Text Models

Structured Outputs

Pass response_format={"type": "json_object"} to force valid JSON output. Describe the shape you want in a system message — the model returns a JSON string you can parse directly. For strict schema enforcement (typed fields, required keys), use Function Calling instead.

structured.py

response = client.chat.completions.create(
    model="<model-id>",
    messages=[
        {#60a5fa]">class="text-emerald-400">"role": class="text-emerald-400">"system", class="text-emerald-400">"content": class="text-emerald-400">"Extract invoice data as JSON."},
        {#60a5fa]">class="text-emerald-400">"role": class="text-emerald-400">"user", class="text-emerald-400">"content": class="text-emerald-400">"Invoice #A-2042, total AED 1,250, due 2026-05-01."},
    ],
    response_format={#60a5fa]">class="text-emerald-400">"type": class="text-emerald-400">"json_object"},
)

# response.choices[0].message.content is a JSON string
#60a5fa]">import json
data = json.loads(response.choices[0].message.content)

JSON mode support varies by model. If the model doesn't support it, the upstream provider returns an error and you get a 400 back. Check the model's documentation before relying on JSON mode in production.