Merge 8014bbba23 into 7bc80994b7

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

@@ -71,13 +71,14 @@ def _should_route_to_responses_api(
     custom_llm_provider: str | None,
     requested_model: str | None = None,
     resolved_model: str | None = None,
+    use_chat_completions_api: bool | None = None,
 ) -> bool:
     """Return True when the request should use the Responses API path.
 
     Set ``litellm.use_chat_completions_url_for_anthropic_messages = True`` to
     opt out and route OpenAI/Azure requests through chat/completions instead.
     """
-    if litellm.use_chat_completions_url_for_anthropic_messages:
+    if litellm.use_chat_completions_url_for_anthropic_messages or use_chat_completions_api is True:
         return False
     if custom_llm_provider in _RESPONSES_API_PROVIDERS:
         return True
@@ -589,7 +590,12 @@ def anthropic_messages_handler(
             custom_llm_provider=custom_llm_provider,
             **kwargs,
         )
-        if _should_route_to_responses_api(custom_llm_provider, original_model, model):
+        if _should_route_to_responses_api(
+            custom_llm_provider=custom_llm_provider,
+            requested_model=original_model,
+            resolved_model=model,
+            use_chat_completions_api=kwargs.get("use_chat_completions_api"),
+        ):
             return LiteLLMMessagesToResponsesAPIHandler.anthropic_messages_handler(**_shared_kwargs)
 
         # The in-gateway context_management polyfill runs inside

tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py

@@ -874,26 +874,30 @@ def _gate_stubs(monkeypatch):
     from litellm.llms.anthropic.experimental_pass_through.messages import handler
 
     captured = {}
-    translation_calls = {"count": 0}
+    translation_calls = {"responses": 0, "chat_completions": 0}
 
     def fake_native(**kwargs):
         captured["config"] = kwargs.get("anthropic_messages_provider_config")
         return "native-passthrough"
 
-    def fake_translation(**kwargs):
-        translation_calls["count"] += 1
-        return "translated"
+    def fake_responses(**kwargs):
+        translation_calls["responses"] += 1
+        return "responses"
+
+    def fake_chat_completions(**kwargs):
+        translation_calls["chat_completions"] += 1
+        return "chat-completions"
 
     monkeypatch.setattr(handler.base_llm_http_handler, "anthropic_messages_handler", fake_native)
     monkeypatch.setattr(
         handler.LiteLLMMessagesToResponsesAPIHandler,
         "anthropic_messages_handler",
-        staticmethod(fake_translation),
+        staticmethod(fake_responses),
     )
     monkeypatch.setattr(
         handler.LiteLLMMessagesToCompletionTransformationHandler,
         "anthropic_messages_handler",
-        staticmethod(fake_translation),
+        staticmethod(fake_chat_completions),
     )
     return captured, translation_calls
 
@@ -921,7 +925,8 @@ def test_gate_passthrough_when_supported_endpoints_opts_in(monkeypatch):
 
     assert result == "native-passthrough"
     assert isinstance(captured["config"], OpenAILikeAnthropicMessagesConfig)
-    assert translation_calls["count"] == 0
+    assert translation_calls["responses"] == 0
+    assert translation_calls["chat_completions"] == 0
 
 
 def test_gate_translates_when_supported_endpoints_absent(monkeypatch):
@@ -941,8 +946,32 @@ def test_gate_translates_when_supported_endpoints_absent(monkeypatch):
         api_base="https://host/v1",
     )
 
-    assert result == "translated"
-    assert translation_calls["count"] == 1
+    assert result == "responses"
+    assert translation_calls["responses"] == 1
+    assert translation_calls["chat_completions"] == 0
+    assert "config" not in captured
+
+
+def test_gate_uses_chat_completions_when_requested(monkeypatch):
+    """The per-deployment chat-completions opt-in must bypass Responses API."""
+    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+        anthropic_messages_handler,
+    )
+
+    captured, translation_calls = _gate_stubs(monkeypatch)
+
+    result = anthropic_messages_handler(
+        max_tokens=100,
+        messages=[{"role": "user", "content": "Hello"}],
+        model="openai/some-model",
+        api_key="sk-test",
+        api_base="https://host/v1",
+        use_chat_completions_api=True,
+    )
+
+    assert result == "chat-completions"
+    assert translation_calls["responses"] == 0
+    assert translation_calls["chat_completions"] == 1
     assert "config" not in captured
 
 
@@ -964,8 +993,9 @@ def test_gate_passthrough_skipped_when_only_chat_completions_supported(monkeypat
         model_info={"supported_endpoints": ["/v1/chat/completions"]},
     )
 
-    assert result == "translated"
-    assert translation_calls["count"] == 1
+    assert result == "responses"
+    assert translation_calls["responses"] == 1
+    assert translation_calls["chat_completions"] == 0
     assert "config" not in captured