← Back to community benchmarks

Qwen3.6-27B-mxfp8

M3 Ultra (80c) · 512 GB · 8bit · 2026-04-24
Performance
32k
tokens
378.8
PP tok/s
18.1
TG tok/s
86496
TTFT (ms)
33.4
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.3.7
macOS macOS 26.4.1
Context 32,768
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 379.0 22.2 27.6 GB view
4k 414.5 21.7 29.0 GB view
8k 413.6 21.0 29.6 GB view
16k 402.7 19.9 30.9 GB view
32k 378.8 18.1 33.4 GB current
64k 330.4 15.4 38.4 GB view
128k 254.7 11.7 48.8 GB view
195k 204.2 9.2 59.5 GB view
Batching Results
Batch Size TG tok/s Speedup
22.2 1.00×
37.6 1.69×
43.8 1.97×