I tried some things in the web demo and it is really good.

What people haven't realised yet is that Koala (another model they did not publish about for now) is also available in the web demo and it is CRAZY GOOD! It's also really fast because I guess I'm the only one using it right now haha.

I really recommand to try it, it looks like Vicuna is a bit bellow GPT3.5 and Koala a bit above but I did not test it enough to be sure right now.