← Back to community benchmarks

Qwen3.5-397B-A17B

M3 Ultra (80c) · 512 GB · 4bit · 2026-03-12

Performance

8k

tokens

517.2

PP tok/s

31.7

TG tok/s

15838

TTFT (ms)

211.0

Peak mem (GB)

Hardware

Chip M3 Ultra (80c)

Memory 512 GB

GPU Cores 80

Software

oMLX v0.2.9

macOS macOS 26.3

Context 8,192

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	448.8	35.1	210.4 GB	view
4k	513.2	33.8	210.7 GB	view
8k	517.2	31.7	211.0 GB	current
16k	495.3	28.9	211.8 GB	view

Batching Results

Batch Size	TG tok/s	Speedup
1×	35.1	1.00×
2×	55.8	1.59×
4×	81.4	2.32×