Structured Outputs - Kyma API

JSON mode

Force the model to return valid JSON by setting response_format:

from openai import OpenAI

client = OpenAI(
    base_url="https://kymaapi.com/v1",
    api_key="kyma-your-api-key"
)

response = client.chat.completions.create(
    model="qwen-3-32b",
    messages=[{"role": "user", "content": "List 3 programming languages with name and year. Return a JSON object with a `languages` array."}],
    response_format={"type": "json_object"}
)

import json
data = json.loads(response.choices[0].message.content)
print(data)
# {"languages": [{"name": "Python", "year": 1991}, {"name": "JavaScript", "year": 1995}, ...]}

JSON Schema (strict mode)

Define the exact shape of the response:

response = client.chat.completions.create(
    model="qwen-3.6-plus",
    messages=[{"role": "user", "content": "List 2 colors"}],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "colors",
            "schema": {
                "type": "object",
                "properties": {
                    "colors": {
                        "type": "array",
                        "items": {"type": "string"}
                    }
                },
                "required": ["colors"]
            }
        }
    }
)

data = json.loads(response.choices[0].message.content)
print(data)
# {"colors": ["red", "blue"]}

Most models follow the schema closely, but nothing on the way validates the answer against it, so keep parsing in a try and validate the result client-side (the Pydantic example below does both).

Which models support structured outputs?

Model	json_object	json_schema
Qwen 3 32B	✅	✅
Qwen 3 Coder	✅	✅
Qwen 3.6 Plus	✅	✅
Llama 3.3 70B	✅	✅
Kimi K2.6	✅	✅
GPT-OSS 120B	✅	✅
Gemini 3 Flash	✅	✅
Gemma 4 31B	✅	⚠️ Limited

If a model doesn’t support json_schema, use json_object mode and describe the format in your prompt. Check supports_structured_outputs on GET /v1/models first — a few models, including the sonar search models, return false and support neither mode.

With Pydantic (Python)

from pydantic import BaseModel
from openai import OpenAI

class MovieReview(BaseModel):
    title: str
    rating: float
    summary: str

client = OpenAI(base_url="https://kymaapi.com/v1", api_key="kyma-your-api-key")

response = client.chat.completions.create(
    model="qwen-3-32b",
    messages=[{"role": "user", "content": "Review the movie Inception"}],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "movie_review",
            "schema": MovieReview.model_json_schema()
        }
    }
)

review = MovieReview.model_validate_json(response.choices[0].message.content)
print(f"{review.title}: {review.rating}/10")

With TypeScript/Zod

import OpenAI from 'openai';
import { z } from 'zod';
import { zodToJsonSchema } from 'zod-to-json-schema';

const ColorSchema = z.object({
  colors: z.array(z.string())
});

const client = new OpenAI({
  baseURL: 'https://kymaapi.com/v1',
  apiKey: 'kyma-your-api-key'
});

const response = await client.chat.completions.create({
  model: 'qwen-3.6-plus',
  messages: [{ role: 'user', content: 'List 3 colors' }],
  response_format: {
    type: 'json_schema',
    json_schema: {
      name: 'colors',
      schema: zodToJsonSchema(ColorSchema)
    }
  }
});

const data = ColorSchema.parse(JSON.parse(response.choices[0].message.content));