I think (maybe I missed something) that identical size and quant versions of Qwen 3.5 and 3.8 should run at the same speed. It’s the exact same architecture.