Structured outputs

Constrain a response to a JSON schema.

Ask for JSON that matches a schema with response_format and type: "json_schema". The simpler { "type": "json_object" } asks for a JSON object without a schema and is forwarded as sent. Either works only on models that list response_format or structured_outputs in supported_parameters.

Request a schema

curl https://deference.si/v1/chat/completions \
  -H "Authorization: Bearer $DEFERENCE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-5.5",
    "messages": [{ "role": "user", "content": "Extract: Ada paid $12.40 on Oct 8." }],
    "response_format": {
      "type": "json_schema",
      "json_schema": {
        "name": "payment",
        "strict": true,
        "schema": {
          "type": "object",
          "properties": {
            "name": { "type": "string" },
            "amount_usd": { "type": "number" },
            "date": { "type": "string" }
          },
          "required": ["name", "amount_usd", "date"],
          "additionalProperties": false
        }
      }
    }
  }'

import json, os
from openai import OpenAI

client = OpenAI(base_url="https://deference.si/v1", api_key=os.environ["DEFERENCE_API_KEY"])

response = client.chat.completions.create(
    model="anthropic/claude-sonnet-5.5",
    messages=[{"role": "user", "content": "Extract: Ada paid $12.40 on Oct 8."}],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "payment",
            "strict": True,
            "schema": {
                "type": "object",
                "properties": {
                    "name": {"type": "string"},
                    "amount_usd": {"type": "number"},
                    "date": {"type": "string"},
                },
                "required": ["name", "amount_usd", "date"],
                "additionalProperties": False,
            },
        },
    },
)
print(json.loads(response.choices[0].message.content))

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://deference.si/v1",
  apiKey: process.env.DEFERENCE_API_KEY,
});

const response = await client.chat.completions.create({
  model: "anthropic/claude-sonnet-5.5",
  messages: [{ role: "user", content: "Extract: Ada paid $12.40 on Oct 8." }],
  response_format: {
    type: "json_schema",
    json_schema: {
      name: "payment",
      strict: true,
      schema: {
        type: "object",
        properties: {
          name: { type: "string" },
          amount_usd: { type: "number" },
          date: { type: "string" },
        },
        required: ["name", "amount_usd", "date"],
        additionalProperties: false,
      },
    },
  },
});
console.log(JSON.parse(response.choices[0].message.content ?? "{}"));

Check the result

The message content is a JSON string. Parse it, and validate it on your side. A response that stops early with finish_reason: "length" can be cut-off JSON, so raise max_tokens if parsing fails.

Models without support

A model that does not support response_format answers with the provider's error, usually HTTP 400, and an invalid schema fails the same way. Check supported_parameters for response_format or structured_outputs before sending the field — provider is rejected on this API, so a request cannot force a supporting endpoint. If no model qualifies, put the schema in the prompt and validate the result.