Parse structured OpenAI output with Pydantic

Steps

1. Use a model that supports Structured Outputs. Install pydantic alongside openai in your project environment.
2. Pass a Pydantic model as text_format to responses.parse. Handle refusals and incomplete responses before consuming parsed data.

Example

Python · Python API examples require OPENAI_API_KEY and OPENAI_MODEL in your server environment. API calls may incur OpenAI charges.

import os
from openai import OpenAI
from pydantic import BaseModel

class Summary(BaseModel):
    title: str
    points: list[str]

response = OpenAI().responses.parse(
    model=os.environ["OPENAI_MODEL"],
    input="Summarize why HTTP timeouts matter.", text_format=Summary,
)
if response.status != "completed" or response.output_parsed is None:
    raise RuntimeError("No completed structured result; inspect refusal or failure")
print(response.output_parsed.model_dump())

Check the result

Expect a title and a list of points. Schema conformance does not prove the facts are correct; validate the content before using it for a consequential action.

Official source

Structured model outputs

Check current provider documentation and model support before deployment. This guide is separate from the approved evidence records listed in the catalog.