From 859cde4a432f5874d3a24457f115660e1fc3ed9e Mon Sep 17 00:00:00 2001
From: Calvin Div <diverpet@icloud.com>
Date: Fri, 28 Aug 2026 23:10:31 +0800
Subject: [PATCH] fix(anthropic): respect chat completions override for
 thinking

---
 .../adapters/handler.py                        |  4 +++-
 ...perimental_pass_through_messages_handler.py | 18 ++++++++++++++++++
 2 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py b/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py
index 9d61701d26d..cd07eb41623 100644
--- a/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py
+++ b/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py
@@ -322,6 +322,7 @@ class LiteLLMMessagesToCompletionTransformationHandler:
         completion_kwargs: _CompletionKwargs,
         *,
         thinking: Mapping[str, object] | None,
+        use_chat_completions_url: bool = False,
     ) -> None:
         """
         When users call `litellm.anthropic.messages.*` with a non-Anthropic model and
@@ -344,7 +345,7 @@ class LiteLLMMessagesToCompletionTransformationHandler:
             except Exception:
                 custom_llm_provider = None
 
-        if custom_llm_provider != "openai":
+        if custom_llm_provider != "openai" or use_chat_completions_url:
             return
 
         if not isinstance(thinking, dict) or thinking.get("type") != "enabled":
@@ -544,6 +545,7 @@ class LiteLLMMessagesToCompletionTransformationHandler:
         LiteLLMMessagesToCompletionTransformationHandler._route_openai_thinking_to_responses_api_if_needed(
             completion_kwargs,
             thinking=thinking,
+            use_chat_completions_url=litellm.use_chat_completions_url_for_anthropic_messages,
         )
 
         return completion_kwargs, tool_name_mapping
diff --git a/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py b/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py
index ad4c3d6bfbb..12acbd1891a 100644
--- a/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py
+++ b/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py
@@ -434,6 +434,24 @@ class TestThinkingParameterTransformation:
 class TestThinkingSummaryPreservation:
     """Tests for thinking.summary preservation and reasoning_auto_summary flag."""
 
+    def test_thinking_respects_chat_completions_url_override(self):
+        from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+            LiteLLMMessagesToCompletionTransformationHandler,
+        )
+
+        completion_kwargs = {
+            "model": "openai/gpt-5.6",
+            "custom_llm_provider": "openai",
+            "reasoning_effort": "medium",
+        }
+        LiteLLMMessagesToCompletionTransformationHandler._route_openai_thinking_to_responses_api_if_needed(
+            completion_kwargs,
+            thinking={"type": "enabled", "budget_tokens": 5000},
+            use_chat_completions_url=True,
+        )
+
+        assert completion_kwargs["model"] == "openai/gpt-5.6"
+
     def test_thinking_summary_concise_preserved_for_openai(self):
         """User-provided summary='concise' should not be replaced with 'detailed'."""
         from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (