Merge pull request #26248 from BerriAI/litellm_anthropic_messages_call_type_fix

fix(proxy): preserve anthropic_messages call type for /v1/messages logging

litellm/llms/anthropic/chat/transformation.py

@@ -1534,12 +1534,16 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
                 f"Invalid effort value: {effort}. Must be one of: "
                 f"'high', 'medium', 'low', 'xhigh', 'max'"
             )
-        # ``max`` is Claude Opus 4.6 only (not Sonnet 4.6, not Opus 4.5/4.7).
-        # Keep this hardcoded so the error message is specific and stable.
-        if effort == "max" and not self._is_opus_4_6_model(model):
+        # ``max`` is for Opus 4.6+ output effort (not Sonnet 4.6, not Opus 4.5).
+        # Accept known Opus 4.6/4.7 id patterns and/or ``supports_max_reasoning_effort``
+        # in the model map (same pattern as ``xhigh`` below).
+        if effort == "max" and not (
+            self._is_opus_4_6_model(model)
+            or self._is_opus_4_7_model(model)
+            or self._supports_effort_level(model, "max")
+        ):
             raise ValueError(
-                f"effort='max' is only supported by Claude Opus 4.6. "
-                f"Got model: {model}"
+                f"effort='max' is not supported by this model. Got model: {model}"
             )
         # ``xhigh`` is data-driven via ``supports_xhigh_reasoning_effort`` so
         # enabling it for a new model is a pure model-map change.

litellm/llms/anthropic/experimental_pass_through/adapters/handler.py

@@ -235,7 +235,7 @@ class LiteLLMMessagesToCompletionTransformationHandler:
             ):
                 from litellm.types.utils import CallTypes
 
-                setattr(value, "call_type", CallTypes.completion.value)
+                setattr(value, "call_type", CallTypes.anthropic_messages.value)
                 setattr(
                     value, "stream_options", completion_kwargs.get("stream_options")
                 )

litellm/llms/anthropic/experimental_pass_through/responses_adapters/handler.py

@@ -105,7 +105,7 @@ def _build_responses_kwargs(
                 # Reclassify as acompletion so the success handler doesn't try to
                 # validate the Responses API event as an AnthropicResponse.
                 # (Mirrors the pattern used in LiteLLMMessagesToCompletionTransformationHandler.)
-                setattr(value, "call_type", CallTypes.acompletion.value)
+                setattr(value, "call_type", CallTypes.anthropic_messages.value)
             responses_kwargs[key] = value
         elif key not in excluded and key not in responses_kwargs and value is not None:
             responses_kwargs[key] = value

litellm/model_prices_and_context_window_backup.json

@@ -9262,6 +9262,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
+        "supports_max_reasoning_effort": true,
         "tool_use_system_prompt_tokens": 346,
         "provider_specific_entry": {
             "us": 1.1,
@@ -9296,6 +9297,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
+        "supports_max_reasoning_effort": true,
         "tool_use_system_prompt_tokens": 346,
         "provider_specific_entry": {
             "us": 1.1,

model_prices_and_context_window.json

@@ -9276,6 +9276,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
+        "supports_max_reasoning_effort": true,
         "tool_use_system_prompt_tokens": 346,
         "provider_specific_entry": {
             "us": 1.1,
@@ -9310,6 +9311,7 @@
         "supports_tool_choice": true,
         "supports_vision": true,
         "supports_xhigh_reasoning_effort": true,
+        "supports_max_reasoning_effort": true,
         "tool_use_system_prompt_tokens": 346,
         "provider_specific_entry": {
             "us": 1.1,

tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py

@@ -1654,7 +1654,7 @@ def test_max_effort_rejected_for_opus_45():
     messages = [{"role": "user", "content": "Test"}]
 
     with pytest.raises(
-        ValueError, match="effort='max' is only supported by Claude Opus 4.6"
+        ValueError, match="effort='max' is not supported by this model"
     ):
         optional_params = {"output_config": {"effort": "max"}}
         config.transform_request(
@@ -2213,12 +2213,12 @@ def test_reasoning_effort_does_not_set_output_config_for_older_models():
 
 
 def test_max_effort_rejected_for_sonnet_46():
-    """Test that effort='max' is rejected for Sonnet 4.6 (only Opus 4.6 supports max)."""
+    """Test that effort='max' is rejected for Sonnet 4.6 (Opus-only effort level)."""
     config = AnthropicConfig()
     messages = [{"role": "user", "content": "Test"}]
 
     with pytest.raises(
-        ValueError, match="effort='max' is only supported by Claude Opus 4.6"
+        ValueError, match="effort='max' is not supported by this model"
     ):
         config.transform_request(
             model="claude-sonnet-4-6-20260219",
@@ -2245,6 +2245,22 @@ def test_max_effort_accepted_for_opus_46():
     assert result["output_config"]["effort"] == "max"
 
