Video of tensorfold running Qwen3.8-Flash-Next-MLX-oQ8-MTP high on M3 Ultra ~80 t/s! It's a great experience, less errors and better quality. Where possible I suggest to use the highest possible quants!