Here is it, Opus at home on a $400 GPU.
Qwen3.8 27b (4 bit) running at 41 tok/s decode and 413 tok/s prefill on a 9060xt 16gb
A single 16gb GPU fully offloading qwen3.8 with near zero quality loss, tbh I didn't think we would get to this point so fast, but here we are.