Merge f7ea88f3da into de42b67252

litellm/litellm_core_utils/exception_mapping_utils.py

@@ -103,6 +103,8 @@ class ExceptionCheckers:
             "exceeds the available context size",  # llama.cpp/Lemonade
             "exceeds the maximum number of tokens allowed",  # Gemini
             '"code": "context_length_exceeded"',
+            "exceeds the context window of this model",  # OpenAI
+            "exceeded model token limit",  # Moonshot
         ]
         for substring in known_exception_substrings:
             if substring in _error_str_lowercase:

tests/unit/litellm_core_utils/test_exception_mapping_utils.py

@@ -242,6 +242,13 @@ context_window_test_cases = [
         "Input should be 'error' or 'truncate', field: 'context_length_exceeded_behavior', value: 'foo'",
         False,
     ),
+    # OpenAI's current wording (code context_length_exceeded)
+    (
+        "Your input exceeds the context window of this model. Please adjust your input and try again.",
+        True,
+    ),
+    # Moonshot
+    ("Invalid request: Your request exceeded model token limit: 262144", True),
     # Negative cases (should return False)
     ("A generic API error occurred.", False),
     ("Invalid API Key provided.", False),
@@ -439,6 +446,37 @@ def test_gemini_context_window_error_mapping(error_message, should_raise_context
             )
 
 
+@pytest.mark.parametrize(
+    "custom_llm_provider, model, error_message",
+    [
+        (
+            "openai",
+            "gpt-5",
+            "Error code: 400 - {'error': {'message': 'Your input exceeds the context window of this model. "
+            "Please adjust your input and try again.', 'type': 'invalid_request_error', 'param': 'input', "
+            "'code': 'context_length_exceeded'}}",
+        ),
+        (
+            "moonshot",
+            "kimi-k2",
+            "Error code: 400 - {'error': {'message': 'Invalid request: Your request exceeded model token limit: "
+            "262144', 'type': 'invalid_request_error', 'param': None, 'code': None}}",
+        ),
+    ],
+)
+def test_openai_compatible_context_window_error_mapping(custom_llm_provider, model, error_message):
+    """These 400s must map to ContextWindowExceededError so context_window_fallbacks fire."""
+    with pytest.raises(litellm.ContextWindowExceededError) as excinfo:
+        exception_type(
+            model=model,
+            original_exception=OpenAIError(status_code=400, message=error_message),
+            custom_llm_provider=custom_llm_provider,
+        )
+
+    assert excinfo.value.status_code == 400
+    assert excinfo.value.llm_provider == custom_llm_provider
+
+
 def test_lemonade_context_window_error_mapping():
     """Lemonade's llama.cpp backend should map context overflows to LiteLLM's standard error."""
 

tests/unit/llms/openai/responses/test_openai_responses_http.py

@@ -596,7 +596,7 @@ async def test_responses_stream_error_event_raises_bad_request() -> None:
         stream: Final = await litellm.aresponses(
             model="openai/gpt-5-mini", api_key="sk-test", input="oversized prompt", stream=True
         )
-        with pytest.raises(litellm.BadRequestError) as exc_info:
+        with pytest.raises(litellm.ContextWindowExceededError) as exc_info:
             _ = [event async for event in stream]
         requests: Final = tuple(route.calls)