Structured outputs

JSON that matches a schema

response_format of type json_schema asks the model for JSON that follows your schema. The router sends such a request only to routes that support structured output (the Structured output capability in the catalog); a model without one is a 400 "no route".

{
  "model": "openai/gpt-4o-mini",
  "messages": [{ "role": "user", "content": "Extract the city and the date: meet me in Kazan on 3 May." }],
  "response_format": {
    "type": "json_schema",
    "json_schema": {
      "name": "meeting",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": { "city": { "type": "string" }, "date": { "type": "string" } },
        "required": ["city", "date"],
        "additionalProperties": false
      }
    }
  }
}

The answer's message.content is a JSON string: parse it.

Any valid JSON

{"type": "json_object"} asks for valid JSON without a schema. The router does not check that the model supports it: say in the prompt that you want JSON, and handle a parsing error.

Tips

• Keep the schema small: it is billed as input tokens.
• Even with strict, check the result in your code; an answer cut by max_tokens (finish_reason: "length") is not valid JSON.