From 3dd76bd5160f6bc3d899656a2cc563ab8ba979b2 Mon Sep 17 00:00:00 2001 From: shrey kharbanda Date: Wed, 23 Sep 2026 06:43:23 +0000 Subject: [PATCH] fix(responses): read prompt-managed reasoning params from local_vars --- litellm/responses/main.py | 10 ++- ...responses_transformation_transformation.py | 4 +- .../test_responses_api_request_body.py | 61 ++++++++++++++++++- .../test_responses_websocket_all_providers.py | 4 +- 4 files changed, 72 insertions(+), 7 deletions(-) diff --git a/litellm/responses/main.py b/litellm/responses/main.py index 8aa76360f3e..884ea7e217d 100644 --- a/litellm/responses/main.py +++ b/litellm/responses/main.py @@ -1313,10 +1313,14 @@ def responses( _raise_responses_compatibility_failure(compatibility_failure, model, custom_llm_provider) local_vars.update(kwargs) + current_reasoning: Final = cast( # cast-ok: prompt-managed reasoning arrives as a plain dict + Reasoning | None, local_vars.get("reasoning") + ) + reasoning_effort: Final = local_vars.get("reasoning_effort") request_reasoning: Final = ( - LiteLLMResponsesTransformationHandler()._map_reasoning_effort(kwargs["reasoning_effort"]) - if reasoning is None and kwargs.get("reasoning_effort") is not None - else reasoning + LiteLLMResponsesTransformationHandler()._map_reasoning_effort(reasoning_effort) + if current_reasoning is None and reasoning_effort is not None + else current_reasoning ) response_api_optional_params: Final[ResponsesAPIOptionalRequestParams] = ( ResponsesAPIRequestUtils.get_requested_response_api_optional_param( diff --git a/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py b/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py index bfcc4cc872f..d0f9bad795d 100644 --- a/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py +++ b/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py @@ -2495,7 +2495,9 @@ def test_transform_request_bedrock_mantle_tools_keeps_reasoning_effort(monkeypat "reasoning_effort", [5, ["low"], "hgih", "", {"effort": 5}, {"effort": "max"}, *get_args(REASONING_EFFORT)], ) -def test_transform_request_never_drops_reasoning_effort(monkeypatch, reasoning_effort): +def test_transform_request_never_drops_reasoning_effort( + monkeypatch: pytest.MonkeyPatch, reasoning_effort: int | list[str] | str | dict[str, object] +): monkeypatch.setattr(litellm, "reasoning_auto_summary", False) monkeypatch.delenv("LITELLM_REASONING_AUTO_SUMMARY", raising=False) handler: Final = LiteLLMResponsesTransformationHandler() diff --git a/tests/test_litellm/responses/test_responses_api_request_body.py b/tests/test_litellm/responses/test_responses_api_request_body.py index f3ddaafe707..98e74955c6f 100644 --- a/tests/test_litellm/responses/test_responses_api_request_body.py +++ b/tests/test_litellm/responses/test_responses_api_request_body.py @@ -224,7 +224,9 @@ async def test_aresponses_drops_stream_options(): @pytest.mark.asyncio -async def test_aresponses_forwards_non_enum_reasoning_effort(monkeypatch, respx_mock: respx.MockRouter): +async def test_aresponses_forwards_non_enum_reasoning_effort( + monkeypatch: pytest.MonkeyPatch, respx_mock: respx.MockRouter +): monkeypatch.setenv("OPENAI_API_KEY", "fake-api-key") monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) litellm.in_memory_llm_clients_cache.flush_cache() @@ -242,7 +244,7 @@ async def test_aresponses_forwards_non_enum_reasoning_effort(monkeypatch, respx_ @pytest.mark.asyncio async def test_acompletion_with_tools_forwards_non_enum_reasoning_effort_over_the_bridge( - monkeypatch, respx_mock: respx.MockRouter + monkeypatch: pytest.MonkeyPatch, respx_mock: respx.MockRouter ): monkeypatch.setenv("OPENAI_API_KEY", "fake-api-key") monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) @@ -272,6 +274,61 @@ async def test_acompletion_with_tools_forwards_non_enum_reasoning_effort_over_th assert response.id == "resp_bridge_int" +@pytest.mark.asyncio +async def test_aresponses_forwards_prompt_managed_reasoning_effort( + monkeypatch: pytest.MonkeyPatch, respx_mock: respx.MockRouter +): + from litellm.responses.main import _AsyncPromptManagementOutcome + + monkeypatch.setenv("OPENAI_API_KEY", "fake-api-key") + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + litellm.in_memory_llm_clients_cache.flush_cache() + upstream: Final = respx_mock.post("https://api.openai.com/v1/responses").mock( + return_value=httpx.Response(200, json=_minimal_responses_api_payload("resp_prompt_effort", "gpt-5.4")) + ) + + response: Final = await litellm.aresponses( + model="openai/gpt-5.4", + input="hi", + _async_prompt_merged_params=_AsyncPromptManagementOutcome( + merged_optional_params={"reasoning_effort": 5}, deployment_model_info=None + ), + ) + + assert upstream.call_count == 1 + request_body: Final = json.loads(upstream.calls[0].request.read()) + assert request_body["reasoning"] == {"effort": 5} + assert "reasoning_effort" not in request_body + assert response.output[0].content[0].text == "Done." + + +@pytest.mark.asyncio +async def test_aresponses_forwards_prompt_managed_reasoning_dict( + monkeypatch: pytest.MonkeyPatch, respx_mock: respx.MockRouter +): + from litellm.responses.main import _AsyncPromptManagementOutcome + + monkeypatch.setenv("OPENAI_API_KEY", "fake-api-key") + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + litellm.in_memory_llm_clients_cache.flush_cache() + upstream: Final = respx_mock.post("https://api.openai.com/v1/responses").mock( + return_value=httpx.Response(200, json=_minimal_responses_api_payload("resp_prompt_reasoning", "gpt-5.4")) + ) + + response: Final = await litellm.aresponses( + model="openai/gpt-5.4", + input="hi", + _async_prompt_merged_params=_AsyncPromptManagementOutcome( + merged_optional_params={"reasoning": {"effort": "high", "summary": "detailed"}}, deployment_model_info=None + ), + ) + + assert upstream.call_count == 1 + request_body: Final = json.loads(upstream.calls[0].request.read()) + assert request_body["reasoning"] == {"effort": "high", "summary": "detailed"} + assert response.output[0].content[0].text == "Done." + + @pytest.mark.asyncio async def test_aresponses_keeps_include_obfuscation_in_stream_options(): """include_obfuscation is a valid Responses API stream option and must survive the include_usage strip.""" diff --git a/tests/test_litellm/responses/test_responses_websocket_all_providers.py b/tests/test_litellm/responses/test_responses_websocket_all_providers.py index a5ab9eda237..3888a84fb5d 100644 --- a/tests/test_litellm/responses/test_responses_websocket_all_providers.py +++ b/tests/test_litellm/responses/test_responses_websocket_all_providers.py @@ -1315,7 +1315,9 @@ class TestNativeWebSocketDeploymentDefaults: assert dict(defaults.fill_missing) == {"reasoning": {"effort": "xhigh", "summary": "auto"}} @pytest.mark.parametrize("reasoning_effort", [5, ["low"], "hgih"]) - def test_builder_forwards_non_enum_reasoning_effort_like_the_http_path(self, reasoning_effort): + def test_builder_forwards_non_enum_reasoning_effort_like_the_http_path( + self, reasoning_effort: int | list[str] | str + ): from litellm.responses.main import _build_responses_websocket_request_defaults defaults = _build_responses_websocket_request_defaults({"model": "gpt-5-pro", "reasoning_effort": reasoning_effort})