3.8 is a huge step up from 3.5, quantized or not. I do all of my programming on Qwen 3.8 27B Q4 these days.