I’ve played with some models now that I’m a little more impressed with. Nous-Hermes is very nice—still a little worse than GPT 3.5, but that’s not bad at all IMO.