There’s a big price difference, and in my understanding if it’s just text, they are the same?
Curious on how do you make the decision
2 Likes
Hi @glaksmono !
So the mini model is essentially a modernization of the gpt-3.5 turbo. In general what I found it useful for is summarization tasks, simple entity extraction tasks, providing structured outputs that are as flat and shallow as possible, and operating on images that require simple descriptions, or coarse segmentation.
On tasks the require generating more elaborate summaries and texts, or more reasoning, or more complex entity extraction and structured output, then the more expensive models such as gpt-4o and o1 are the way to go.
One approach is to just start with the mini by default, and evaluate its performance against your business context, if good then keep it, if not good enough, up it to gpt-4o and re-evaluate.
1 Like
So in practice, you always have both instances, and use each respectively depending on what u wanna do?
Either as you say, route to mini or the other depending on the problem (there are even “routers” out there that can classify your problem and send it to one or the other). Or, you start with only mini, and if it doesn’t meet your needs for certain cases, upgrade to the larger one, or apply this routing concept