← Back to community benchmarks

Qwen3.5-397B-A17B

M3 Ultra (80c) · 512 GB · 8bit · 2026-03-10
Performance
1k
tokens
433.6
PP tok/s
26.3
TG tok/s
2362
TTFT (ms)
394.9
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.2.6
macOS macOS 26.3.1
Context 1,024
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 433.6 26.3 394.9 GB current
4k 493.6 25.6 395.2 GB view
8k 498.9 24.1 395.6 GB view
16k 479.0 21.0 396.3 GB view
32k 436.2 16.5 397.8 GB view
64k 342.9 11.8 400.8 GB view
Batching Results
Batch Size TG tok/s Speedup
26.3 1.00×
43.6 1.66×
54.0 2.05×