Flash-Next on a Mac Studio M3 Ultra 256GB: 113 tok/s (mlx-serve)
Qwen3.8-Flash-Next on mlx-serve, this Mac Studio, nothing else loaded.
Think off: 113 tok/s decode
Think xhigh: 70 tok/s decode
64k prompt: 960 tok/s prefill
258k prompt: 892 tok/s prefill
That is