Structured outputs

Force the model to return JSON with response_format — ideal for extraction, classification and anything you parse programmatically.

ℹ Note

response_format is forwarded to OpenAI-compatible providers; AIx auto-routes requests that use it. JSON-schema strictness depends on the underlying model — for guaranteed schema adherence pick a model that supports json_schema.

JSON object mode

from openai import OpenAI
client = OpenAI(base_url="https://api.aix.theaimart.co/v1", api_key="$AIX_KEY")

resp = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V3.2",
    messages=[
        {"role": "system", "content": "Extract the fields as JSON."},
        {"role": "user", "content": "Invoice #421 from Acme for $1,299 due 2026-07-01"},
    ],
    response_format={"type": "json_object"},
)
import json
data = json.loads(resp.choices[0].message.content)

JSON schema (when supported)

schema = {
    "type": "object",
    "properties": {
        "invoice_no": {"type": "string"},
        "vendor": {"type": "string"},
        "amount_usd": {"type": "number"},
        "due_date": {"type": "string"},
    },
    "required": ["invoice_no", "vendor", "amount_usd"],
    "additionalProperties": False,
}

resp = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V3.2",
    messages=[{"role": "user", "content": "Invoice #421 from Acme for $1,299 due 2026-07-01"}],
    response_format={"type": "json_schema", "json_schema": {"name": "invoice", "schema": schema}},
)

✦ Tip

Always wrap json.loads in a try/except and retry once on failure — even in JSON mode a model can occasionally emit invalid JSON. Ask for the schema in the system prompt too.

← Tool & function calling

Vision →