Okay akutally Qwen 3.8 might be the first qwen model that I can actually work with. All new models have a bias but i've been using it for a few hours so far today. A little slow, but hopefully I can get that tweaked. I lose cloud tokens at the end of next week, so lets see if it can keep up :) 27b-mtp-q8_0 kv: 256k-q8_0 - 41gb in use.