KC(@karanC_12):This should not be possible. Qwen3.8-27B just hit 70 tokens per second on a MacBook Pro M5 Max. Same quality. Up to 4.6× faster than normal decoding. A frontier-level open model running this fast on a laptop. Local AI just became actually usable. 👀

2026.08.19 09:55

This should not be possible. Qwen3.8-27B just hit 70 tokens per second on a MacBook Pro M5 Max. Same quality. Up to 4.6× faster than normal decoding. A frontier-level open model running this fast on a laptop. Local AI just became actually usable. 👀

Zhijian Liu@zhijianliu_

2026.08.18 22:08

DFlash 2 is here! Qwen3.8-27B at 70 tok/s on an M5 Max MacBook Pro. ⚡ Up to 4.6× the speed of autoregressive decoding, with the same output. This is the next generation of DFlash, seeded at Z Lab and upgraded at Inco AI. Get one more accepted token on every pass, for free!