Pydantic AI doesn't work with openai>=1.99.2 due to changes in the tool types mappings

[restless]

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

I'm using pydantic 0.6.2 and after upgrading to the latest openai (greater than 1.99.1) I started getting the following error:

File "/usr/local/lib/python3.13/site-packages/pydantic_ai/models/openai.py", line 315, in _completions_create
openai_messages = await self._map_messages(messages)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.13/site-packages/pydantic_ai/models/openai.py", line 486, in _map_messages
     tool_calls.append(self._map_tool_call(item))
                       ~~~~~~~~~~~~~~~~~~~^^^^^^
   File "/usr/local/lib/python3.13/site-packages/pydantic_ai/models/openai.py", line 508, in _map_tool_call
     return chat.ChatCompletionMessageToolCallParam(
            ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
         id=_guard_tool_call_id(t=t),
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
         type='function',
         ^^^^^^^^^^^^^^^^
         function={'name': t.tool_name, 'arguments': t.args_as_json_str()},
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     )
     ^
   File "/usr/local/lib/python3.13/typing.py", line 1317, in __call__
     result = self.__origin__(*args, **kwargs)
   File "/usr/local/lib/python3.13/typing.py", line 560, in __call__
     raise TypeError(f"Cannot instantiate {self!r}")
 TypeError: Cannot instantiate typing.Union

My code is similar to the one from here: https://ai.pydantic.dev/message-history/#storing-and-loading-messages-to-json - I'm simply saving and restoring messages history.

What happens:

1. The first message I send to my PydanticAI agent is processed correctly
2. The response to the above call is a "final_result" tool call
3. the next message I send to the agent, while using messages_history that contains the final_result tool call fails with the above error.

Sample messages:

[
  {
    "kind": "request",
    "parts": [
      {
        "content": "hi there",
        "part_kind": "user-prompt",
        "timestamp": "2025-08-11T11:20:22.522615Z"
      }
    ],
    "instructions": "..."
  },
  {
    "kind": "response",
    "parts": [
      {
        "args": "{\"message\":\"Hello!\"}",
        "part_kind": "tool-call",
        "tool_name": "final_result",
        "tool_call_id": "call_VFqZ3Dm1uhCbOIJmHXaL0BCL"
      }
    ],
    "usage": {
      "details": {
        "audio_tokens": 0,
        "cached_tokens": 0,
        "reasoning_tokens": 0,
        "accepted_prediction_tokens": 0,
        "rejected_prediction_tokens": 0
      },
      "requests": 1,
      "total_tokens": 820,
      "request_tokens": 805,
      "response_tokens": 15
    },
    "timestamp": "2025-08-11T11:20:22Z",
    "vendor_id": "chatcmpl-C3KreKto8WewObphvxayf2AukUu6q",
    "model_name": "gpt-4.1-mini-2025-04-14",
    "vendor_details": null
  },
  {
    "kind": "request",
    "parts": [
      {
        "content": "Final result processed.",
        "metadata": null,
        "part_kind": "tool-return",
        "timestamp": "2025-08-11T11:20:23.254464Z",
        "tool_name": "final_result",
        "tool_call_id": "call_VFqZ3Dm1uhCbOIJmHXaL0BCL"
      }
    ],
    "instructions": null
  },

Now, when I use this as message_history for the next message the aforementioned exception is raised.

I tracked this down to the 1.99.2 version of Open AI where this change seems suspicious to me: openai/openai-python@0c57bd7

Python, Pydantic AI & LLM client version

pydantic AI: 0.6.2

[lpb-conseil]

Same here

[erik-wang21]

Getting same issue.

[HamzaFarhan]

uv add openai==1.99.7 should work

[restless]

Thanks @HamzaFarhan. Things are back to normal with openai>=1.99.7. Closing the issue.