Getting JSON Object Response Back In OpenAI GPT4o

Getting back valid json object returned as a response from any LLM models api has always been a challenge.

No matter how many times you add explicitly add in your prompt , "Only return the response in a json object and nothing else" etc and call that api endpoint 10 times, it is mostly likely 3 out of them will fail to return a valid json object and there goes the JSON.parse error.

So it was a great news when when OpenAI shared that you can now get back a valid json object response from their api (atleast some models) endpoints.

I have been testing this with GPT4o and it works like a charm. Here is very simple example on how you can get back a valid json object response from OpenAI GPT4o model.

const completion = await openai.chat.completions.create({
      model: "gpt-4o",
      messages: [
        {
          role: "user",
          content: "Say this is a test , return it in a json object",
        },
      ],
      response_format: { type: "json_object" },
    });
    console.log(completion.choices[0]?.message?.content);   
  } catch (error) {
    console.error(error);   
  }

Response :

{
  "message": "This is a test."
}

Few things to note:

• Make sure you are using the latest version of OpenAI api client.

• Make sure you are using the correct model name.

• And most importantly, make sure you explicitly mention in your prompt that you want the response in a json object. Otherwise it will fail with following error :

  BadRequestError: 400 'messages' must contain the word 'json' in some form, to use 'response_format' of type 'json_object'.
  

So following prompt doesn't works and throws above error:

content: "Say this is a test",

Following prompt works & returns a valid json object:

content: "Say this is a test , return it in a json object",

Hope this helps!

Back To All Blogs