Smaller models like Curie can work quite well too. But they are less "instruct-like" models so you will need to properly use few-shot learning (aka "prompt engi

Smaller models like Curie can work quite well too. But they are less "instruct-like" models so you will need to properly use few-shot learning (aka "prompt engineering") in order to get good results: https://nlpcloud.com/effectively-using-gpt-j-gpt-neo-gpt-3-alternatives-few-shot-learning.html https://nlpcloud.com/effectively-using-gpt-j-gpt-neo-gpt-3-a... I takes a bit more work though, and it makes your requests bigger so more expensive.