)
+      applyAIMessageUpdates({
+        processDurationMs: durationMs,
+        contentBlocks: persistProcessDurationMs(message.contentBlocks, durationMs),
+      })
+    }
+
+    const appendThinkToBlocks = (text: string, tag?: string, durationMs?: number) => {
+      if (!text && durationMs === undefined) return
+      if (text) streamTextBatcher.push({ type: 'think', content: text, thinkTag: tag })
+      if (durationMs !== undefined) {
+        flushPendingText()
+        appendThinkImmediately('', tag, durationMs)
+      }
     }
 
     const appendChartsToBlocks = (charts: ChartPayload[]) => {
       if (charts.length === 0) return
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       blocks.push(...toChartContentBlocks(charts))
@@ -753,6 +795,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
     }
 
     const appendEvidenceToBlocks = (evidence: ChatEvidencePayload) => {
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       blocks.push(toEvidenceContentBlock(evidence))
@@ -761,24 +804,28 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
 
     const appendPlanDraftToBlocks = (delta: string) => {
       if (!delta) return
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       updateAIMessage({ contentBlocks: [...appendPlanDraftDelta(blocks, delta)] })
     }
 
     const appendPlanToBlocks = (plan: PlanContentBlock) => {
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       updateAIMessage({ contentBlocks: [...replacePlanDraftWithPlan(blocks, plan)] })
     }
 
     const removePlanDraftsFromBlocks = () => {
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       updateAIMessage({ contentBlocks: [...removePlanDraftBlocks(blocks)] })
     }
 
     const updatePlanBlockStatus = (status: PlanBlockStatus) => {
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       for (let index = blocks.length - 1; index >= 0; index--) {
@@ -792,6 +839,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
     }
 
     const appendErrorToBlocks = (error: SerializedErrorInfo) => {
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       blocks.push({ type: 'error', error })
@@ -799,6 +847,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
     }
 
     const addToolBlock = (toolName: string, params?: Record<string, unknown>, toolCallId?: string) => {
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       blocks.push({
@@ -811,6 +860,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
     const addRenderOnlyToolPendingBlock = (toolName: string, params?: Record<string, unknown>, toolCallId?: string) => {
       const pendingBlock = createRenderOnlyToolPendingBlock(toolName, params, toolCallId)
       if (!pendingBlock) return
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       blocks.push(pendingBlock)
@@ -823,6 +873,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
       charts: ChartPayload[],
       errorBlock: ReturnType<typeof toRenderOnlyToolErrorBlock>
     ) => {
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
 
@@ -836,6 +887,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
       status: 'done' | 'error',
       completion?: { toolCallId?: string; result?: string; displayResult?: string; isError?: boolean }
     ) => {
+      flushPendingText()
       const idx = getAiMessageIndex()
       const blocks = targetBuffer.messages[idx].contentBlocks || []
       const isRunningTool = (block: ContentBlock): block is Extract<ContentBlock, { type: 'tool' }> =>
@@ -886,6 +938,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
       updateToolBlockStatus,
       completeTurn,
       settleProcessDuration,
+      flushPendingText,
     }
   }
 
@@ -1013,6 +1066,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
         updateToolBlockStatus,
         completeTurn,
         settleProcessDuration,
+        flushPendingText,
       } = createStreamBlockHelpers(targetBuffer, () => aiMessageIndex)
 
       const currentAssistantId = targetBuffer.assistantId ?? getDefaultGeneralAssistantId(state.locale)
@@ -1113,6 +1167,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
               break
 
             case 'tool_update':
+              flushPendingText()
               if (chunk.toolName && chunk.toolProgress) {
                 state.currentToolStatus = toolLifecycle.update({
                   name: chunk.toolName,
@@ -1187,12 +1242,14 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
               break
 
             case 'status':
+              flushPendingText()
               if (chunk.status && (!state.agentStatus || chunk.status.updatedAt >= state.agentStatus.updatedAt)) {
                 state.agentStatus = chunk.status
               }
               break
 
