← Back to community benchmarks

Qwen3.5-27B-heretic-oQ4

M5 Max (40c) · 128 GB · 4bit · 2026-04-15
Performance
8k
tokens
864.3
PP tok/s
30.3
TG tok/s
9478
TTFT (ms)
18.4
Peak mem (GB)
Hardware
Chip M5 Max (40c)
Memory 128 GB
GPU Cores 40
Software
oMLX v0.3.5.dev1
macOS macOS 26.4.1
Context 8,192
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 769.0 32.2 16.3 GB view
4k 891.1 31.2 17.7 GB view
8k 864.3 30.3 18.4 GB current
16k 761.0 28.3 19.6 GB view
32k 682.3 25.4 22.1 GB view
64k 543.5 18.7 27.1 GB view
Batching Results
Batch Size TG tok/s Speedup
32.2 1.00×
45.7 1.42×
56.2 1.75×