The new GPT-Instruct model was just released. It looks like it might be quite good at certain tasks, example here is solid Chess playing.
https://twitter.com/GrantSlatton/status/1703913578036904431
I'm getting mixed results from folks testing, will be interesting to see how it does on various tasks. It looks like it was released with the old completion API and not the Chat/Completion API. It makes testing it slightly harder given that API is legacy.