← Back to community benchmarks

Qwen3.5-27B

M1 Ultra (48c) · 64 GB · 4bit · 2026-03-09
Performance
16k
tokens
130.9
PP tok/s
16.6
TG tok/s
125167
TTFT (ms)
18.3
Peak mem (GB)
Hardware
Chip M1 Ultra (48c)
Memory 64 GB
GPU Cores 48
Software
oMLX v0.2.6
macOS macOS 26.3.1
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 133.1 24.3 16.6 GB view
4k 134.2 22.7 17.0 GB view
8k 133.5 20.5 17.4 GB view
16k 130.9 16.6 18.3 GB current
32k 125.8 12.5 20.1 GB view
Batching Results
Batch Size TG tok/s Speedup
24.3 1.00×
29.6 1.22×
32.7 1.35×