Automatically use OpenAI strict mode for strict-compatible native output types (#2447)

Author: DouweM

pydantic_ai_slim/pydantic_ai/models/__init__.py

@@ -790,12 +790,18 @@ def get_user_agent() -> str:
 def _customize_tool_def(transformer: type[JsonSchemaTransformer], t: ToolDefinition):
     schema_transformer = transformer(t.parameters_json_schema, strict=t.strict)
     parameters_json_schema = schema_transformer.walk()
-    if t.strict is None:
-        t = replace(t, strict=schema_transformer.is_strict_compatible)
-    return replace(t, parameters_json_schema=parameters_json_schema)
+    return replace(
+        t,
+        parameters_json_schema=parameters_json_schema,
+        strict=schema_transformer.is_strict_compatible if t.strict is None else t.strict,
+    )
 
 
 def _customize_output_object(transformer: type[JsonSchemaTransformer], o: OutputObjectDefinition):
-    schema_transformer = transformer(o.json_schema, strict=True)
-    son_schema = schema_transformer.walk()
-    return replace(o, json_schema=son_schema)
+    schema_transformer = transformer(o.json_schema, strict=o.strict)
+    json_schema = schema_transformer.walk()
+    return replace(
+        o,
+        json_schema=json_schema,
+        strict=schema_transformer.is_strict_compatible if o.strict is None else o.strict,
+    )

tests/models/test_openai.py

@@ -12,7 +12,7 @@
 import pytest
 from dirty_equals import IsListOrTuple
 from inline_snapshot import snapshot
-from pydantic import AnyUrl, BaseModel, Discriminator, Field, Tag
+from pydantic import AnyUrl, BaseModel, ConfigDict, Discriminator, Field, Tag
 from typing_extensions import NotRequired, TypedDict
 
 from pydantic_ai import Agent, ModelHTTPError, ModelRetry, UnexpectedModelBehavior
@@ -1777,6 +1777,44 @@ class MyModel(BaseModel):
     )
 
 
+def test_native_output_strict_mode(allow_model_requests: None):
+    class CityLocation(BaseModel):
+        city: str
+        country: str
+
+    c = completion_message(
+        ChatCompletionMessage(content='{"city": "Mexico City", "country": "Mexico"}', role='assistant'),
+    )
+    mock_client = MockOpenAI.create_mock(c)
+    model = OpenAIModel('gpt-4o', provider=OpenAIProvider(openai_client=mock_client))
+
+    # Explicit strict=True
+    agent = Agent(model, output_type=NativeOutput(CityLocation, strict=True))
+
+    agent.run_sync('What is the capital of Mexico?')
+    assert get_mock_chat_completion_kwargs(mock_client)[-1]['response_format']['json_schema']['strict'] is True
+
+    # Explicit strict=False
+    agent = Agent(model, output_type=NativeOutput(CityLocation, strict=False))
+
+    agent.run_sync('What is the capital of Mexico?')
+    assert get_mock_chat_completion_kwargs(mock_client)[-1]['response_format']['json_schema']['strict'] is False
+
+    # Strict-compatible
+    agent = Agent(model, output_type=NativeOutput(CityLocation))
+
+    agent.run_sync('What is the capital of Mexico?')
+    assert get_mock_chat_completion_kwargs(mock_client)[-1]['response_format']['json_schema']['strict'] is True
+
+    # Strict-incompatible
+    CityLocation.model_config = ConfigDict(extra='allow')
+
+    agent = Agent(model, output_type=NativeOutput(CityLocation))
+
+    agent.run_sync('What is the capital of Mexico?')
+    assert get_mock_chat_completion_kwargs(mock_client)[-1]['response_format']['json_schema']['strict'] is False
+
+
 async def test_openai_instructions(allow_model_requests: None, openai_api_key: str):
     m = OpenAIModel('gpt-4o', provider=OpenAIProvider(api_key=openai_api_key))
     agent = Agent(m, instructions='You are a helpful assistant.')