pull down to refresh

I get 25TPS. Codex/Claude is significantly better, but I was happily surprised how well this works without relying on one of these centralised companies. Now I’m very much looking forward to the next generation of local models