← Back to community benchmarks

Qwen3.5-27B

M5 Max (40c) · 128 GB · 8bit · 2026-03-21
Performance
64k
tokens
569.2
PP tok/s
11.0
TG tok/s
115130
TTFT (ms)
40.3
Peak mem (GB)
Hardware
Chip M5 Max (40c)
Memory 128 GB
GPU Cores 40
Software
oMLX v0.2.19
macOS macOS 26.3.1
Context 65,536
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 809.6 18.1 29.2 GB view
4k 855.2 17.8 30.8 GB view
8k 784.5 17.3 31.4 GB view
16k 725.0 13.0 32.7 GB view
32k 673.9 15.1 35.2 GB view
64k 569.2 11.0 40.3 GB current
Batching Results
Batch Size TG tok/s Speedup
18.1 1.00×
34.4 1.90×
48.9 2.70×
57.8 3.19×