← Back to community benchmarks

Qwen3.5-397B-A17B9bit

M3 Ultra (80c) · 512 GB · 8bit · 2026-03-15
Performance
8k
tokens
490.1
PP tok/s
23.7
TG tok/s
16715
TTFT (ms)
417.8
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.2.12
macOS macOS 26.3.1
Context 8,192
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 424.3 26.4 417.1 GB view
4k 488.0 25.7 417.4 GB view
8k 490.1 23.7 417.8 GB current
Batching Results
Batch Size TG tok/s Speedup
26.4 1.00×
42.0 1.59×
55.6 2.11×