From b498b0a81c48523e767c60d1c1b32cd053f2c757 Mon Sep 17 00:00:00 2001 From: ninglinLiu Date: Thu, 29 Oct 2026 12:32:12 +0800 Subject: [PATCH] fix(responses): forward text.verbosity through the chat-completions bridge Responses API text.verbosity was dropped when a request was bridged to chat completions, so Codex and other clients silently lost low/high verbosity. Map it to top-level verbosity, and also forward prompt_cache_key / prompt_cache_retention like service_tier. Fixes #48996 --- .../transformation.py | 5 +++++ .../test_litellm_completion_responses.py | 21 +++++++++++++++++++ 2 files changed, 26 insertions(+) diff --git a/litellm/responses/litellm_completion_transformation/transformation.py b/litellm/responses/litellm_completion_transformation/transformation.py index 587cdfcee25..f9d56d29cde 100644 --- a/litellm/responses/litellm_completion_transformation/transformation.py +++ b/litellm/responses/litellm_completion_transformation/transformation.py @@ -909,6 +909,11 @@ class LiteLLMCompletionResponsesConfig: "metadata": kwargs.get("metadata"), "service_tier": responses_api_request.get("service_tier"), "safety_identifier": responses_api_request.get("safety_identifier"), + "prompt_cache_key": responses_api_request.get("prompt_cache_key"), + "prompt_cache_retention": responses_api_request.get("prompt_cache_retention"), + "verbosity": ( + text_param.get("verbosity") if isinstance(text_param, dict) else None + ), "web_search_options": web_search_options, "response_format": response_format, "reasoning_effort": reasoning.effort, diff --git a/tests/unit/responses/litellm_completion_transformation/test_litellm_completion_responses.py b/tests/unit/responses/litellm_completion_transformation/test_litellm_completion_responses.py index cc5d5c62739..78257b58540 100644 --- a/tests/unit/responses/litellm_completion_transformation/test_litellm_completion_responses.py +++ b/tests/unit/responses/litellm_completion_transformation/test_litellm_completion_responses.py @@ -1244,6 +1244,27 @@ class TestFunctionCallTransformation: assert result["safety_identifier"] == "user-7f3a" + def test_text_verbosity_forwarded_as_top_level_verbosity(self) -> None: + result: Final = LiteLLMCompletionResponsesConfig.transform_responses_api_request_to_chat_completion_request( + model="openai/gpt-5.6", + input="Explain what an HTTP 400 is.", + responses_api_request={"text": {"verbosity": "high"}}, + custom_llm_provider="openai", + ) + + assert result["verbosity"] == "high" + + def test_prompt_cache_key_forwarded_to_chat_completion_request(self) -> None: + result: Final = LiteLLMCompletionResponsesConfig.transform_responses_api_request_to_chat_completion_request( + model="openai/gpt-5.6", + input="hi", + responses_api_request={"prompt_cache_key": "sess-abc", "prompt_cache_retention": "24h"}, + custom_llm_provider="openai", + ) + + assert result["prompt_cache_key"] == "sess-abc" + assert result["prompt_cache_retention"] == "24h" + def test_parallel_tool_calls_dropped_when_no_chat_tools_remain(self) -> None: transform: Final = LiteLLMCompletionResponsesConfig.transform_responses_api_request_to_chat_completion_request codex_tool_search: Final = {