← Back to community benchmarks

Qwen3.5-27B4bit-MTP

M5 Max (40c) · 128 GB · 4bit · 2026-04-02
Performance
16k
tokens
722.1
PP tok/s
26.6
TG tok/s
22690
TTFT (ms)
19.3
Peak mem (GB)
Hardware
Chip M5 Max (40c)
Memory 128 GB
GPU Cores 40
Software
oMLX v0.3.0
macOS macOS 26.4
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 756.3 30.6 15.8 GB view
4k 894.8 28.4 17.4 GB view
8k 825.4 27.2 18.0 GB view
16k 722.1 26.6 19.3 GB current
Batching Results
Batch Size TG tok/s Speedup
30.6 1.00×
27.5 0.90×
40.3 1.32×
47.0 1.54×