             case 'compression_done':
+              flushPendingText()
               if (chunk.compressionResult) {
                 const summaryMsg: ChatMessage = {
                   id: `summary-${Date.now()}`,
@@ -1257,6 +1314,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
       setActiveTaskMeta(chatKey, content, agentReqId, resolvedAIChatId)
 
       const result = await agentPromise
+      flushPendingText()
       if (state.isAborted) {
         clearActiveTask(chatKey, agentReqId)
         return { success: false, reason: 'aborted' }
@@ -1635,6 +1693,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
       updateToolBlockStatus,
       completeTurn,
       settleProcessDuration,
+      flushPendingText,
     } = createStreamBlockHelpers(targetBuffer, () => aiMessageIndex)
 
     try {
@@ -1726,6 +1785,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
               }
               break
             case 'tool_update':
+              flushPendingText()
               if (chunk.toolName && chunk.toolProgress) {
                 state.currentToolStatus = toolLifecycle.update({
                   name: chunk.toolName,
@@ -1786,6 +1846,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
               completeTurn(Boolean(chunk.hadToolCalls))
               break
             case 'status':
+              flushPendingText()
               if (chunk.status && (!state.agentStatus || chunk.status.updatedAt >= state.agentStatus.updatedAt)) {
                 state.agentStatus = chunk.status
               }
@@ -1823,6 +1884,7 @@ export const useAIChatStore = defineStore('aiChatRuntime', () => {
       setActiveTaskMeta(chatKey, content, agentReqId, state.currentAIChatId)
 
