← Back to community benchmarks

Qwen3.5-122B-A10B-5bit

M5 Max (40c) · 128 GB · 5bit · 2026-04-05
Performance
16k
tokens
2,466
PP tok/s
17.3
TG tok/s
6643
TTFT (ms)
84.4
Peak mem (GB)
Hardware
Chip M5 Max (40c)
Memory 128 GB
GPU Cores 40
Software
oMLX v0.3.2
macOS macOS 26.4
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,008 58.6 80.8 GB view
4k 1,447 56.0 82.2 GB view
8k 1,469 53.7 82.8 GB view
16k 2,466 17.3 84.4 GB current
32k 3,098 17.1 85.8 GB view
64k 2,700 16.8 87.6 GB view
128k 2,099 16.2 91.1 GB view
Batching Results
Batch Size TG tok/s Speedup
58.6 1.00×
85.7 1.46×
102.5 1.75×
122.0 2.08×