replied toย recently:
i think the best you can get right now is qwen 3.8 27B. youre gonna need 24 gigs of some fast memory though, i heard theres decent quantizations that can also make it run on 16. im quite excited about open/local models, but theres big investment overhead with the prices right now