       const result = await agentPromise
+      flushPendingText()
       if (state.isAborted) {
         restoreOriginal()
         clearActiveTask(chatKey, agentReqId)
diff --git a/src/stores/aiChatStreamBatcher.test.ts b/src/stores/aiChatStreamBatcher.test.ts
new file mode 100644
index 00000000..0e681908
--- /dev/null
+++ b/src/stores/aiChatStreamBatcher.test.ts
@@ -0,0 +1,74 @@
+import assert from 'node:assert/strict'
+import test from 'node:test'
+import { createAIStreamTextBatcher, type AIStreamTextDelta } from './aiChatStreamBatcher'
+
+function createManualScheduler() {
+  let nextHandle = 1
+  const callbacks = new Map<number, () => void>()
+  return {
+    scheduler: {
+      request(callback: () => void) {
+        const handle = nextHandle++
+        callbacks.set(handle, callback)
+        return handle
+      },
+      cancel(handle: number) {
+        callbacks.delete(handle)
+      },
+    },
+    runFrame() {
+      const pending = [...callbacks.values()]
+      callbacks.clear()
+      pending.forEach((callback) => callback())
+    },
+    pendingCount() {
+      return callbacks.size
+    },
+  }
+}
+
+test('merges adjacent text and thinking deltas within one frame', () => {
+  const frame = createManualScheduler()
+  const applied: AIStreamTextDelta[][] = []
+  const batcher = createAIStreamTextBatcher((deltas) => applied.push(deltas), frame.scheduler)
+
+  batcher.push({ type: 'content', content: 'A' })
+  batcher.push({ type: 'content', content: 'B' })
+  batcher.push({ type: 'think', thinkTag: 'analysis', content: 'C' })
+  batcher.push({ type: 'think', thinkTag: 'analysis', content: 'D' })
+  batcher.push({ type: 'content', content: 'E' })
+
+  assert.equal(frame.pendingCount(), 1)
+  frame.runFrame()
+  assert.deepEqual(applied, [
+    [
+      { type: 'content', content: 'AB' },
+      { type: 'think', thinkTag: 'analysis', content: 'CD' },
+      { type: 'content', content: 'E' },
+    ],
+  ])
+})
+
+test('flushes queued deltas synchronously before a stream boundary', () => {
+  const frame = createManualScheduler()
+  const applied: AIStreamTextDelta[][] = []
+  const batcher = createAIStreamTextBatcher((deltas) => applied.push(deltas), frame.scheduler)
+
+  batcher.push({ type: 'content', content: 'before tool' })
+  batcher.flush()
+
+  assert.equal(frame.pendingCount(), 0)
+  assert.deepEqual(applied, [[{ type: 'content', content: 'before tool' }]])
+})
+
+test('cancels the pending frame without applying stale deltas', () => {
+  const frame = createManualScheduler()
+  const applied: AIStreamTextDelta[][] = []
+  const batcher = createAIStreamTextBatcher((deltas) => applied.push(deltas), frame.scheduler)
+
+  batcher.push({ type: 'content', content: 'stale' })
+  batcher.cancel()
+  frame.runFrame()
+
+  assert.deepEqual(applied, [])
+})
diff --git a/src/stores/aiChatStreamBatcher.ts b/src/stores/aiChatStreamBatcher.ts
new file mode 100644
index 00000000..02b4f2be
--- /dev/null
+++ b/src/stores/aiChatStreamBatcher.ts
@@ -0,0 +1,74 @@
+export type AIStreamTextDelta =
+  | { type: 'content'; content: string }
+  | { type: 'think'; content: string; thinkTag?: string }
+
+interface FrameScheduler {
+  request: (callback: () => void) => number
+  cancel: (handle: number) => void
+}
+
+function createDefaultFrameScheduler(): FrameScheduler {
+  if (typeof requestAnimationFrame === 'function' && typeof cancelAnimationFrame === 'function') {
+    return {
+      request: (callback) => requestAnimationFrame(callback),
+      cancel: (handle) => cancelAnimationFrame(handle),
+    }
+  }
+
+  return {
+    request: (callback) => setTimeout(callback, 16) as unknown as number,
+    cancel: (handle) => clearTimeout(handle),
+  }
+}
+
+export function createAIStreamTextBatcher(
+  apply: (deltas: AIStreamTextDelta[]) => void,
+  scheduler: FrameScheduler = createDefaultFrameScheduler()
+): {
+  push: (delta: AIStreamTextDelta) => void
