beta.chat.completions.parse is not working with response_format json object or json schema after upgrading to 1.55.3

[SunkadGit]

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

Getting Type error
TypeError: No method for generating JsonSchema for core_schema.type='invalid' (expected: GenerateJsonSchema.invalid_schema)

To Reproduce

Use beta.chat.completions.parse with pydantic object in response_format

Code snippets

No response

OS

windows

Python version

3.11

Library version

1.55.3

[RobertCraigie]

Can you share the BaseModel class you're using?

[paulhyuk]

Doesn't work for me either on v1.57.1. The following code (not including imports):

class Response(BaseModel):
    operations: List[Dict]
    explanation: str

completion = client.beta.chat.completions.parse(
    model="gpt-4o-2024-08-06",
    messages=[
        {"role": "system", "content": "You are a data transformation expert that provides precise, actionable JSON responses."},
        {"role": "user", "content": prompt}
    ],
    response_format=Response,
)
response = completion.choices[0].message.parsed

Returns the following error:

    raise self._make_status_error_from_response(err.response) from None
openai.BadRequestError: Error code: 400 - {'error': {'message': "Invalid schema for response_format 'Response': In context=(), 'required' is required to be supplied and to be an array including every key in properties. Extra required key 'operations' supplied.", 'type': 'invalid_request_error', 'param': 'response_format', 'code': None}}

[RobertCraigie]

@paulhyuk unfortunately arbitrary properties aren't supported in structured outputs yet https://platform.openai.com/docs/guides/structured-outputs#additionalproperties-false-must-always-be-set-in-objects, so you'll need to give a different definition for opeerations: List[Dict]

[paulhyuk]

Thanks @RobertCraigie -- does the API support Unions for the structured output? For example the below doesn't seem to work as a response_format

class OperationResponse(BaseModel):
    operation: str
    parameters: Rename|Drop|Create|FillNa|Datetime

[SunkadGit]

For me it worked when downgraded openai library to 1.51.0

This is the Basemodel:
from pydantic import BaseModel
from typing import List

class FieldExtraction(BaseModel):
fieldname: str
fieldvalue: str

class Classification(BaseModel):
page_number: int
classification: str
similarity: float
all_similarities: List[float]
extraction: List[FieldExtraction]
employeeID: str
emailID: str

class Classifications(BaseModel):
classifications: List[Classification]