fix(messages): route to Responses API based on model_info.mode

When a /v1/messages request arrives for a non-native model (no
Anthropic messages config), the handler decides between Responses API
and chat/completions adapter. Previously this was provider-only:
only models under the "openai" provider went to Responses API.

Now also check model_info["mode"] == "responses", so models like
github_copilot GPT (configured with mode: responses in config.yaml)
go directly through messages → responses translation instead of
the two-layer messages → chat/completions → responses bridge.

Also fixes tool translation for web search:
- tool_choice: web_search name maps to {"type": "web_search"}
  instead of {"type": "function", "name": "web_search"}
- tools: Anthropic allowed_domains on web_search tools are translated
  to Responses API filters.allowed_domains (blocked_domains dropped
  as Responses API does not support them)
- Use "web_search" tool type instead of deprecated "web_search_preview"
  (only "web_search" supports filters.allowed_domains)

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

@@ -51,14 +51,23 @@ from .utils import AnthropicMessagesRequestUtils, mock_response
 _RESPONSES_API_PROVIDERS = frozenset({"openai"})
 
 
-def _should_route_to_responses_api(custom_llm_provider: Optional[str]) -> bool:
+def _should_route_to_responses_api(
+    custom_llm_provider: Optional[str],
+    model_info: Optional[Dict] = None,
+) -> bool:
     """Return True when the provider should use the Responses API path.
 
+    Routes to the Responses API when:
+    - The provider is in _RESPONSES_API_PROVIDERS (e.g. openai), OR
+    - The model has mode=responses in model_info (e.g. github_copilot GPT models)
+
     Set ``litellm.use_chat_completions_url_for_anthropic_messages = True`` to
-    opt out and route OpenAI/Azure requests through chat/completions instead.
+    opt out and route all requests through chat/completions instead.
     """
     if litellm.use_chat_completions_url_for_anthropic_messages:
         return False
+    if model_info and model_info.get("mode") == "responses":
+        return True
     return custom_llm_provider in _RESPONSES_API_PROVIDERS
 
 
@@ -558,7 +567,7 @@ def anthropic_messages_handler(
             custom_llm_provider=custom_llm_provider,
             **kwargs,
         )
-        if _should_route_to_responses_api(custom_llm_provider):
+        if _should_route_to_responses_api(custom_llm_provider, kwargs.get("model_info")):
             return LiteLLMMessagesToResponsesAPIHandler.anthropic_messages_handler(**_shared_kwargs)
 
         # The in-gateway context_management polyfill runs inside

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

@@ -185,7 +185,11 @@ class LiteLLMAnthropicToResponsesAPIAdapter:
             tool_name = tool_dict.get("name", "")
             # web_search tool
             if (isinstance(tool_type, str) and tool_type.startswith("web_search")) or tool_name == "web_search":
-                result.append({"type": "web_search_preview"})
+                web_tool: Dict[str, Any] = {"type": "web_search"}
+                allowed = tool_dict.get("allowed_domains")
+                if allowed:
+                    web_tool["filters"] = {"allowed_domains": allowed}
+                result.append(web_tool)
                 continue
             func_tool: Dict[str, Any] = {"type": "function", "name": tool_name}
             if "description" in tool_dict:
@@ -204,7 +208,11 @@ class LiteLLMAnthropicToResponsesAPIAdapter:
         if tc_type == "any":
             return "required"
         elif tc_type == "tool":
-            return {"type": "function", "name": tool_choice.get("name", "")}
+            name = tool_choice.get("name", "")
+            # web_search is a hosted tool, not a function
+            if name == "web_search":
+                return {"type": "web_search"}
+            return {"type": "function", "name": name}
         elif tc_type == "none":
             return "none"
         return "auto"

tests/test_litellm/llms/anthropic/experimental_pass_through/responses_adapters/test_responses_adapters_transformation.py

@@ -527,16 +527,16 @@ class TestTranslateToolsToResponsesAPI:
         assert "parameters" not in result[0]
 
     def test_web_search_tool_by_name(self):
-        """Tool named 'web_search' maps to web_search_preview."""
+        """Tool named 'web_search' maps to web_search."""
         tools = [{"name": "web_search", "type": "custom"}]
         result = _ADAPTER.translate_tools_to_responses_api(tools)  # type: ignore[arg-type]
-        assert result == [{"type": "web_search_preview"}]
+        assert result == [{"type": "web_search"}]
 
     def test_web_search_tool_by_type_prefix(self):
-        """Tool with type starting with 'web_search' maps to web_search_preview."""
+        """Tool with type starting with 'web_search' maps to web_search."""
         tools = [{"name": "search", "type": "web_search_20250305"}]
         result = _ADAPTER.translate_tools_to_responses_api(tools)  # type: ignore[arg-type]
-        assert result == [{"type": "web_search_preview"}]
+        assert result == [{"type": "web_search"}]
 
     def test_multiple_tools_order_preserved(self):
         """Multiple tools are converted in order."""
@@ -548,7 +548,7 @@ class TestTranslateToolsToResponsesAPI:
         result = _ADAPTER.translate_tools_to_responses_api(tools)  # type: ignore[arg-type]
         assert len(result) == 3
         assert result[0]["name"] == "tool_a"
-        assert result[1] == {"type": "web_search_preview"}
+        assert result[1] == {"type": "web_search"}
         assert result[2]["name"] == "tool_b"
 
     def test_empty_tools_list(self):