Error when using GPT-5 with `output_type` and built-in tools because `tool_choice=required` is not supported

[JacobTheis]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI
• I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue

Description

GPT-5 does not support tool_choice = "required"

ModelHTTPError: status_code: 400, model_name: gpt-5, body: {'message': "Tool choices other than 'auto' are not supported with model 'gpt-5' and the following tool types: 'web_search_preview'.", 'type': 'invalid_request_error', 'param': 'tool_choice', 'code': None}

Running a multi-agent setup where one of the called agents also attempts to use the OpenAI built-in websearch tool always yields this error.

Example Code

from pydantic_ai import Agent, RunContext
from pydantic_ai.models.openai import OpenAIResponsesModel, OpenAIResponsesModelSettings
from openai.types.responses import WebSearchToolParam

from pydantic_ai.usage import UsageLimits

openai_model = OpenAIResponsesModel('gpt-5')
openai_model_settings = OpenAIResponsesModelSettings(
    openai_builtin_tools=[WebSearchToolParam(type='web_search_preview')],
)

joke_selection_agent = Agent(
    model=openai_model,
    system_prompt=(
        'Use the `joke_factory` to generate some jokes, then choose the best. '
        'You must return just a single joke.'
    ),
)
joke_generation_agent = Agent(
    model=openai_model,
    model_settings=openai_model_settings,
    output_type=list[str]
)


@joke_selection_agent.tool
async def joke_factory(ctx: RunContext[None], count: int) -> list[str]:
    r = await joke_generation_agent.run(
        f'Please generate {count} jokes. Search the web for the best jokes.',
        usage=ctx.usage,
    )
    return r.output


result = joke_selection_agent.run_sync(
    'Tell me a joke.',
    usage_limits=UsageLimits(request_limit=5, total_tokens_limit=90000),
)
print(result.output)
#> Did you hear about the toothpaste scandal? They called it Colgate.
print(result.usage())
#> Usage(requests=3, request_tokens=204, response_tokens=24, total_tokens=228)

Python, Pydantic AI & LLM client version

Python 13.12.2
Pydantic 0.6.2
LLM OpenAI Responses Api

[DouweM]

@JacobTheis This is a consequence of output_type using the default tool output mode which forces tool_choice=required.

You can work around this by using the native output mode instead: https://ai.pydantic.dev/output/#native-output

from pydantic_ai.output import NativeOutput

# ...

joke_generation_agent = Agent(
    model=openai_model,
    model_settings=openai_model_settings,
    output_type=NativeOutput(list[str]),
)

Judging by openai/openai-python#2537, this is also the case for built-in tools code_interpreter and file_search.

With #2102, these errors can also be hit when using Agent(builtin_tools=...) rather than OpenAIResponsesModelSettings(openai_builtin_tools=...).

I think we should raise an error instructing the user to use NativeOutput when this conflict is found on a model that's known to have this restriction (which we can track using a new OpenAIModelProfile option).

@JacobTheis Would you be up for filing a PR to that effect? :)