diff --git a/app/api/endpoints/llm.py b/app/api/endpoints/llm.py
index 871e96c4..387fe319 100644
--- a/app/api/endpoints/llm.py
+++ b/app/api/endpoints/llm.py
@@ -68,6 +68,14 @@ def _sanitize_llm_test_error(message: str, api_key: Optional[str] = None) -> str
         "Authorization: ***",
         sanitized,
     )
+
+    normalized_message = sanitized.lower().replace("_", "").replace(" ", "")
+    if "str" in normalized_message and "modeldump" in normalized_message:
+        return (
+            "LLM 调用失败：服务返回内容不是兼容的模型响应，"
+            "请检查基础地址是否填写为 API Base URL，不要填写网页地址或完整的 "
+            "chat/completions 路径"
+        )
     return sanitized
 
 
diff --git a/tests/test_llm_endpoint_error_messages.py b/tests/test_llm_endpoint_error_messages.py
new file mode 100644
index 00000000..04098847
--- /dev/null
+++ b/tests/test_llm_endpoint_error_messages.py
@@ -0,0 +1,26 @@
+import asyncio
+from unittest.mock import AsyncMock, patch
+
+from app.api.endpoints import llm as llm_endpoint
+
+
+def test_llm_test_maps_internal_model_dump_error_to_base_url_hint():
+    """LLM 测试遇到 SDK 内部响应解析错误时应提示检查基础地址。"""
+    with patch.object(llm_endpoint.settings, "AI_AGENT_ENABLE", True), patch.object(
+        llm_endpoint.settings, "LLM_PROVIDER", "openai"
+    ), patch.object(llm_endpoint.settings, "LLM_MODEL", "gpt-4o-mini"), patch.object(
+        llm_endpoint.settings, "LLM_API_KEY", "sk-test"
+    ), patch.object(
+        llm_endpoint.settings, "LLM_BASE_URL", "https://example.com/not-api"
+    ), patch.object(
+        llm_endpoint.LLMHelper,
+        "test_current_settings",
+        AsyncMock(side_effect=RuntimeError("'str' object has no attribute 'model_dump'")),
+        create=True,
+    ):
+        resp = asyncio.run(llm_endpoint.llm_test(_="token"))
+
+    assert not resp.success
+    assert "基础地址" in resp.message
+    assert "API Base URL" in resp.message
+    assert "model_dump" not in resp.message