JSON Response Format

Force the model to respond with valid JSON by setting response_format: "json".

Modes

Mode	How to use
Simple object	Ask for JSON in the prompt, set response_format: "json"
Array / multiple items	Ask for a JSON array in the prompt
Custom schema	Include your schema in the system or user message

Simple JSON

Python SDKcURL

response = client.chat.completions.create(
    model="gemma-3-12b-it-qat",
    messages=[{"role": "user", "content": 'Return {"name": "Alice", "age": 30} as JSON'}],
    response_format="json",
)
import json
data = json.loads(response["choices"][0]["message"]["content"])

curl https://api.studiolm.dev/v1/chat/completions \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma-3-12b-it-qat",
    "messages": [{"role":"user","content":"Return {\"name\":\"Alice\",\"age\":30} as JSON"}],
    "response_format": "json"
  }'

Custom schema

Include the schema in your prompt and the model will conform to it:

response = client.chat.completions.create(
    model="gemma-3-12b-it-qat",
    messages=[{
        "role": "user",
        "content": (
            "List 3 programming languages. "
            'Respond as JSON matching this schema: '
            '{"languages": [{"name": string, "year": number}]}'
        ),
    }],
    response_format="json",
)

Streaming with JSON mode

Buffer the full response before parsing — partial JSON is not valid:

chunks = []
for chunk in client.chat.completions.create(
    model="gemma-3-12b-it-qat",
    messages=[{"role": "user", "content": "Give me a JSON object with a joke."}],
    stream=True,
    response_format="json",
):
    chunks.append(chunk["choices"][0].get("delta", {}).get("content", ""))

import json
data = json.loads("".join(chunks))

Tip

Always tell the model in your prompt what JSON shape you expect — the instruction in the prompt is what guides the structure.