strict` mode on function call is currently not supported for grok models

[Luca-Blight]

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

Any ideas how to work around this? Works for mistral okay.

500: Unexpected error: 500: Error processing AI query: 500: Error running data analyst: status_code: 400, model_name: grok-3-mini-fast-latest, body: Invalid request content: strict mode on function call is currently not supported.

Example Code

Python, Pydantic AI & LLM client version

python version is 3.12
pydantic-ai version is 0.2.10
LLM client is xai and model is 'grok-3-latest'(tried other models)

[DouweM]

@Luca-Blight I'm working on fixing exactly this class of issues in #1835 and #1842! Can you verify you're currently doing exactly what's in https://ai.pydantic.dev/models/openai/#grok-xai, with OpenAIModel and OpenAIProvider with custom base_url? In the first PR I'm fixing it for that scenario, and in the second PR there's a new GrokProvider that'll also get this behavior.

[Luca-Blight]

That's great to hear! Here's the code i'm working with, it appears correct based on the docs.

		provider = OpenAIProvider(
			api_key=config.secrets.get("GROK_API_KEY"),
			base_url="https://api.x.ai/v1",
			http_client=httpx.AsyncClient(),
		)
		model = OpenAIModel(
			model_name,
			provider=provider,
		)

switched to mistral for now, so it seems to be particular to grok.

[akarca]

I solved it using pydantic-ai==0.2.12 and the code below @Luca-Blight

model = OpenAIModel(
        "grok-3",
        provider=OpenAIProvider(
            api_key="***",
            base_url="https://api.x.ai/v1",
        ),
)
model.profile.openai_supports_strict_tool_definition = False  # This line fixes

[Luca-Blight]

@akarca thanks for sharing! Seems like a simple bug to fix.

[DouweM]

@Luca-Blight @akarca If you use the new GrokProvider this should work automatically!