Custom model provider agent does not use tool calling when an output type is specified

[anakin-05]

Please read this first

• Have you read the custom model provider docs, including the 'Common issues' section? Model provider docs
• Have you searched for related issues? Others may have faced similar issues.

Yes, I have reviewed all the relevant documentation.

Describe the question

When using a custom model provider, the agent generates a response without calling a tool if an output type is specified (1st run). However, when I remove the output type, the model correctly calls the intended tool (2nd run). I also attempted to enforce tool usage by setting tool_use_behavior to 'required' and specifying a named tool (3rd run), but this did not resolve the issue.

With the OpenAI model, the agent behaves as expected.

Debug information

• Agents SDK version: v0.0.6
• Python version: 3.12.9
• Model provider: Ollama
• Model name: qwq:32b (supports tool calling)

Repro steps

from __future__ import annotations

import asyncio
import os
from dataclasses import dataclass

import mlflow
from openai import AsyncOpenAI

from agents import (
    Agent,
    Model,
    ModelProvider,
    OpenAIChatCompletionsModel,
    RunConfig,
    Runner,
    function_tool,
    set_tracing_disabled,
)

BASE_URL = os.getenv("EXAMPLE_BASE_URL")
API_KEY = os.getenv("EXAMPLE_API_KEY")
MODEL_NAME = os.getenv("EXAMPLE_MODEL_NAME")

if not BASE_URL or not API_KEY or not MODEL_NAME:
    raise ValueError(
        "Please set EXAMPLE_BASE_URL, EXAMPLE_API_KEY, EXAMPLE_MODEL_NAME via env var or code."
    )

client = AsyncOpenAI(base_url=BASE_URL, api_key=API_KEY)


class CustomModelProvider(ModelProvider):
    def get_model(self, model_name: str | None) -> Model:
        return OpenAIChatCompletionsModel(
            model=model_name or MODEL_NAME, openai_client=client
        )


CUSTOM_MODEL_PROVIDER = CustomModelProvider()


@dataclass
class Weather:
    city: str
    date: str
    weather: str
    temperature: int | None = None


@function_tool
def get_weather(city: str, date: str):
    print(f"[debug] getting weather for {city} on {date}")
    return f"The weather in {city} on {date} is sunny with a high of 75°F."


async def main():
    agent = Agent(
        name="Assistant",
        instructions="You are a helpful assistant.",
        tools=[get_weather],
        output_type=Weather,
        # tool_use_behavior="required",
        # tool_use_behavior="get_weather",
    )

    # This will use the custom model provider
    result = await Runner.run(
        agent,
        "What's the weather in Tokyo on September 3rd 2021?",
        run_config=RunConfig(model_provider=CUSTOM_MODEL_PROVIDER),
    )
    print(result.final_output)


if __name__ == "__main__":
    asyncio.run(main())

Expected behavior

The agent should call the intended tool and generate a response in the specified output format, just as it does with the OpenAI model.

[rm-openai]

@anakin-05 unfortunately this looks like an issue with the custom model you're using? The SDK correctly passes through the params for tool choice, but the model is ignoring it.