I've gone down this rabbit hole and I want to reaffirm what the other commenters are saying: even if you use a massive model and have the compute to back it up

I've gone down this rabbit hole and I want to reaffirm what the other commenters are saying: even if you use a massive model and have the compute to back it up at a reasonable pace (you likely don't), it sucks, can't even hold a candle to GPT 3.5