Qwen Flash Next Runs Locally on 64GB Mac mini M5

A Reddit user ran Qwen Flash Next Q4 locally on a 64GB Mac mini M5, achieving about 17.5 tks decoding and 360 tks prompt processing, with hot MoE experts cached and the rest streamed from SSD—though decoding still waits on disk about 27% of the time.

2026-10-05 ~ 2026-10-05 · 2 related posts