+  flush: () => void
+  cancel: () => void
+} {
+  let pendingFrame: number | null = null
+  let pendingDeltas: AIStreamTextDelta[] = []
+
+  function flush() {
+    if (pendingFrame !== null) {
+      scheduler.cancel(pendingFrame)
+      pendingFrame = null
+    }
+    if (pendingDeltas.length === 0) return
+
+    const deltas = pendingDeltas
+    pendingDeltas = []
+    apply(deltas)
+  }
+
+  function push(delta: AIStreamTextDelta) {
+    const previous = pendingDeltas[pendingDeltas.length - 1]
+    if (previous?.type === 'content' && delta.type === 'content') {
+      previous.content += delta.content
+    } else if (previous?.type === 'think' && delta.type === 'think' && previous.thinkTag === delta.thinkTag) {
+      previous.content += delta.content
+    } else {
+      pendingDeltas.push({ ...delta })
+    }
+
+    if (pendingFrame === null) {
+      pendingFrame = scheduler.request(() => {
+        pendingFrame = null
+        flush()
+      })
+    }
+  }
+
+  function cancel() {
+    if (pendingFrame !== null) {
+      scheduler.cancel(pendingFrame)
+      pendingFrame = null
+    }
+    pendingDeltas = []
+  }
+
+  return { push, flush, cancel }
+}
diff --git a/tests/ai-chat-performance/BenchmarkApp.vue b/tests/ai-chat-performance/BenchmarkApp.vue
new file mode 100644
index 00000000..ee200e09
--- /dev/null
+++ b/tests/ai-chat-performance/BenchmarkApp.vue
@@ -0,0 +1,216 @@
+<script setup lang="ts">
+import { computed, nextTick, onMounted, ref } from 'vue'
+import ChatMessage from '@/components/AIChat/chat/ChatMessage.vue'
+import { useProgressiveChatHistory } from '@/components/AIChat/composables/useProgressiveChatHistory'
+import { groupMessagesToQAPairs } from '@/components/AIChat/utils/chatMessages'
+import type { ChatMessage as ChatMessageData, ContentBlock } from '@/stores/aiChat'
+import { createAIStreamTextBatcher } from '@/stores/aiChatStreamBatcher'
+
+interface BenchmarkReport {
+  pairs: number
+  mountedPairs: number
+  blocksPerAnswer: number
+  answerCharacters: number
+  initialRenderMs: number
+  domNodes: number
+  stream?: {
+    chunks: number
+    intervalMs: number
+    batched: boolean
+    renderUpdates: number
+    durationMs: number
+    longTasks: number
+    longTaskDurationMs: number
+  }
+}
+
+declare global {
+  interface Window {
+    __AI_CHAT_BENCHMARK_STARTED_AT__: number
+    __AI_CHAT_BENCHMARK__?: BenchmarkReport
+  }
+}
+
+const params = new URLSearchParams(window.location.search)
+const pairCount = readPositiveInteger('pairs', 100)
+const blocksPerAnswer = readPositiveInteger('blocks', 10)
+const answerCharacters = readPositiveInteger('chars', 10_000)
+const streamChunks = readPositiveInteger('streamChunks', 0)
+const streamIntervalMs = readPositiveInteger('streamInterval', 4)
+const useStreamBatcher = params.get('batch') === '1'
+const useProgressiveHistory = params.get('progressive') === '1'
+const report = ref<BenchmarkReport | null>(null)
+const messages = ref(createMessages(pairCount, blocksPerAnswer, answerCharacters))
+const qaPairs = computed(() => groupMessagesToQAPairs(messages.value))
+const benchmarkConversationId = ref<string | null>('benchmark')
+const benchmarkScrollContainer = ref<HTMLElement | null>(null)
+const progressiveHistory = useProgressiveChatHistory(qaPairs, benchmarkConversationId, benchmarkScrollContainer)
+const renderedPairs = computed(() => (useProgressiveHistory ? progressiveHistory.visiblePairs.value : qaPairs.value))
+
+function readPositiveInteger(key: string, fallback: number): number {
+  const value = Number.parseInt(params.get(key) ?? '', 10)
+  return Number.isInteger(value) && value > 0 ? value : fallback
+}
+
+function buildText(length: number, seed: number): string {
+  const paragraph = `第 ${seed + 1} 段合成回答只用于性能测试，不包含任何用户数据。这里包含 **Markdown**、列表和换行。\n\n- 项目一\n- 项目二\n\n`
+  return paragraph.repeat(Math.ceil(length / paragraph.length)).slice(0, length)
+}
+
