fix(bedrock): honor the json_mode kwarg on invoke Nova structured output and keep gpt-4o-mini off chat web search

litellm/llms/bedrock/chat/converse_transformation.py

@@ -1805,6 +1805,7 @@ class AmazonConverseConfig(BaseConfig):
             data=request_data,
             messages=messages,
             encoding=encoding,
+            json_mode=json_mode,
         )
 
     def _transform_reasoning_content(self, reasoning_content_blocks: list[BedrockConverseReasoningContentBlock]) -> str:
@@ -2237,6 +2238,7 @@ class AmazonConverseConfig(BaseConfig):
         data: dict | str,
         messages: list,
         encoding,
+        json_mode: bool | None = None,
     ) -> ModelResponse:
         ## LOGGING
         if logging_obj is not None:
@@ -2247,7 +2249,9 @@ class AmazonConverseConfig(BaseConfig):
                 additional_args={"complete_input_dict": data},
             )
 
-        json_mode: Final[bool | None] = optional_params.get("json_mode", None)
+        resolved_json_mode: Final[bool | None] = (
+            json_mode if json_mode is not None else optional_params.get("json_mode", None)
+        )
         ## RESPONSE OBJECT
         try:
             completion_response: Final = ConverseResponseBlock(**response.json())
@@ -2339,7 +2343,7 @@ class AmazonConverseConfig(BaseConfig):
             chat_completion_message["thinking_blocks"] = self._transform_thinking_blocks(reasoningContentBlocks)
         chat_completion_message["content"] = content_str
         filtered_tools: Final = self._filter_json_mode_tools(
-            json_mode=json_mode,
+            json_mode=resolved_json_mode,
             tools=tools,
             chat_completion_message=chat_completion_message,
         )
@@ -2363,7 +2367,7 @@ class AmazonConverseConfig(BaseConfig):
         # When json_mode filtered out all synthetic tool calls the response
         # is plain content, not a pending tool invocation. Fix finish_reason
         # so callers (e.g. OpenAI SDK) don't misinterpret it.
-        if json_mode and not filtered_tools and tools:
+        if resolved_json_mode and not filtered_tools and tools:
             initial_finish_reason = "stop"
 
         (

litellm/model_prices_and_context_window_backup.json

@@ -29407,8 +29407,7 @@
             "search_context_size_high": 0.025,
             "search_context_size_low": 0.025,
             "search_context_size_medium": 0.025
-        },
-        "supports_web_search": true
+        }
     },
     "gpt-4o-mini-2024-07-18": {
         "cache_read_input_token_cost": 7.5e-08,
@@ -29436,8 +29435,7 @@
         "supports_response_schema": true,
         "supports_system_messages": true,
         "supports_tool_choice": true,
-        "supports_vision": true,
-        "supports_web_search": true
+        "supports_vision": true
     },
     "gpt-4o-mini-audio-preview": {
         "deprecation_date": "2026-05-07",

model_prices_and_context_window.json

@@ -29407,8 +29407,7 @@
             "search_context_size_high": 0.025,
             "search_context_size_low": 0.025,
             "search_context_size_medium": 0.025
-        },
-        "supports_web_search": true
+        }
     },
     "gpt-4o-mini-2024-07-18": {
         "cache_read_input_token_cost": 7.5e-08,
@@ -29436,8 +29435,7 @@
         "supports_response_schema": true,
         "supports_system_messages": true,
         "supports_tool_choice": true,
-        "supports_vision": true,
-        "supports_web_search": true
+        "supports_vision": true
     },
     "gpt-4o-mini-audio-preview": {
         "deprecation_date": "2026-05-07",

tests/test_litellm/llms/bedrock/chat/test_converse_transformation.py

@@ -6727,3 +6727,41 @@ def test_forced_tool_choice_forwarded_on_converse_models_that_support_it(
     )
 
     assert result == {"any": {}}
+
+
+def test_transform_response_honors_json_mode_kwarg_when_optional_params_lack_it():
+    response_json = {
+        "metrics": {"latencyMs": 900},
+        "output": {
+            "message": {
+                "content": [
+                    {
+                        "toolUse": {
+                            "input": {"city": "Paris", "population": 2100000},
+                            "name": "json_tool_call",
+                            "toolUseId": "tooluse_invoke_nova_json",
+                        }
+                    }
+                ],
+                "role": "assistant",
+            }
+        },
+        "stopReason": "tool_use",
+        "usage": {"inputTokens": 40, "outputTokens": 20, "totalTokens": 60},
+    }
+    raw_response = httpx.Response(200, json=response_json, request=httpx.Request("POST", "https://bedrock.test"))
+    logging_obj = MagicMock()
+    result = AmazonConverseConfig().transform_response(
+        model="bedrock/invoke/us.amazon.nova-micro-v1:0",
+        raw_response=raw_response,
+        model_response=ModelResponse(),
+        logging_obj=logging_obj,
+        request_data={},
+        messages=[],
+        optional_params={"tools": [{"type": "function", "function": {"name": "json_tool_call", "parameters": {}}}]},
+        litellm_params={},
+        encoding=None,
+        json_mode=True,
+    )
+    assert result.choices[0].message.tool_calls is None
+    assert json.loads(result.choices[0].message.content) == {"city": "Paris", "population": 2100000}

whitelisted_bedrock_models.txt

@@ -6,6 +6,7 @@ ai21.jamba-instruct-v1:0
 twelvelabs.pegasus-1-2-v1:0
 us.twelvelabs.pegasus-1-2-v1:0
 eu.twelvelabs.pegasus-1-2-v1:0
+global.twelvelabs.pegasus-1-2-v1:0
 amazon.titan-text-express-v1
 amazon.titan-text-lite-v1
 amazon.titan-text-premier-v1:0