TypeError: Cannot instantiate typing.Union when calling tools due to misuse of TypeAlias in chatcmpl_converter.py

[Aligadit1]

Description:
When running an agent with a tool call, the agents SDK crashes with:
TypeError: Cannot instantiate typing.Union

Python: 3.13
openai (version : 1.99.6)
openai-agents (version: 0.0.16 )
OS ( Windows:11)

Full Traceback (shortened):
  File "...\agents\models\chatcmpl_converter.py", line 443, in items_to_messages
    new_tool_call = ChatCompletionMessageToolCallParam(...)
TypeError: Cannot instantiate typing.Union

Root cause:
ChatCompletionMessageToolCallParam is defined in openai.types.chat as:

ChatCompletionMessageToolCallParam: TypeAlias = Union[
    ChatCompletionMessageFunctionToolCallParam,
    ChatCompletionMessageCustomToolCallParam
]
This is a TypeAlias for a Union, not a class.
The converter calls it as if it were a constructor, which is invalid in Python:

new_tool_call = ChatCompletionMessageToolCallParam(...)  # ❌
This triggers the error because typing.Union is not instantiable.

Expected behavior:

The converter should inspect the data, determine whether it matches
ChatCompletionMessageFunctionToolCallParam or ChatCompletionMessageCustomToolCallParam,
and then instantiate the correct concrete type:

Solution:

elif func_call := cls.maybe_function_tool_call(item):
                asst = ensure_assistant_message()
                tool_calls = list(asst.get("tool_calls", []))
                arguments = func_call["arguments"] if func_call["arguments"] else "{}"
                if func_call.get("type") == "function_call":
                    new_tool_call = ChatCompletionMessageFunctionToolCallParam(
                        id=func_call["call_id"],
                        type="function",
                        function={
                            "name": func_call["name"],
                            "arguments": arguments,
                        },
                    )
                else:
                    new_tool_call = ChatCompletionMessageCustomToolCallParam(
                        id=func_call["call_id"],
                        type="function",
                        custom={
                            "name": func_call["name"],
                            "arguments": arguments,
                        },
                    )
                tool_calls.append(new_tool_call)
                asst["tool_calls"] = tool_calls

Steps to reproduce:

Create an agent with Runner.run()

Add a tool call to the conversation.

Run and observe the crash when the model returns a tool call.

Additional note:
I’m happy to open a PR with a fix for this in the repo. Please let me know if that would help — I can contribute the change and relevant test cases.I have attached a photo which shows my code worked after doing the solution i provide.

[Aligadit1]

I will be very happy to contribute

[seratch]

Thanks for reporting this issue. We're aware of this and #1415 will resolve it in the next release.