+function createBlocks(blockCount: number, totalCharacters: number, seed: number): ContentBlock[] {
+  const textPerBlock = Math.max(1, Math.floor(totalCharacters / blockCount))
+  return Array.from({ length: blockCount }, (_, index): ContentBlock => {
+    const text = buildText(textPerBlock, seed * blockCount + index)
+    if (index % 3 === 0) return { type: 'think', tag: 'analysis', text, durationMs: 500 }
+    if (index % 3 === 1) {
+      return {
+        type: 'tool',
+        tool: {
+          name: 'benchmark_tool',
+          displayName: 'benchmark_tool',
+          status: 'done',
+          result: text,
+          displayResult: text,
+        },
+      }
+    }
+    return { type: 'text', text }
+  })
+}
+
+function createMessages(pairs: number, blockCount: number, characters: number): ChatMessageData[] {
+  const result: ChatMessageData[] = []
+  const timestamp = Date.now()
+  for (let index = 0; index < pairs; index += 1) {
+    result.push({
+      id: `user-${index}`,
+      role: 'user',
+      content: `这是第 ${index + 1} 个合成问题。`,
+      timestamp: timestamp + index * 2,
+    })
+    result.push({
+      id: `assistant-${index}`,
+      role: 'assistant',
+      content: buildText(characters, index),
+      contentBlocks: createBlocks(blockCount, characters, index),
+      processDurationMs: 1_500,
+      timestamp: timestamp + index * 2 + 1,
+    })
+  }
+  return result
+}
+
+onMounted(async () => {
+  await nextTick()
+  await new Promise<void>((resolve) => requestAnimationFrame(() => requestAnimationFrame(() => resolve())))
+  const result: BenchmarkReport = {
+    pairs: pairCount,
+    mountedPairs: renderedPairs.value.length,
+    blocksPerAnswer,
+    answerCharacters,
+    initialRenderMs: performance.now() - window.__AI_CHAT_BENCHMARK_STARTED_AT__,
+    domNodes: document.getElementsByTagName('*').length,
+  }
+  report.value = result
+  window.__AI_CHAT_BENCHMARK__ = result
+
+  if (streamChunks > 0) {
+    await runStreamingBenchmark(result)
+  }
+})
+
+async function runStreamingBenchmark(result: BenchmarkReport): Promise<void> {
+  const longTaskDurations: number[] = []
+  const observer =
+    typeof PerformanceObserver === 'undefined'
+      ? null
+      : new PerformanceObserver((entries) => {
+          longTaskDurations.push(...entries.getEntries().map((entry) => entry.duration))
+        })
+  observer?.observe({ type: 'longtask', buffered: false })
+
+  const assistantIndex = messages.value.length - 1
+  const assistant = messages.value[assistantIndex]
+  const initialBlocks = assistant.contentBlocks ?? []
+  messages.value[assistantIndex] = {
+    ...assistant,
+    isStreaming: true,
+    contentBlocks: [...initialBlocks, { type: 'text', text: '' }],
+  }
+  await nextTick()
+
+  const chunkText = '合成流式内容 **Markdown**。'
+  const startedAt = performance.now()
+  let emitted = 0
+  let renderUpdates = 0
+  const applyText = (text: string) => {
+    const current = messages.value[assistantIndex]
+    const blocks = current.contentBlocks ?? []
+    const lastBlock = blocks[blocks.length - 1]
+    if (lastBlock?.type === 'text') lastBlock.text += text
+    messages.value[assistantIndex] = {
+      ...current,
+      content: current.content + text,
+      contentBlocks: [...blocks],
+    }
+    renderUpdates += 1
+  }
+  const batcher = createAIStreamTextBatcher((deltas) => {
+    applyText(deltas.map((delta) => delta.content).join(''))
+  })
+  await new Promise<void>((resolve) => {
+    const timer = window.setInterval(() => {
+      if (useStreamBatcher) batcher.push({ type: 'content', content: chunkText })
+      else applyText(chunkText)
+      emitted += 1
+      if (emitted < streamChunks) return
+      window.clearInterval(timer)
+      batcher.flush()
+      resolve()
+    }, streamIntervalMs)
+  })
+
+  await nextTick()
+  await new Promise<void>((resolve) => requestAnimationFrame(() => requestAnimationFrame(() => resolve())))
