Which size model are you using? Large isn't terribly good, but Next is alright. It's not close at all to GPT-4, but I can see some use cases I'd try it for (and will be).