← Back to community benchmarks

Qwen3.5-27B

M3 Pro (18c) · 36 GB · 4bit · 2026-03-09
Performance
16k
tokens
94.3
PP tok/s
7.3
TG tok/s
173756
TTFT (ms)
18.3
Peak mem (GB)
Hardware
Chip M3 Pro (18c)
Memory 36 GB
GPU Cores 18
Software
oMLX v0.2.6
macOS macOS 26.3
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 97.4 8.6 16.6 GB view
4k 98.0 8.1 17.0 GB view
8k 97.2 8.0 17.4 GB view
16k 94.3 7.3 18.3 GB current
Batching Results
Batch Size TG tok/s Speedup
8.6 1.00×
17.0 1.98×
25.5 2.97×
25.8 3.00×