+  observer?.disconnect()
+  result.stream = {
+    chunks: streamChunks,
+    intervalMs: streamIntervalMs,
+    batched: useStreamBatcher,
+    renderUpdates,
+    durationMs: performance.now() - startedAt,
+    longTasks: longTaskDurations.length,
+    longTaskDurationMs: longTaskDurations.reduce((sum, duration) => sum + duration, 0),
+  }
+  report.value = { ...result }
+  window.__AI_CHAT_BENCHMARK__ = report.value
+}
+</script>
+
+<template>
+  <UApp>
+    <main class="mx-auto max-w-4xl px-6 py-8">
+      <pre id="benchmark-report" class="mb-8 rounded-lg bg-gray-100 p-4 text-xs dark:bg-gray-900">{{ report }}</pre>
+      <template v-for="pair in renderedPairs" :key="pair.id">
+        <div v-if="!pair.standalone" class="space-y-6 pb-4">
+          <ChatMessage
+            v-if="pair.user"
+            :message-id="pair.user.id"
+            :role="pair.user.role"
+            :content="pair.user.content"
+            :timestamp="pair.user.timestamp"
+          />
+          <ChatMessage
+            v-if="pair.assistant"
+            :message-id="pair.assistant.id"
+            :role="pair.assistant.role"
+            :content="pair.assistant.content"
+            :content-blocks="pair.assistant.contentBlocks"
+            :process-duration-ms="pair.assistant.processDurationMs"
+            :timestamp="pair.assistant.timestamp"
+          />
+        </div>
+      </template>
+    </main>
+  </UApp>
+</template>
diff --git a/tests/ai-chat-performance/index.html b/tests/ai-chat-performance/index.html
new file mode 100644
index 00000000..fd73578e
--- /dev/null
+++ b/tests/ai-chat-performance/index.html
@@ -0,0 +1,15 @@
+<!doctype html>
+<html lang="zh-CN">
+  <head>
+    <meta charset="UTF-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>AI Chat Performance Benchmark</title>
+    <script>
+      window.__AI_CHAT_BENCHMARK_STARTED_AT__ = performance.now()
+    </script>
+  </head>
+  <body>
+    <div id="app"></div>
+    <script type="module" src="/main.ts"></script>
+  </body>
+</html>
diff --git a/tests/ai-chat-performance/main.ts b/tests/ai-chat-performance/main.ts
new file mode 100644
index 00000000..71f25caf
--- /dev/null
+++ b/tests/ai-chat-performance/main.ts
@@ -0,0 +1,13 @@
+import '@/icons/disable-iconify-api'
+import 'virtual:nuxt-icon-bundle/register'
+
+import { createApp } from 'vue'
+import ui from '@nuxt/ui/vue-plugin'
+import { createMemoryHistory, createRouter } from 'vue-router'
+import i18n from '@/i18n'
+import BenchmarkApp from './BenchmarkApp.vue'
+import '@/assets/styles/main.css'
+
+const router = createRouter({ history: createMemoryHistory(), routes: [] })
+
+createApp(BenchmarkApp).use(router).use(ui).use(i18n).mount('#app')
diff --git a/tests/ai-chat-performance/vite.config.mts b/tests/ai-chat-performance/vite.config.mts
new file mode 100644
index 00000000..4bcc2de2
--- /dev/null
+++ b/tests/ai-chat-performance/vite.config.mts
@@ -0,0 +1,30 @@
+import { resolve } from 'node:path'
+import { defineConfig } from 'vite'
+import vue from '@vitejs/plugin-vue'
+import ui from '@nuxt/ui/vite'
+import { chatlabIconBundle } from '../../vite.icon-bundle.config'
+
+const repositoryRoot = resolve(import.meta.dirname, '../..')
+
+export default defineConfig({
+  root: import.meta.dirname,
+  resolve: {
+    alias: {
+      '@': resolve(repositoryRoot, 'src'),
+      '~': resolve(repositoryRoot, 'src'),
+      '@openchatlab': resolve(repositoryRoot, 'packages'),
+      '@electron/shared': resolve(repositoryRoot, 'apps/desktop/shared'),
+      '@electron/preload': resolve(repositoryRoot, 'apps/desktop/preload'),
+    },
+  },
+  define: {
+    __IS_ELECTRON__: JSON.stringify(false),
+    __IS_WEB_WASM__: JSON.stringify(false),
+    __APP_VERSION__: JSON.stringify('benchmark'),
+  },
+  plugins: [vue(), ui({ dts: false }), chatlabIconBundle(repositoryRoot)],
+  server: {
+    host: '127.0.0.1',
+    port: 3190,
+  },
+})