← Back to community benchmarks

Qwen3.5-397B-A17B-8bit-gs32

M3 Ultra (80c) · 512 GB · 8bit · 2026-03-11
Performance
16k
tokens
473.2
PP tok/s
22.3
TG tok/s
34621
TTFT (ms)
418.5
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.2.6
macOS macOS 26.2
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 416.8 26.1 417.1 GB view
4k 487.0 25.5 417.4 GB view
8k 489.6 24.4 417.8 GB view
16k 473.2 22.3 418.5 GB current
32k 432.9 19.2 420.0 GB view