Ensuring JSON-Only Responses from Perplexity's Sonar-Pro Model

[ManthanCitrusbug]

I have an answer sheet that I read and pass as input to Perplexity’s Sonar-Pro model via OpenAI’s SDK. The model should:

1. Extract all questions from the answer sheet.
2. Detect the language of the text.
3. Return a properly formatted JSON response in the following structure:

{
    "language": "<detected_language>",
    "questions": [
        "question-1",
        "question-2",
        "question-3",
        "question-4"
    ]
}

I am using Perplexity’s base URL in OpenAI’s SDK and need to ensure the response strictly follows this JSON format.

How can I configure my request (parameters, settings, or prompt instructions) to achieve this?

[alexkbeck]

See the docs: https://docs.perplexity.ai/api-reference/chat-completions#body-response-format

You need a higher API tier to access this parameter. For now I am chaining API calls to have a second, smaller, model format the first model's response in to JSON.