Docs/Structured output

Structured output

Structured output makes the model return data in the JSON format you specify, suitable for scenarios such as data extraction, classification labeling, and form filling.

JSON Mode#

The simplest form of structured output, forcing the model to return valid JSON:

Python

Code

from openai import OpenAI
 
client = OpenAI(
    base_url="https://as.apinoma.com/v1",
    api_key="<YOUR APINOMA_API_KEY>"
)
 
response = client.chat.completions.create(
    model="openai/gpt-4o",
    messages=[
        {"role": "system", "content": "You are a data extraction assistant. Return the result in JSON format."},
        {"role": "user", "content": "Extract the person name, company, and title from the following text: John Smith is a senior engineer at Alibaba"}
    ],
    response_format={"type": "json_object"}
)
 
import json
result = json.loads(response.choices[0].message.content)
print(result)
# {"name": "John Smith", "company": "Alibaba", "title": "Senior Engineer"}

TypeScript

When using JSON Mode, the system prompt must contain the keyword “JSON”, otherwise some models may ignore the format requirement.

JSON Schema constraints#

Control the output structure more precisely, ensuring field names and types match expectations:

Code

response = client.chat.completions.create(
    model="openai/gpt-4o",
    messages=[
        {"role": "user", "content": "Analyze the sentiment of this user review: This product is amazing and really easy to use!"}
    ],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "sentiment_analysis",
            "schema": {
                "type": "object",
                "properties": {
                    "sentiment": {
                        "type": "string",
                        "enum": ["positive", "negative", "neutral"],
                        "description": "Sentiment polarity"
                    },
                    "confidence": {
                        "type": "number",
                        "description": "Confidence, 0-1"
                    },
                    "keywords": {
                        "type": "array",
                        "items": {"type": "string"},
                        "description": "Key sentiment words"
                    }
                },
                "required": ["sentiment", "confidence", "keywords"],
                "additionalProperties": False
            }
        }
    }
)

Output:

Real-world use cases#

Data extraction

Code

# Extract structured data from unstructured text
response = client.chat.completions.create(
    model="openai/gpt-4o",
    messages=[{
        "role": "user",
        "content": """Extract the following order information:
        Customer Jane Doe placed an order on January 15, 2025 for 3 MacBook Pros,
        at a unit price of 18999 CNY, with the shipping address 123 XXX Road, Chaoyang District, Beijing"""
    }],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "order_info",
            "schema": {
                "type": "object",
                "properties": {
                    "customer": {"type": "string"},
                    "date": {"type": "string"},
                    "items": {
                        "type": "array",
                        "items": {
                            "type": "object",
                            "properties": {
                                "name": {"type": "string"},
                                "quantity": {"type": "integer"},
                                "unit_price": {"type": "number"}
                            }
                        }
                    },
                    "address": {"type": "string"}
                },
                "required": ["customer", "date", "items", "address"]
            }
        }
    }
)

Classification labeling

Supported models#

Model

JSON Mode

JSON Schema

`openai/gpt-4o`

✅

`openai/gpt-4o-mini`

✅

`anthropic/claude-sonnet-4.6`

✅

—

`google/gemini-pro-compatible-flash-lite-preview`

✅

Models that do not support JSON Schema can achieve a similar effect by describing the expected JSON format in detail in the system prompt.

Last updated on June 23, 2026