Description
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
When using Pydantic’s Field to include metadata such as title or description in nested models, the generated JSON schema does not properly set additionalProperties: false for $ref-referenced types when they are inlined. This causes a BadRequestError (400) from the API, with the following error message:
BadRequestError: Error code: 400 - {'error': {'message': "Invalid schema for response_format 'Universe': In context=('properties', 'largest_star'), 'additionalProperties' is required to be supplied and to be false.", 'type': 'invalid_request_error', 'param': 'response_format', 'code': None}}
The root cause is that when nested objects are referenced using $defs in the JSON schema and subsequently inlined via $ref, the additionalProperties: false setting is omitted, which violates the expected strict schema requirements.
The JSON schema generated from the Universe class used in the code snippet below looks like this. (additionalProperties: false is not set for Galaxy.largest_star.)
{
"$defs": {
"Galaxy": {
"properties": {
"name": {
"description": "The name of the galaxy.",
"title": "Name",
"type": "string"
},
"largest_star": {
# `"additionalProperties": false` is missing.
"description": "The largest star in the galaxy.",
"properties": {
"name": {
"description": "The name of the star.",
"title": "Name",
"type": "string"
}
},
"required": ["name"],
"title": "Star",
"type": "object"
}
},
"required": ["name", "largest_star"],
"title": "Galaxy",
"type": "object",
"additionalProperties": false
},
"Star": {
"properties": {
"name": {
"description": "The name of the star.",
"title": "Name",
"type": "string"
}
},
"required": ["name"],
"title": "Star",
"type": "object",
"additionalProperties": false
}
},
"properties": {
"name": {
"description": "The name of the universe.",
"title": "Name",
"type": "string"
},
"galaxy": {
"description": "A galaxy in the universe.",
"properties": {
"name": {
"description": "The name of the galaxy.",
"title": "Name",
"type": "string"
},
"largest_star": {
"description": "The largest star in the galaxy.",
"properties": {
"name": {
"description": "The name of the star.",
"title": "Name",
"type": "string"
}
},
"required": ["name"],
"title": "Star",
"type": "object"
}
},
"required": ["name", "largest_star"],
"title": "Galaxy",
"type": "object",
"additionalProperties": false
}
},
"required": ["name", "galaxy"],
"title": "Universe",
"type": "object",
"additionalProperties": false
}This issue occurs when using the AzureOpenAI client and is likely reproducible with the standard OpenAI client as well, as they both share the underlying schema handling mechanism.
To Reproduce
- Initialize an AzureOpenAI client (or an OpenAI client).
- Create a nested Pydantic model structure with Field metadata (title and description).
- Pass the top-level model as the response_format argument.
- Observe the 400 error indicating the missing additionalProperties: false for nested objects.
Code snippets
from typing import Annotated
from openai import AzureOpenAI
from pydantic import BaseModel, Field
class Star(BaseModel):
name: Annotated[str, Field(description="The name of the star.")]
class Galaxy(BaseModel):
name: Annotated[str, Field(description="The name of the galaxy.")]
largest_star: Annotated[Star, Field(description="The largest star in the galaxy.")]
class Universe(BaseModel):
name: Annotated[str, Field(description="The name of the universe.")]
galaxy: Annotated[Galaxy, Field(description="A galaxy in the universe.")]
client = AzureOpenAI(azure_endpoint="endpoint", api_key="api-key", api_version="api-version")
prompt = "Create a fictional universe for a science fiction novel."
completion = client.beta.chat.completions.parse(
messages=[
{"role": "user", "content": prompt},
],
model="gpt-4o",
response_format=Universe,
)OS
macOS
Python version
Python v3.12.8
Library version
openai v1.59.7