← Back to community benchmarks

Qwen3.5-397B-A17B-4bitREAP-CRACK

M3 Ultra (80c) · 512 GB · 4bit · 2026-04-17

Performance

16k

tokens

444.9

PP tok/s

24.4

TG tok/s

36824

TTFT (ms)

236.5

Peak mem (GB)

Hardware

Chip M3 Ultra (80c)

Memory 512 GB

GPU Cores 80

Software

oMLX v0.3.6

macOS macOS 26.5

Context 16,384

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	120.0	0.2	233.5 GB	view
4k	42.5	27.3	234.7 GB	view
16k	444.9	24.4	236.5 GB	current