diff --git a/backend/server.js b/backend/server.js
index 51a61b9..6ad474c 100644
--- a/backend/server.js
+++ b/backend/server.js
@@ -227,7 +227,7 @@ async function searchWeb(query, maxResults = 3) {
 
 const LLM_API_URL = process.env.LLM_API_URL || 'https://api.deepseek.com/v1/chat/completions';
 const LLM_API_KEY = process.env.LLM_API_KEY || '';
-const LLM_MODEL = process.env.LLM_MODEL || 'deepseek-chat';
+const LLM_MODEL = process.env.LLM_MODEL || 'deepseek-v4-flash';
 
 async function callLLM(systemPrompt, userPrompt) {
     if (!LLM_API_KEY) {
diff --git a/functions/api/chat.js b/functions/api/chat.js
index 8f64be6..0d67298 100644
--- a/functions/api/chat.js
+++ b/functions/api/chat.js
@@ -49,11 +49,10 @@ export async function onRequestPost(context) {
                 .slice(0, 5)
                 .map(s => s.row);
 
-                bestScore = scored.length > 0 ? scored[0].score : 0;
+            bestScore = scored.length > 0 ? scored[0].score : 0;
 
-                if (bestScore >= 30 && scored[0].row.answer) {
-                    sources = [scored[0].row.id];
-                }
+            if (bestScore >= 30 && scored[0].row.answer) {
+                sources = [scored[0].row.id];
             }
         }
 
@@ -63,7 +62,7 @@ export async function onRequestPost(context) {
             webResults = await searchWeb(env, question, 3);
         }
 
-        if (bestScore >= 60 && qaContext.length > 0) {
+        if (bestScore >= 30 && qaContext.length > 0) {
             answer = qaContext[0].answer;
         } else if (env.LLM_API_KEY && env.LLM_API_URL) {
             try {
@@ -79,8 +78,8 @@ export async function onRequestPost(context) {
 
         if (!answer) {
             answer = env.LLM_API_KEY
-                ? '\u62b1\u6b49\uff0c\u668a\u80fd\u95ee\u7b54\u670d\u52a1\u6682\u65f6\u4e0d\u53ef\u7528\uff0c\u8bf7\u7a0d\u540e\u91cd\u8bd5\u3002'
-                : '\u667a\u80fd\u95ee\u7b54\u670d\u52a1\u6682\u672a\u914d\u7f6e\uff0c\u8bf7\u8054\u7cfb\u7ba1\u7406\u5458\u3002';
+                ? '抱歉，智能问答服务暂时不可用，请稍后重试。'
+                : '智能问答服务尚未配置，请联系管理员。';
         }
 
         if (db) {
@@ -99,7 +98,7 @@ export async function onRequestPost(context) {
         return jsonResponse({ answer, sources });
     } catch (err) {
         return jsonResponse({
-            answer: '\u62b1\u6b49\uff0c\u667a\u80fd\u95ee\u7b54\u670d\u52a1\u6682\u65f6\u4e0d\u53ef\u7528\u3002',
+            answer: '抱歉，智能问答服务暂时不可用。',
             error: err.message,
         }, 500);
     }
@@ -127,18 +126,18 @@ export async function onRequestOptions() {
 }
 
 const STOP_WORDS = new Set([
-    '\u7684', '\u4e86', '\u5728', '\u662f', '\u6211', '\u6709', '\u548c', '\u5c31', '\u4e0d',
-    '\u4eba', '\u90fd', '\u4e00', '\u4e0a', '\u4e5f', '\u5f88', '\u5230',
-    '\u8bf4', '\u8981', '\u53bb', '\u4f60', '\u4f1a', '\u7740', '\u6ca1\u6709', '\u770b',
-    '\u597d', '\u81ea\u5df1', '\u8fd9', '\u4ed6', '\u5979', '\u5b83', '\u4eec', '\u90a3',
-    '\u4e9b', '\u5417', '\u554a', '\u5462', '\u5427', '\u55ef', '\u54e6',
-    '\u600e\u4e48', '\u4ec0\u4e48', '\u5982\u4f55', '\u54ea\u91cc', '\u54ea\u4e2a', '\u54ea\u4e9b',
-    '\u4f55\u65f6', '\u591a\u5c11', '\u51e0', '\u5565', '\u548b', '\u4e3a\u5565', '\u4e3a\u4ec0\u4e48',
-    '\u8bf7\u95ee', '\u8bf7',
+    '的', '了', '在', '是', '我', '有', '和', '就', '不',
+    '人', '都', '一', '上', '也', '很', '到',
+    '说', '要', '去', '你', '会', '着', '没有', '看',
+    '好', '自己', '这', '他', '她', '它', '们', '那',
+    '些', '吗', '啊', '呢', '吧', '嗯', '哦',
+    '怎么', '什么', '如何', '哪里', '哪个', '哪些',
+    '何时', '多少', '几', '啥', '咋', '为啥', '为什么',
+    '请问', '请',
 ]);
 
 function tokenize(text) {
-    const raw = text.toLowerCase().split(/[\s,.\u3002\uff0c\uff01\uff1f\u3001\uff1b\uff1a\u201c\u201d\u2018\u2019\uff08\uff09()\u3010\u3011\u300a\u300b/\\|]+/);
+    const raw = text.toLowerCase().split(/[\s,.。，！？、；：“”‘’（）()【】《》/\\|]+/);
     const tokens = [];
     for (const token of raw) {
         if (/[\u4e00-\u9fff]/.test(token)) {