 
+def test_max_effort_accepted_for_opus_47():
+    """Test that effort='max' works for Opus 4.7."""
+    config = AnthropicConfig()
+    messages = [{"role": "user", "content": "Test"}]
+
+    result = config.transform_request(
+        model="claude-opus-4-7",
+        messages=messages,
+        optional_params={"output_config": {"effort": "max"}},
+        litellm_params={},
+        headers={},
+    )
+
+    assert result["output_config"]["effort"] == "max"
+
+
 def test_effort_beta_header_not_injected_for_46_models():
     """
     Test that is_effort_used returns False for Claude 4.6 models.

tests/test_litellm/proxy/test_litellm_pre_call_utils.py

@@ -1749,6 +1749,95 @@ async def test_add_litellm_metadata_from_request_headers():
         litellm.callbacks = original_callbacks
 
 
+@pytest.mark.asyncio
+async def test_anthropic_messages_standard_logging_object_matches_fixture():
+    """
+    Regression: /v1/messages calls routed to non-Anthropic providers should keep
+    call_type=anthropic_messages in standard logging payloads.
+    """
+    litellm._turn_on_debug()
+    test_logger = TestCustomLogger()
+    original_callbacks = litellm.callbacks
+    litellm.callbacks = [test_logger]
+
+    try:
+        data = {
+            "model": "gemini/gemini-2.5-flash",
+            "messages": [{"role": "user", "content": "Hi."}],
+            "stream": False,
+            "mock_response": "Hello! How can I help you today?",
+            "api_key": "fake-key",
+            "max_tokens": 4096,
+        }
+
+        mock_request = MagicMock(spec=Request)
+        mock_request.headers = {"user-agent": "PostmanRuntime/7.53.0"}
+        mock_request.url.path = "/v1/messages"
+        mock_request.url = MagicMock()
+        mock_request.url.__str__.return_value = "http://localhost/v1/messages"
+        mock_request.method = "POST"
+        mock_request.query_params = {}
+        mock_request.client = MagicMock()
+        mock_request.client.host = "127.0.0.1"
+
+        mock_fastapi_response = MagicMock(spec=Response)
+        mock_user_api_key_dict = UserAPIKeyAuth(
+            api_key="test-key", user_id="default_user_id"
+        )
+
+        mock_proxy_logging_obj = MagicMock(spec=ProxyLogging)
+
+        async def mock_during_call_hook(*args, **kwargs):
+            return None
+
+        async def mock_pre_call_hook(*args, **kwargs):
+            return data
+
+        async def mock_post_call_success_hook(*args, **kwargs):
+            return kwargs.get("response", args[2] if len(args) > 2 else None)
+
+        mock_proxy_logging_obj.during_call_hook = mock_during_call_hook
+        mock_proxy_logging_obj.pre_call_hook = mock_pre_call_hook
+        mock_proxy_logging_obj.post_call_success_hook = mock_post_call_success_hook
+
+        processor = ProxyBaseLLMRequestProcessing(data=data)
+        await processor.base_process_llm_request(
+            request=mock_request,
+            fastapi_response=mock_fastapi_response,
+            user_api_key_dict=mock_user_api_key_dict,
+            route_type="anthropic_messages",
+            proxy_logging_obj=mock_proxy_logging_obj,
+            general_settings={},
+            proxy_config=MagicMock(),
+            select_data_generator=None,
+            llm_router=None,
+            model="gemini/gemini-2.5-flash",
+            is_streaming_request=False,
+        )
+
+        await asyncio.sleep(3)
+
+        assert test_logger.standard_logging_object is not None
+        actual = test_logger.standard_logging_object
+
+        expected = {
+            "call_type": "anthropic_messages",
+            "status": "success",
+            "model": "gemini/gemini-2.5-flash",
+        }
+
+        # Compare only stable fields from the saved proxy log snapshot.
+        actual_projection = {
+            "call_type": actual.get("call_type"),
+            "status": actual.get("status"),
+            "model": actual.get("model"),
+        }
+        assert actual_projection == expected
+        assert actual.get("call_type") == "anthropic_messages"
+    finally:
+        litellm.callbacks = original_callbacks
+
+
 def test_add_litellm_metadata_from_request_headers_x_litellm_trace_id_sets_chain_id():
     """x-litellm-trace-id sets both metadata and top-level litellm_session_id/litellm_trace_id for call chaining."""
     headers = {"x-litellm-trace-id": "foo"}