Error when supplying model with function (tool) call result obtained via the `responses.parse` method

[JovanVeljanoski]

Confirm this is an issue with the Python library and not an underlying OpenAI API

• This is an issue with the Python library

Describe the bug

Hi team!

I believe there is a bug when trying to update the message list with a function call results when one is using the responses.parse method.

Below I submit a working example that you can run to see the error. I also have a workaround.
The issue comes here:
When attaching the output to the message history, if one is using responses.parsed the output is of type ParsedResponseFunctionToolCall which contains the parsed_arguments property. The existence of this property raises and error.

openai.BadRequestError: Error code: 400 - {'error': {'message': "Unknown parameter: 'input[2].parsed_arguments'.", 'type': 'invalid_request_error', 'param': 'input[2].parsed_arguments', 'code': 'unknown_parameter'}}

Workaround:
This is a simple workaround: all you need to do is re-create the response output while removing the parsed_arguments:

mapping = dict(response.output[0])
del mapping['parsed_arguments']
messages.append(ParsedResponseFunctionToolCall(**mapping))

After this workaround - everything works as expected - but I think this should not be needed!

---

Honestly, i do not know if this is an issue with the openai python library or with the API, but I assume it is with the library.

---

To Reproduce

Execute the following code below to reproduce the issue.

The main bug here:

import openai
from openai.types.responses import ParsedResponseFunctionToolCall

client = openai.OpenAI()


messages = [
    {'role': 'developer', 'content': 'You are a pirate captain. You are a seasoned captain, and you give clear commands and updates.'},
    {'role': 'user', 'content': 'Do we expect a storm tonight captain?'},
    ParsedResponseFunctionToolCall(arguments='{}', call_id='call_DXbTc6kVdiZw16mSoNvWrYWD', name='CheckWeather', type='function_call', id='fc_682f45de491881988e8ab6de03e55c170900eec136ebb9a8', status='completed', parsed_arguments={'function': 'CheckWeather'}),
    {'type': 'function_call_output', 'call_id': 'call_DXbTc6kVdiZw16mSoNvWrYWD', 'output': 'Steady wind from the north, sunny, no clouds at all.'}
]


response = client.responses.parse(
    input=messages,
    model='gpt-4.1-nano-2025-04-14',
    store=False,
)

print(response.output_text)

The following example also fails, even though parsed_arguments is None (but it is explicitly specified)

messages = [
    {'role': 'developer', 'content': 'You are a pirate captain. You are a seasoned captain, and you give clear commands and updates.'},
    {'role': 'user', 'content': 'Do we expect a storm tonight captain?'},
    ParsedResponseFunctionToolCall(arguments='{}', call_id='call_DXbTc6kVdiZw16mSoNvWrYWD', name='CheckWeather', type='function_call', id='fc_682f45de491881988e8ab6de03e55c170900eec136ebb9a8', status='completed', parsed_arguments=None),
    {'type': 'function_call_output', 'call_id': 'call_DXbTc6kVdiZw16mSoNvWrYWD', 'output': 'Steady wind from the north, sunny, no clouds at all.'}
]


response = client.responses.parse(
    input=messages,
    model='gpt-4.1-nano-2025-04-14',
    store=False,
)

However, if we do not specify parsed_arguments, everything works as expected (hence the workaround)

messages = [
    {'role': 'developer', 'content': 'You are a pirate captain. You are a seasoned captain, and you give clear commands and updates.'},
    {'role': 'user', 'content': 'Do we expect a storm tonight captain?'},
    ParsedResponseFunctionToolCall(arguments='{}', call_id='call_DXbTc6kVdiZw16mSoNvWrYWD', name='CheckWeather', type='function_call', id='fc_682f45de491881988e8ab6de03e55c170900eec136ebb9a8', status='completed'),
    {'type': 'function_call_output', 'call_id': 'call_DXbTc6kVdiZw16mSoNvWrYWD', 'output': 'Steady wind from the north, sunny, no clouds at all.'}
]


response = client.responses.parse(
    input=messages,
    model='gpt-4.1-nano-2025-04-14',
    store=False,
)

Code snippets

import openai
from openai.types.responses import ParsedResponseFunctionToolCall

client = openai.OpenAI()


messages = [
    {'role': 'developer', 'content': 'You are a pirate captain. You are a seasoned captain, and you give clear commands and updates.'},
    {'role': 'user', 'content': 'Do we expect a storm tonight captain?'},
    ParsedResponseFunctionToolCall(arguments='{}', call_id='call_DXbTc6kVdiZw16mSoNvWrYWD', name='CheckWeather', type='function_call', id='fc_682f45de491881988e8ab6de03e55c170900eec136ebb9a8', status='completed', parsed_arguments={'function': 'CheckWeather'}),
    {'type': 'function_call_output', 'call_id': 'call_DXbTc6kVdiZw16mSoNvWrYWD', 'output': 'Steady wind from the north, sunny, no clouds at all.'}
]


response = client.responses.parse(
    input=messages,
    model='gpt-4.1-nano-2025-04-14',
    store=False,
)

print(response.output_text)

OS

macOS 15.4.1

Python version

Python v3.12.0

Library version

openai v1.81.0