Merge 2ec59eebf6 into d23bec84c4

litellm/llms/xai/chat/transformation.py

@@ -219,13 +219,19 @@ class XAIChatConfig(OpenAIGPTConfig):
         litellm_params: dict,
         headers: dict,
     ) -> dict:
-        """
-        Handle https://github.com/BerriAI/litellm/issues/9720
+        """Handle https://github.com/BerriAI/litellm/issues/9720"""
+        if "web_search_options" in optional_params:
+            verbose_logger.warning(
+                "XAI no longer supports web search on /chat/completions (Live Search is deprecated). "
+                "Dropping 'web_search_options'. Use the Responses API for XAI web search."
+            )
 
-        Filter out 'name' from messages
-        """
-        messages = strip_name_from_messages(messages)
-        return super().transform_request(model, messages, optional_params, litellm_params, headers)
+        chat_params: Final = {  # mutable-ok: base transform_request takes a plain dict of optional params
+            key: value for key, value in optional_params.items() if key != "web_search_options"
+        }
+        return super().transform_request(
+            model, strip_name_from_messages(messages), chat_params, litellm_params, headers
+        )
 
     @staticmethod
     def _fix_choice_finish_reason_for_tool_calls(choice: Choices) -> None:

litellm/main.py

@@ -1036,10 +1036,6 @@ def responses_api_bridge_check(
             mode = "responses"
             model_info["mode"] = mode
 
-        if web_search_options is not None and custom_llm_provider == "xai":
-            model_info["mode"] = "responses"
-            model = model.replace("responses/", "")
-
     except Exception as e:
         verbose_logger.debug("Error getting model info: %s", e)
 
@@ -1048,6 +1044,11 @@ def responses_api_bridge_check(
             mode = "responses"
             model_info["mode"] = mode
 
+    # xAI retired Live Search on /v1/chat/completions (410), so web search only works on /v1/responses
+    if web_search_options is not None and custom_llm_provider == "xai":
+        model_info["mode"] = "responses"
+        model = model.replace("responses/", "")
+
     # OpenAI/Azure GPT-5 chat-completions that need Responses-only fields (e.g.
     # ``reasoningSummary`` in ``extra_body``) must be bridged; Chat Completions rejects
     # those keys.

tests/test_litellm/llms/xai/test_xai_chat_transformation.py

@@ -124,6 +124,24 @@ class TestXAIParallelToolCalls:
         assert result["messages"][0]["role"] == "user"
 
 
+class TestXAIChatWebSearchOptions:
+    """XAI answers /chat/completions requests carrying web_search_options with a 410 (Live Search retired)"""
+
+    def test_transform_request_drops_web_search_options(self):
+        config = XAIChatConfig()
+
+        result = config.transform_request(
+            model="xai/grok-4.6",
+            messages=[{"role": "user", "content": "newest litellm version?"}],
+            optional_params={"web_search_options": {"search_context_size": "medium"}, "temperature": 0.5},
+            litellm_params={},
+            headers={},
+        )
+
+        assert "web_search_options" not in result
+        assert result["temperature"] == 0.5
+
+
 class TestXAIUsageNormalization:
     def test_preserves_reasoning_tokens_in_total_usage(self):
         usage = Usage(prompt_tokens=100, completion_tokens=50, total_tokens=200)

tests/test_litellm/test_xai_responses_auto_routing.py

@@ -204,6 +204,17 @@ class TestXAIResponsesAutoRouting:
         assert model_info.get("mode") == "responses"
         assert updated_model == model
 
+    def test_responses_api_bridge_check_with_web_search_options_on_unmapped_model(self):
+        """web search must reach /responses even for a model missing from the cost map, chat returns 410"""
+        model_info, updated_model = responses_api_bridge_check(
+            model="grok-not-in-cost-map",
+            custom_llm_provider="xai",
+            web_search_options={"search_context_size": "medium"},
+        )
+
+        assert model_info.get("mode") == "responses"
+        assert updated_model == "grok-not-in-cost-map"
+
     @patch("litellm.completion_extras.responses_api_bridge.completion")
     def test_completion_with_tools_routes_to_responses_api(
         self, mock_responses_completion