diff --git a/src/langbot/pkg/provider/modelmgr/requesters/litellmchat.py b/src/langbot/pkg/provider/modelmgr/requesters/litellmchat.py
index 6b087916..63536a2c 100644
--- a/src/langbot/pkg/provider/modelmgr/requesters/litellmchat.py
+++ b/src/langbot/pkg/provider/modelmgr/requesters/litellmchat.py
@@ -75,22 +75,33 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
                 continue
         return False
 
+    @staticmethod
+    def _positive_int(value: typing.Any) -> int | None:
+        if isinstance(value, bool):
+            return None
+        if isinstance(value, int) and value > 0:
+            return value
+        if isinstance(value, str) and value.isdigit():
+            parsed_value = int(value)
+            if parsed_value > 0:
+                return parsed_value
+        return None
+
     def _context_length_from_scan_payload(self, model_payload: dict[str, typing.Any] | None) -> int | None:
         if not model_payload:
             return None
 
         for field_name in ('context_length', 'context_window', 'max_context_length'):
-            value = model_payload.get(field_name)
-            if isinstance(value, bool):
-                continue
-            if isinstance(value, int) and value > 0:
-                return value
-            if isinstance(value, str) and value.isdigit():
-                parsed_value = int(value)
-                if parsed_value > 0:
-                    return parsed_value
+            context_length = self._positive_int(model_payload.get(field_name))
+            if context_length is not None:
+                return context_length
         return None
 
+    def _context_length_from_litellm_model_info(self, model_info: typing.Any) -> int | None:
+        if isinstance(model_info, dict):
+            return self._positive_int(model_info.get('max_input_tokens'))
+        return self._positive_int(getattr(model_info, 'max_input_tokens', None))
+
     def _metadata_provider_candidates(self, model_name: str) -> list[str]:
         normalized_model_name = (model_name or '').lower()
         candidates = []
@@ -126,7 +137,7 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
         return None
 
     def _safe_context_length(self, model_name: str) -> int | None:
-        helper = getattr(litellm, 'get_max_tokens', None)
+        helper = getattr(litellm, 'get_model_info', None)
         if not callable(helper):
             return self._known_context_length_fallback(model_name)
 
@@ -143,11 +154,12 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
                 continue
             tried_candidates.append(candidate)
             try:
-                max_tokens = helper(candidate)
+                model_info = helper(candidate)
             except Exception:
                 continue
-            if isinstance(max_tokens, int) and max_tokens > 0:
-                return max_tokens
+            context_length = self._context_length_from_litellm_model_info(model_info)
+            if context_length is not None:
+                return context_length
         return self._known_context_length_fallback(model_name)
 
     def _supports_function_calling(self, model_name: str) -> bool:
diff --git a/tests/unit_tests/provider/test_litellmchat.py b/tests/unit_tests/provider/test_litellmchat.py
index 1ec12d82..2878d827 100644
--- a/tests/unit_tests/provider/test_litellmchat.py
+++ b/tests/unit_tests/provider/test_litellmchat.py
@@ -1034,11 +1034,28 @@ class TestScanModels:
             },
         )
 
-        with patch.object(litellmchat.litellm, 'get_max_tokens') as mock_get_max_tokens:
-            mock_get_max_tokens.side_effect = lambda model: 131072 if model == 'moonshot/moonshot-v1-128k' else None
+        with patch.object(litellmchat.litellm, 'get_model_info') as mock_get_model_info:
+            mock_get_model_info.side_effect = (
+                lambda model: {'max_input_tokens': 131072}
+                if model == 'moonshot/moonshot-v1-128k'
+                else {}
+            )
 
             assert requester._safe_context_length('moonshot-v1-128k') == 131072
 
+    def test_safe_context_length_uses_litellm_max_input_tokens(self):
+        """LiteLLM max_output_tokens must not be treated as the context window."""
+        requester = litellmchat.LiteLLMRequester(ap=Mock(), config={})
+
+        with patch.object(litellmchat.litellm, 'get_model_info') as mock_get_model_info:
+            mock_get_model_info.return_value = {
+                'max_input_tokens': 128000,
+                'max_output_tokens': 16384,
+                'max_tokens': 16384,
+            }
+
+            assert requester._safe_context_length('gpt-4o') == 128000
+
     def test_litellm_bool_helper_tries_moonshot_metadata_alias(self):
         """OpenAI-compatible Moonshot endpoints still use Moonshot metadata for abilities."""
         requester = litellmchat.LiteLLMRequester(
@@ -1102,7 +1119,7 @@ class TestScanModels:
             },
         )
 
-        with patch.object(litellmchat.litellm, 'get_max_tokens', side_effect=Exception('not mapped')):
+        with patch.object(litellmchat.litellm, 'get_model_info', side_effect=Exception('not mapped')):
             assert requester._safe_context_length('deepseek-v4-pro') == 1_000_000
             assert requester._safe_context_length('deepseek-v4-flash') == 1_000_000