Getting Structured JSON Output from an LLM

Step 1: Define the shape of the data

Describe what you want back with a Pydantic model. Each field has a name and a type, and the model doubles as validation and documentation.

from pydantic import BaseModel
from typing import List

class ContactInfo(BaseModel):
    name: str
    email: str
    plan: str
    interests: List[str]
    demo_requested: bool

Step 2: Parse the response into your model

The SDK's parse helper takes your model and returns a validated instance, so you work with real Python objects instead of text.

import anthropic

client = anthropic.Anthropic()

response = client.messages.parse(
    model="claude-opus-5-5",
    max_tokens=16000,
    messages=[{
        "role": "user",
        "content": "Extract: Jane Doe (jane@co.com) wants Enterprise, interested in API and SDKs, wants a demo.",
    }],
    output_format=ContactInfo,
)

contact = response.parsed_output
print(contact.name, contact.interests)

Step 3: Or supply a raw JSON schema

If you are not using Pydantic, pass a JSON schema in the output configuration. The response is then guaranteed to be valid JSON matching it, which you load with the json module. Set additionalProperties to false and list required fields.

import json

response = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=16000,
    messages=[{"role": "user", "content": "Extract info: John Smith (john@example.com) wants the Enterprise plan."}],
    output_config={
        "format": {
            "type": "json_schema",
            "schema": {
                "type": "object",
                "properties": {
                    "name": {"type": "string"},
                    "email": {"type": "string"},
                    "plan": {"type": "string"},
                },
                "required": ["name", "email", "plan"],
                "additionalProperties": False,
            },
        }
    },
)

text = next(b.text for b in response.content if b.type == "text")
data = json.loads(text)

Step 4: Still validate the meaning

A schema guarantees the shape, not the truth. A valid email field can still hold the wrong email. Add business checks in your code, such as format rules and cross-checks with other data, and decide what to do with records that fail.

Step 5: Handle failures and keep the schema simple

Check the stop reason before trusting a response, since a request can end early or be declined. Keep schemas small and fields well named, use enums for fixed choices, and make fields optional only when the data may legitimately be absent.