diff --git a/packages/agentlab/src/catalog.ts b/packages/agentlab/src/catalog.ts
index 9369845..a0e25da 100644
--- a/packages/agentlab/src/catalog.ts
+++ b/packages/agentlab/src/catalog.ts
@@ -26,6 +26,8 @@ export const AGENTLAB_PROVIDER_CATALOG: AgentLabProviderCatalogEntry[] = [
     models: [
       { id: 'deepseek-chat', label: 'deepseek-chat', capabilities: ['chat'] },
       { id: 'deepseek-reasoner', label: 'deepseek-reasoner', capabilities: ['chat'] },
+      { id: 'deepseek-v4-flash', label: 'deepseek-v4-flash', capabilities: ['chat'] },
+      { id: 'deepseek-v4-pro', label: 'deepseek-v4-pro', capabilities: ['chat'] },
     ],
   },
   {
diff --git a/packages/agentlab/src/extract.ts b/packages/agentlab/src/extract.ts
index b629cb4..77a25be 100644
--- a/packages/agentlab/src/extract.ts
+++ b/packages/agentlab/src/extract.ts
@@ -12,7 +12,7 @@ import type {
   AgentLabExpression,
   AgentLabFewShotPair,
 } from './types';
-import { reportUsage } from './http';
+import { reportUsage, pickMessageText } from './http';
 
 function coerceString(value: unknown): string {
   if (typeof value === 'string') return value.trim();
@@ -63,9 +63,9 @@ async function chatCompletion(
   if (!res.ok) {
     throw new Error(`AgentLab 提炼接口调用失败: HTTP ${res.status}`);
   }
-  const data = (await res.json()) as { choices?: Array<{ message?: { content?: string } }> };
+  const data = (await res.json()) as { choices?: Array<{ message?: { content?: string; reasoning_content?: string } }> };
   reportUsage(endpoint, data);
-  return data.choices?.[0]?.message?.content ?? '';
+  return pickMessageText(data.choices?.[0]?.message);
 }
 
 /**
@@ -103,9 +103,9 @@ async function visionCompletion(
   if (!res.ok) {
     throw new Error(`AgentLab 视觉接口调用失败: HTTP ${res.status}`);
   }
-  const data = (await res.json()) as { choices?: Array<{ message?: { content?: string } }> };
+  const data = (await res.json()) as { choices?: Array<{ message?: { content?: string; reasoning_content?: string } }> };
   reportUsage(endpoint, data);
-  return data.choices?.[0]?.message?.content ?? '';
+  return pickMessageText(data.choices?.[0]?.message);
 }
 
 /** generateText + 宽松解析 + 失败重试一次。 */
diff --git a/packages/agentlab/src/http.ts b/packages/agentlab/src/http.ts
index 442b6fb..5c86391 100644
--- a/packages/agentlab/src/http.ts
+++ b/packages/agentlab/src/http.ts
@@ -345,20 +345,31 @@ async function postJson<T>(
   return (await res.json()) as T;
 }
 
+/** OpenAI 兼容消息里同时可能存在 content 和 reasoning_content（推理模型专用），取首个非空。 */
+export function pickMessageText(msg: { content?: unknown; reasoning_content?: unknown } | undefined): string {
+  if (!msg) return '';
+  const c = typeof msg.content === 'string' ? msg.content : '';
+  const r = typeof msg.reasoning_content === 'string' ? msg.reasoning_content : '';
+  return (c || r).trim();
+}
+
 /**
  * 设置页「测试连通性」：用最小的 chat 请求探活，返回 ok + 详细错误（含 HTTP 状态码与响应体）。
  * 不抛错——把失败包装成 { ok:false, error }，让前端直接展示「模型不对 / key 不对」。
+ * 推理模型（如 deepseek-v4 系列）会把思考过程放在 reasoning_content 中，
+ * 当 max_tokens 较小时 content 可能为空——因此同时检查两个字段。
  */
 export async function testChatEndpoint(
   endpoint: AgentLabEndpoint,
 ): Promise<{ ok: boolean; error?: string; reply?: string }> {
   try {
-    const data = await postJson<{ choices?: Array<{ message?: { content?: string } }> }>(
+    const data = await postJson<{ choices?: Array<{ message?: { content?: string; reasoning_content?: string } }> }>(
       endpoint,
       '/chat/completions',
-      { model: endpoint.model, messages: [{ role: 'user', content: '你好' }], max_tokens: 16 },
+      // 推理模型需要更多 token 预算才能产出 content；64 足够短路"你好"又不浪费。
+      { model: endpoint.model, messages: [{ role: 'user', content: '你好' }], max_tokens: 64 },
     );
-    const reply = data.choices?.[0]?.message?.content?.trim();
+    const reply = pickMessageText(data.choices?.[0]?.message);
     if (!reply) return { ok: false, error: '接口可达，但返回内容为空（模型可能不支持 chat/completions）。' };
     return { ok: true, reply };
   } catch (e) {
@@ -564,7 +575,7 @@ export async function runPersonaChat(
   ];
 
