o1 Model fails when making tool calls

[mtessar]

Thanks for integrating the o1 model. I can use o1 agents without tools, but when I try and use tools, I get an unsupported parameter error due to the parallel_tool_calls being set by the pydantic ai framework.

Here is some code that shows the agent working without tools and produces a BadRequestError when using tools:

agent = Agent(
    'openai:o1',
    system_prompt=(
        'You are a helpful assistant.  Today is 2024-12-12'
    ),
)
result = agent.run_sync("What is today's date?")
print(result)
"""
Today is January 24, 2025.
"""

@agent.tool
def get_joke(str) -> str:
    return 'Why did the chicken cross the road? To get to the other side!'

result = agent.run_sync("Call the `get_joke` function")
"""
---------------------------------------------------------------------------
BadRequestError                           Traceback (most recent call last)
<ipython-input-12-512d4a647163> in <cell line: 0>()
     19     return 'Why did the chicken cross the road? To get to the other side!'
     20 
---> 21 result = agent.run_sync("Call the `get_joke` function")
     22 

10 frames
/usr/local/lib/python3.11/dist-packages/openai/_base_client.py in _request(self, cast_to, options, stream, stream_cls, retries_taken)
   1642 
   1643             log.debug("Re-raising status error")
-> 1644             raise self._make_status_error_from_response(err.response) from None
   1645 
   1646         return await self._process_response(

BadRequestError: Error code: 400 - {'error': {'message': "Unsupported parameter: 'parallel_tool_calls' is not supported with this model.", 'type': 'invalid_request_error', 'param': 'parallel_tool_calls', 'code': 'unsupported_parameter'}}
"""

This code works for openai:gpt-4o but not o1. Thanks!

[sydney-runkle]

@mtessar,

Thanks for the report. I'll work on a fix for this today 👍

[mtessar]

Pretty sure you are the one who deserves the thanks for adding this support @sydney-runkle

[sydney-runkle]

@mtessar, could you please confirm your agent works on #764? I'm getting an awfully long delay in responses, but no error 👍

[mtessar]

@sydney-runkle I will check

[mtessar]

I couldn't figure out how to get the branch installed with pip but I did temporarily edit the line you did and it worked (and was also very slow for me :) )

[izzyacademy]

@mtessar you can you the following steps to build and install the package from the git branch

# Install the tools we will use to build the library
pip install hatch hatchling

# Clone the repository
git clone [email protected]:pydantic/pydantic-ai.git

# Navigate to the repository
cd pydantic-ai/

# Checkout the specific branch [main]
git checkout main (or your specific branch) 

# Build the pydantic packages
hatch build

# Navigate to the folder containing your example programs
# Install the *.whl files from the dist folder
pip install ~/Code/pydantic-ai/*/dist/pydantic_*-0.0.19-py3-none-any.whl

Make sure to run pip uninstall on the pydantic-ai, pydantic-ai-slim and pydantic-graph packages first before installing the *.whl packages

[mtessar]

I was able to get the pydantic-ai installed with the instructions above but pydantic-ai-slim and pydantic-ai-graph are still the pip packages

pydantic-ai @ file:///Users/mtessar/src/pydantic-ai/dist/pydantic_ai-0.0.20-py3-none-any.whl#sha256=52553460e55e194d179c486fd38b60e357609b9b55782af8e9afa1cacd62fbaf
pydantic-ai-slim==0.0.20
pydantic-graph==0.0.20

Sorry!

[elmehalawi]

This doesn't seem to work for o1-mini. I get the following error:

openai.BadRequestError: Error code: 400 - {'error': {'message': "Unsupported value: 'messages[0].role' does not support 'developer' with this model.", 'type': 'invalid_request_error', 'param': 'messages[0].role', 'code': 'unsupported_value'}}

I'm setting up my model like this:

open_ai_model = OpenAIModel("o1-mini", system_prompt_role="developer")

scheduler_agent = Agent(
    open_ai_model,
    deps_type=Dependencies,
    result_type=str,
    system_prompt=system_prompt,
    retries=4,
)

[Kludex]

@elmehalawi This is a different issue: #974

[drdavella]

@Kludex even when accounting for the system call issue in #974 this still does not work. Here's a minimal example:

from pydantic import BaseModel
from pydantic_ai import Agent
from pydantic_ai.models.openai import OpenAIModel

model = OpenAIModel(
    model_name='o1-mini',
    system_prompt_role='user',
)

class YesNoAnswer(BaseModel):
    answer: str

agent = Agent(
    model=model,
    system_prompt='You are a helpful assistant. Provide a yes-no answer to the question.',
    result_type=YesNoAnswer,
)

print(agent.run_sync('Is the sky blue?'))

With the following result:

 File "/Users/username/miniconda3/envs/pydantic-ai/lib/python3.13/site-packages/openai/_base_client.py", line 1651, in _request
    raise self._make_status_error_from_response(err.response) from None
openai.NotFoundError: Error code: 404 - {'error': {'message': 'tools is not supported in this model. For a list of supported models, refer to https://platform.openai.com/docs/guides/function-calling#models-supporting-function-calling.', 'type': 'invalid_request_error', 'param': None, 'code': None}}

[Kludex]

If you remove the result_type, it does. 🤔