fix: complete anthropic json mode structured output handling

litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py

@@ -589,6 +589,7 @@ def convert_to_model_response_object(  # noqa: PLR0915
                                 choice["message"]["content"] = json_mode_content_str
                             elif isinstance(choice["message"].get("content"), str):
                                 choice["message"]["content"] += f"\n{json_mode_content_str}"
+                            finish_reason = "tool_calls"
                 if message is None:
                     # Preserve provider_specific_fields if already present
                     # in the response (e.g. from proxy passthrough)

litellm/llms/anthropic/chat/transformation.py

@@ -1456,28 +1456,8 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
             elif param == "top_p":
                 optional_params["top_p"] = value
             elif param == "response_format" and isinstance(value, dict):
-                if any(
-                    substring in model
-                    for substring in {
-                        "sonnet-4.5",
-                        "sonnet-4-5",
-                        "opus-4.1",
-                        "opus-4-1",
-                        "opus-4.5",
-                        "opus-4-5",
-                        "opus-4.6",
-                        "opus-4-6",
-                        "opus-4.7",
-                        "opus-4-7",
-                        "sonnet-4.6",
-                        "sonnet-4-6",
-                        "sonnet_4.6",
-                        "sonnet_4_6",
-                        "haiku-4.5",
-                        "haiku-4-5",
-                        "haiku_4.5",
-                        "haiku_4_5",
-                    }
+                if litellm.supports_response_schema(
+                    model=model, custom_llm_provider="anthropic"
                 ):
                     _output_format = (
                         self.map_response_format_to_anthropic_output_format(value)
@@ -2443,7 +2423,9 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
         _message.provider_specific_fields = provider_specific_fields
 
         if json_mode_message is not None:
-            completion_response["stop_reason"] = "stop"
+            completion_response["stop_reason"] = (
+                "tool_use" if json_mode_message.tool_calls else "stop"
+            )
             _message = json_mode_message
 
         model_response.choices[0].message = _message

tests/llm_translation/test_llm_response_utils/test_convert_dict_to_chat_completion.py

@@ -40,7 +40,7 @@ def test_convert_to_model_response_object_basic():
                     "content": "Hi there! How can I assist you today?",
                     "refusal": None,
                 },
-                "finish_reason": "stop",
+                "finish_reason": None,
             }
         ],
         "usage": {
@@ -344,7 +344,7 @@ def test_convert_to_model_response_object_json_mode():
                         }
                     ],
                 },
-                "finish_reason": None,
+                "finish_reason": "stop",
             }
         ],
         "usage": {"total_tokens": 10, "prompt_tokens": 5, "completion_tokens": 5},
@@ -374,6 +374,61 @@ def test_convert_to_model_response_object_json_mode():
     assert result.usage.completion_tokens == 5
 
 
+def test_convert_to_model_response_object_json_mode_with_parallel_real_tool():
+    model_response_object = ModelResponse(model="gpt-3.5-turbo")
+    from litellm.constants import RESPONSE_FORMAT_TOOL_NAME
+
+    response_object = {
+        "choices": [
+            {
+                "message": {
+                    "role": "assistant",
+                    "content": None,
+                    "tool_calls": [
+                        {
+                            "id": "call_real",
+                            "type": "function",
+                            "function": {
+                                "arguments": '{"movie":"Inception"}',
+                                "name": "get_showtimes",
+                            },
+                        },
+                        {
+                            "id": "call_json",
+                            "type": "function",
+                            "function": {
+                                "arguments": '{"title":"Inception"}',
+                                "name": RESPONSE_FORMAT_TOOL_NAME,
+                            },
+                        },
+                    ],
+                },
+                "finish_reason": "stop",
+            }
+        ],
+        "usage": {"total_tokens": 10, "prompt_tokens": 5, "completion_tokens": 5},
+        "model": "gpt-3.5-turbo",
+    }
+
+    result = convert_to_model_response_object(
+        model_response_object=model_response_object,
+        response_object=response_object,
+        stream=False,
+        start_time=datetime.now(),
+        end_time=datetime.now(),
+        hidden_params=None,
+        _response_headers=None,
+        convert_tool_call_to_json_mode=True,
+    )
+
+    assert result.choices[0].message.content == '{"title":"Inception"}'
+    assert result.choices[0].finish_reason == "tool_calls"
+    tool_calls = result.choices[0].message.tool_calls
+    assert tool_calls is not None
+    assert len(tool_calls) == 1
+    assert tool_calls[0].function.name == "get_showtimes"
+
+
 def test_convert_to_model_response_object_function_output():
     """
     Test conversion with function output.

tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py

@@ -73,6 +73,34 @@ def test_anthropic_json_mode_non_streaming_mixed_internal_and_user_tools():
     assert extra == '{"answer": 42}'
 
 
+def test_haiku_45_uses_model_metadata_for_native_structured_output():
+    config = AnthropicConfig()
+    mapped_params = config.map_openai_params(
+        non_default_params={
+            "response_format": {
+                "type": "json_schema",
+                "json_schema": {
+                    "name": "MovieReview",
+                    "strict": True,
+                    "schema": {
+                        "type": "object",
+                        "properties": {"title": {"type": "string"}},
+                        "required": ["title"],
+                        "additionalProperties": False,
+                    },
+                },
+            }
+        },
+        optional_params={},
+        model="claude-haiku-4-5-20251001",
+        drop_params=False,
+    )
+
+    assert "output_format" in mapped_params
+    assert "tools" not in mapped_params
+    assert "tool_choice" not in mapped_params
+
+
 def test_calculate_usage():
     """
     Do not include cache_creation_input_tokens in the prompt_tokens