   const data = await postJson<{
-    choices?: Array<{ message?: { content?: string } }>;
+    choices?: Array<{ message?: { content?: string; reasoning_content?: string } }>;
     usage?: OpenAiUsage;
   }>(chat, '/chat/completions', {
     model: chat.model,
@@ -573,7 +584,7 @@ export async function runPersonaChat(
     messages,
   });
   reportUsage(chat, data);
-  const raw = data.choices?.[0]?.message?.content?.trim();
+  const raw = pickMessageText(data.choices?.[0]?.message);
   if (!raw) {
     throw new Error('AgentLab chat 返回为空');
   }
diff --git a/packages/agentlab/src/index.ts b/packages/agentlab/src/index.ts
index 2b6176b..4c61ea2 100644
--- a/packages/agentlab/src/index.ts
+++ b/packages/agentlab/src/index.ts
@@ -5,7 +5,7 @@ export {
   modelsWithCapability,
   resolveEndpoint,
 } from './provider';
-export { embedTexts, runPersonaChat, reportUsage, keywordsOf, testChatEndpoint } from './http';
+export { embedTexts, runPersonaChat, reportUsage, keywordsOf, testChatEndpoint, pickMessageText } from './http';
 export { selectStickerByEmotion } from './sticker';
 export { humanizeText, DEFAULT_TYPO_INTENSITY } from './typo';
 export { scoreReplyWillingness } from './willing';
diff --git a/packages/service/src/account/assistant.ts b/packages/service/src/account/assistant.ts
index 86723b2..4289ef4 100644
--- a/packages/service/src/account/assistant.ts
+++ b/packages/service/src/account/assistant.ts
@@ -13,7 +13,7 @@
 import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';
 import { randomUUID } from 'node:crypto';
 import { basename, dirname, extname, join, resolve, sep } from 'node:path';
-import { reportUsage, type AgentLabEndpoint, type AgentLabModelRef, type AgentLabUsage } from '@weq/agentlab';
+import { reportUsage, pickMessageText, type AgentLabEndpoint, type AgentLabModelRef, type AgentLabUsage } from '@weq/agentlab';
 import type { TokenUsageStore } from './agentlab_usage';
 import type { ConversationStore, ConversationTurn } from './agentlab_conversation';
 
@@ -384,8 +384,8 @@ export class AssistantService {
       ],
       [],
     );
-    const raw = data.choices?.[0]?.message?.content;
-    return cleanTitle(typeof raw === 'string' ? raw : '');
+    const raw = pickMessageText(data.choices?.[0]?.message);
+    return cleanTitle(raw);
   }
 
   /** 核心多轮循环：返回最终文本。中途只 emit 过程 step，不 emit final。 */
@@ -424,7 +424,7 @@ export class AssistantService {
       const msg = data.choices?.[0]?.message;
       if (!msg) throw new Error('助手返回为空');
 
-      const content = typeof msg.content === 'string' ? msg.content.trim() : '';
+      const content = pickMessageText(msg);
 
       if (allowTools && msg.tool_calls?.length) {
         // 模型在调用工具前给出的思路 → 作为"思考"展示。