diff --git a/rdagent/oai/backend/deprec.py b/rdagent/oai/backend/deprec.py
index 3fcae490..adfdbc8d 100644
--- a/rdagent/oai/backend/deprec.py
+++ b/rdagent/oai/backend/deprec.py
@@ -153,7 +153,7 @@ class DeprecBackend(APIBackend):
             self.gcr_endpoint_do_sample = LLM_SETTINGS.gcr_endpoint_do_sample
             self.gcr_endpoint_max_token = LLM_SETTINGS.gcr_endpoint_max_token
             if not os.environ.get("PYTHONHTTPSVERIFY", "") and hasattr(ssl, "_create_unverified_context"):
-                ssl._create_default_https_context = ssl._create_unverified_context  # noqa: SLF001
+                ssl._create_default_https_context = ssl._create_unverified_context  # type: ignore[assignment]
             self.chat_model_map = LLM_SETTINGS.chat_model_map
             self.chat_model = LLM_SETTINGS.chat_model
             self.encoder = None
diff --git a/rdagent/oai/backend/litellm.py b/rdagent/oai/backend/litellm.py
index dd55ceaf..164b4827 100644
--- a/rdagent/oai/backend/litellm.py
+++ b/rdagent/oai/backend/litellm.py
@@ -7,7 +7,7 @@ from litellm import (
     completion,
     completion_cost,
     embedding,
-    get_max_tokens,
+    get_model_info,
     supports_function_calling,
     supports_response_schema,
     token_counter,
@@ -205,10 +205,19 @@ class LiteLLMAPIBackend(APIBackend):
 
     @property
     def chat_token_limit(self) -> int:
+        """Suggest an input token limit, ensuring enough space in the context window for the maximum output tokens."""
         try:
-            max_tokens = get_max_tokens(LITELLM_SETTINGS.chat_model)
-            if max_tokens is None:
+            model_info = get_model_info(LITELLM_SETTINGS.chat_model)
+            if model_info is None:
                 return super().chat_token_limit
-            return max_tokens
+
+            max_input = model_info.get("max_input_tokens")
+            max_output = model_info.get("max_output_tokens")
+
+            if max_input is None or max_output is None:
+                return super().chat_token_limit
+
+            max_input_tokens = max_input - max_output
+            return max_input_tokens
         except Exception as e:
             return super().chat_token_limit