Schema validation error

[DamyanBG]

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

Hello,

I can see something interesting. I have Pydantic model:

class ReviewSentiment(BaseModel):
    sentiment: str
    tags: List[str]
    aspect_sentiment: Dict[str, str]

When I am trying that code:

prompt = f"""
    Analyze the following product review and return a JSON object with these fields:
    - sentiment: Overall sentiment as one of 'very positive', 'positive', 'neutral', 'negative', 'very negative'.
    - tags: List of up to 3 main topics or issues (e.g., ['shipping', 'quality', 'customer_service']).
    -     aspect_sentiment: dict[str, str]  # Maps aspect (like 'shipping', 'quality') to its sentiment (e.g., {{'shipping': 'positive', 'quality': 'negative'}}).

    Review rating (1-5): {review.rating}
    Review title: {review.title}
    Review content: {review.content}
    Return JSON only.
    """
    client = OpenAI()

    response = client.responses.parse(
        model="gpt-4o",
        input=[
            {
                "role": "system",
                "content": prompt,
            },
        ],
        text_format=ReviewSentiment,
    )

    review_sentiment = response.output_parsed
    print(review_sentiment)

I am receiveing error:

\.venv\Lib\site-packages\openai\_base_client.py", line 1052, in request
    raise self._make_status_error_from_response(err.response) from None
openai.BadRequestError: Error code: 400 - {'error': {'message': "Invalid schema for response_format 'ReviewSentiment': In context=(), 'required' is required to be supplied and to be an array including every key in properties. Extra required key 'aspect_sentiment' supplied.", 'type': 'invalid_request_error', 'param': 'text.format.schema', 'code': 'invalid_json_schema'}}

I actually debugged the code a little bit and I can see the dict schema, which then openai SDK creates:

{
    "input": [
        {
            "role": "system",
            "content": "\n    Analyze the following product review and return a JSON object with these fields:\n    - sentiment: Overall sentiment as one of 'very positive', 'positive', 'neutral', 'negative', 'very negative'.\n    - tags: List of up to 3 main topics or issues (e.g., ['shipping', 'quality', 'customer_service']).\n    -     aspect_sentiment: dict[str, str]  # Maps aspect (like 'shipping', 'quality') to its sentiment (e.g., {'shipping': 'positive', 'quality': 'negative'}).\n\n    Review rating (1-5): 5\n    Review title: Very good\n    Review content: I luked the product very much, delivery was fast\n    Return JSON only.\n    ",
        }
    ],
    "model": "gpt-4o",
    "text": {
        "format": {
            "type": "json_schema",
            "strict": True,
            "name": "ReviewSentiment",
            "schema": {
                "properties": {
                    "sentiment": {"title": "Sentiment", "type": "string"},
                    "tags": {
                        "items": {"type": "string"},
                        "title": "Tags",
                        "type": "array",
                    },
                    "aspect_sentiment": {
                        "additionalProperties": {"type": "string"},
                        "title": "Aspect Sentiment",
                        "type": "object",
                    },
                },
                "required": ["sentiment", "tags", "aspect_sentiment"],
                "title": "ReviewSentiment",
                "type": "object",
                "additionalProperties": False,
            },
        }
    },
}

Everything looks okay, but as I mentioned - I receive an error.

When I tweak a little but the Pydantic model, I do not have error:

class ReviewSentiment(BaseModel):
    sentiment: str
    tags: List[str]
    aspect_sentiment: Dict[str, str] | None = None

And the request:

{
    "input": [
        {
            "role": "system",
            "content": "\\n    Analyze the following product review and return a JSON object with these fields:\\n    - sentiment: Overall sentiment as one of 'very positive', 'positive', 'neutral', 'negative', 'very negative'.\\n    - tags: List of up to 3 main topics or issues (e.g., ['shipping', 'quality', 'customer_service']).\\n    -     aspect_sentiment: dict[str, str]  # Maps aspect (like 'shipping', 'quality') to its sentiment (e.g., {'shipping': 'positive', 'quality': 'negative'}).\\n\\n    Review rating (1-5): 5\\n    Review title: Very good\\n    Review content: I luked the product very much, delivery was fast\\n    Return JSON only.\\n    "
        }
    ],
    "model": "gpt-4o",
    "text": {
        "format": {
            "type": "json_schema",
            "strict": true,
            "name": "ReviewSentiment",
            "schema": {
                "properties": {
                    "sentiment": { "title": "Sentiment", "type": "string" },
                    "tags": { "items": { "type": "string" }, "title": "Tags", "type": "array" },
                    "aspect_sentiment": {
                        "anyOf": [{ "additionalProperties": { "type": "string" }, "type": "object" }, { "type": "null" }],
                        "title": "Aspect Sentiment"
                    }
                },
                "required": ["sentiment", "tags", "aspect_sentiment"],
                "title": "ReviewSentiment",
                "type": "object",
                "additionalProperties": false
            }
        }
    }
}

So that looks very strange at all. I think some error is in the validation in the API, but would like to clarify with you.

To Reproduce

1. Take the Pydantic model, which I provided
2. Make a call like in the coded, which I provided
3. Error occurs

Code snippets

OS

Windows 11

Python version

Python 3.13.5

Library version

openai 1.98.0

[RobertCraigie]

Thanks for the report, does the model give you any data back in the aspect_sentiment property?

[DamyanBG]

When I change aspect_sentiment to Optional , the model gives me data for aspect_sentiment.