diff --git a/ais_bench/benchmark/tasks/swebench_pro/swebench_pro_infer.py b/ais_bench/benchmark/tasks/swebench_pro/swebench_pro_infer.py
index f8685a44..f6350907 100644
--- a/ais_bench/benchmark/tasks/swebench_pro/swebench_pro_infer.py
+++ b/ais_bench/benchmark/tasks/swebench_pro/swebench_pro_infer.py
@@ -199,7 +199,8 @@ def run(self, task_state_manager: TaskStateManager):
             raise AISBenchImportError(
                 SWEBP_CODES.MINISWEAGENT_IMPORT_ERROR,
                 "SWEBenchProInferTask requires mini-swe-agent. "
-                "Install with: pip install mini-swe-agent",
+                "Refer to the Prerequisites section in the swe_bench_pro.md file. "
+                "Install with: git clone https://github.com/scaleapi/mini-swe-agent.git",
             ) from e
 
         dataset_cfg = self.dataset_cfgs[0]
@@ -270,9 +271,8 @@ def run(self, task_state_manager: TaskStateManager):
         if not (model_name or "").strip():
             raise AISBenchValueError(
                 SWEBP_CODES.MODEL_NOT_SET,
-                "No model set for SWEBenchPro infer. In your config, set "
-                "models[0]['model'], models[0]['url'], and models[0]['api_key']. "
-                "Example for local vLLM: model='hosted_vllm/qwen3', url='http://127.0.0.1:2998/v1', api_key='EMPTY'.",
+                "No model set for SWEBenchPro infer in your config "
+                "Example for local vLLM: model='qwen3', url='http://127.0.0.1:2998/v1', api_key='EMPTY'.",
             )
         our_config.setdefault("environment", {})["environment_class"] = "docker"
         # SWE-bench Pro images have hardcoded entrypoint=bin/bash which prevents sleep command execution, need to override to empty for default bin/sh
diff --git a/docs/source_en/faqs/error_codes.md b/docs/source_en/faqs/error_codes.md
index 3d7fa597..4e604531 100644
--- a/docs/source_en/faqs/error_codes.md
+++ b/docs/source_en/faqs/error_codes.md
@@ -1409,7 +1409,7 @@ If you use a virtual environment, make sure `ais_bench` and `mini-swe-agent` are
 No valid model config is detected for SWE-Bench Pro infer (required fields like `model/url/api_key` are missing or empty).
 ### Solution
 Check `models[0]` in your task config and provide at least:
-1. `model`, for example `hosted_vllm/qwen3`
+1. `model`, for example `qwen3`
 2. `url`, for example `http://127.0.0.1:2998/v1`
 3. `api_key`, `EMPTY` is acceptable for local tests
 
diff --git a/docs/source_zh_cn/faqs/error_codes.md b/docs/source_zh_cn/faqs/error_codes.md
index 1c493ac5..4ca738a6 100644
--- a/docs/source_zh_cn/faqs/error_codes.md
+++ b/docs/source_zh_cn/faqs/error_codes.md
@@ -1375,7 +1375,7 @@ pip install -e .
 SWE-Bench Pro infer 未检测到可用模型配置（`model/url/api_key` 等关键字段缺失或为空）。
 ### 解决办法
 检查任务配置中的 `models[0]`，至少补齐以下字段：
-1. `model`：如 `hosted_vllm/qwen3`
+1. `model`：如 `qwen3`
 2. `url`：如 `http://127.0.0.1:2998/v1`
 3. `api_key`：本地测试可设为 `EMPTY`