← Back to community benchmarks

Qwen3.6-27B-oQ8-mtp

M3 Ultra (80c) · 512 GB · 8bit · 2026-05-13
Performance
1k
tokens
386.9
PP tok/s
29.0
TG tok/s
2647
TTFT (ms)
28.8
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.3.9.dev2
macOS macOS 26.4.1
Context 1,024
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 386.9 29.0 28.8 GB current
4k 420.3 31.9 30.3 GB view
8k 419.8 31.3 31.3 GB view
16k 409.1 28.9 32.8 GB view
32k 383.8 27.4 35.8 GB view
64k 333.6 23.9 41.8 GB view
128k 252.9 4.8 183.6 GB view
195k 200.2 8.5 213.4 GB view
Batching Results
Batch Size TG tok/s Speedup
29.0 1.00×
14.8 0.51×
37.9 1.31×
55.1 1.90×