Which Model of OpenAI API do we need to use?

[spnc]

Hi it's not very clear, after we've turned on Retool AI with our OpenAI API, which model is most efficient?

By default: Retool puts gpt-4

But is this enough? So I try gpt-4-turbo

Obviously it's a lot more expensive than gpt-4 (looking at my cost usage on the OpenAI API page), like 4/5 times more expensive

So what is the standard practice here, what is everyone doing? Is gpt-4 enough?

Also there's quite a few more models:

I'm really not sure what's the best deal in terms of performance and pricing.

Surprisingly Retool is not providing any sort of guidance.

[abusedmedia]

Hi @spnc

Your question is not easy to answer.
It depends of what you need to do. And you need to test them.

There are two main variable that you need to consider:

• the context size, if you need to send more information, 16k or 32k, so it might guide your choice
• the accuracy of the model, it really depends of what you need to do

In general, try the cheaper one, than, switch to more expensive models if the quality is not enough.

Hope this help.