← Back to community benchmarks

Qwen3.5-397B-A17B-4bit_副本

M3 Ultra (80c) · 512 GB · 4bit · 2026-03-15

Performance

8k

tokens

510.1

PP tok/s

34.7

TG tok/s

16060

TTFT (ms)

210.2

Peak mem (GB)

Hardware

Chip M3 Ultra (80c)

Memory 512 GB

GPU Cores 80

Software

oMLX v0.2.12

macOS macOS 26.4

Context 8,192

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	447.9	37.6	209.6 GB	view
4k	510.8	36.3	209.9 GB	view
8k	510.1	34.7	210.2 GB	current
16k	490.6	22.4	211.0 GB	view