Quickstart: get structured output

In this quickstart, you describe the fields you want with a Pydantic class, and the model fills them from a customer message. You get a validated Python object instead of free text. It takes about 5 minutes. For JSON mode and JSON schemas, see Structured outputs.

Before you begin

A Mistral API key. To create one, see Create your first API key.
Python 3.10 or later and uv.

Set up your environment

Create a project, install the Mistral SDK, and export your API key:

uv init structured-output --python 3.12 && cd structured-output
uv add mistralai
export MISTRAL_API_KEY="your-api-key"

uv init structured-output --python 3.12 && cd structured-output
uv add mistralai
export MISTRAL_API_KEY="your-api-key"

1. Write the script

Create structured.py. chat.parse sends the schema of Ticket with the request, then parses and validates the answer:

import os

from mistralai.client import Mistral
from pydantic import BaseModel


class Ticket(BaseModel):
    title: str
    priority: int
    customer: str


client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

response = client.chat.parse(
    model="mistral-small-latest",
    messages=[
        {
            "role": "user",
            "content": "Ada from Acme writes: our checkout page is down, please fix it urgently. "
            "Extract a support ticket with a priority from 1 (urgent) to 3 (low).",
        }
    ],
    response_format=Ticket,
)

ticket = response.choices[0].message.parsed
print(ticket)
print(f"Priority: {ticket.priority}")

import os

from mistralai.client import Mistral
from pydantic import BaseModel


class Ticket(BaseModel):
    title: str
    priority: int
    customer: str


client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

response = client.chat.parse(
    model="mistral-small-latest",
    messages=[
        {
            "role": "user",
            "content": "Ada from Acme writes: our checkout page is down, please fix it urgently. "
            "Extract a support ticket with a priority from 1 (urgent) to 3 (low).",
        }
    ],
    response_format=Ticket,
)

ticket = response.choices[0].message.parsed
print(ticket)
print(f"Priority: {ticket.priority}")

2. Run it

uv run structured.py

uv run structured.py

You should see

A Ticket object with the three fields filled in, then one field read as a Python integer:

title='Checkout page is down' priority=1 customer='Acme'
Priority: 1

title='Checkout page is down' priority=1 customer='Acme'
Priority: 1

The values can vary slightly between runs, but they always match the types of Ticket. If the answer doesn't validate, parse raises a validation error instead of returning a wrong object.

Going further

Structured outputs

Quickstart: call your own functions

Quickstart: run a batch job