← Back to community benchmarks

Qwen3.5-397B-A17B4.1bit

M3 Ultra (80c) · 512 GB · 6bit · 2026-03-09
Performance
8k
tokens
506.0
PP tok/s
36.2
TG tok/s
16191
TTFT (ms)
192.2
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.2.6
macOS macOS 26.2
Context 8,192
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 436.7 40.1 191.6 GB view
4k 499.8 38.2 191.8 GB view
8k 506.0 36.2 192.2 GB current
16k 489.5 33.3 192.9 GB view
32k 448.3 28.2 194.4 GB view
64k 362.1 22.2 197.4 GB view
Batching Results
Batch Size TG tok/s Speedup
40.1 1.00×
59.7 1.49×
71.6 1.79×
89.6 2.23×