diff --git a/backend/app/model_services/chat_services.py b/backend/app/model_services/chat_services.py
index 552a74b..ae5f2fc 100644
--- a/backend/app/model_services/chat_services.py
+++ b/backend/app/model_services/chat_services.py
@@ -54,21 +54,42 @@ class LocalVLLMChatProvider(BaseServiceProvider[BaseChatModel]):
             return False
         
         try:
-            # 尝试创建一个简单的测试调用
-            from langchain_openai import ChatOpenAI
-            from pydantic import SecretStr
+            # 先测试主机名能否解析
+            import httpx
+            from urllib.parse import urlparse
             
-            llm = ChatOpenAI(
-                base_url=VLLM_BASE_URL,
-                api_key=SecretStr(LLM_API_KEY) if LLM_API_KEY else SecretStr("dummy"),
-                model=self._model,
-                timeout=10.0,
-                max_retries=1,
-            )
+            parsed_url = urlparse(VLLM_BASE_URL)
+            host = parsed_url.hostname
+            port = parsed_url.port or (80 if parsed_url.scheme == 'http' else 443)
             
-            # 简单的 ping 测试（不实际调用模型）
-            logger.info(f"本地 VLLM 服务配置正确，准备使用: {self._model}")
-            return True
+            # 测试能否建立 TCP 连接（快速失败）
+            import socket
+            sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+            sock.settimeout(2.0)
+            try:
+                sock.connect((host, port))
+                sock.close()
+            except Exception as e:
+                logger.warning(f"本地 VLLM 服务无法连接: {host}:{port} - {e}")
+                return False
+            
+            # 再尝试调用简单的 API（比如 models 接口）
+            client = httpx.Client(base_url=VLLM_BASE_URL.rstrip('/'), timeout=5.0)
+            headers = {}
+            if LLM_API_KEY:
+                headers["Authorization"] = f"Bearer {LLM_API_KEY}"
+            
+            try:
+                response = client.get("/v1/models", headers=headers)
+                if response.status_code == 200:
+                    logger.info(f"本地 VLLM 服务可用: {self._model}")
+                    return True
+            except Exception:
+                pass
+            
+            # 如果 /v1/models 失败，也认为服务不可用
+            logger.warning(f"本地 VLLM 服务响应异常")
+            return False
         except Exception as e:
             logger.warning(f"本地 VLLM 服务不可用: {e}")
             return False
diff --git a/backend/app/model_services/rerank_services.py b/backend/app/model_services/rerank_services.py
index ecd5855..4cc2052 100644
--- a/backend/app/model_services/rerank_services.py
+++ b/backend/app/model_services/rerank_services.py
@@ -277,22 +277,10 @@ class ZhipuRerankProvider(BaseServiceProvider[BaseRerankService]):
     def is_available(self) -> bool:
         """
         检查智谱 API 重排服务是否可用
+        注意：zhipuai 库 2.0.1 版本没有 rerank API，直接返回 False，降级到 LLM 方案
         """
-        if not ZHIPUAI_API_KEY:
-            logger.warning("ZHIPUAI_API_KEY 未配置")
-            return False
-
-        try:
-            service = ZhipuRerankService(model=self._model)
-            test_scores = service.compute_scores("test query", ["test document"])
-            logger.info(f"智谱重排服务可用")
-            return True
-        except ImportError:
-            logger.warning("zhipuai 库未安装")
-            return False
-        except Exception as e:
-            logger.warning(f"智谱重排服务不可用: {e}")
-            return False
+        logger.warning("智谱 rerank API 在当前 zhipuai 库版本不可用，降级到 LLM 方案")
+        return False
 
     def get_service(self) -> BaseRerankService:
         """