← Back to community benchmarks

Qwen3.5-27B-Heretic

M1 Ultra (64c) · 128 GB · bf16 · 2026-03-13
Performance
16k
tokens
199.4
PP tok/s
9.0
TG tok/s
82172
TTFT (ms)
54.2
Peak mem (GB)
Hardware
Chip M1 Ultra (64c)
Memory 128 GB
GPU Cores 64
Software
oMLX v0.2.10
macOS macOS 26.4
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 201.6 10.4 52.5 GB view
4k 207.0 10.1 52.8 GB view
8k 205.0 9.7 53.3 GB view
16k 199.4 9.0 54.2 GB current
Batching Results
Batch Size TG tok/s Speedup
10.4 1.00×
6.9 0.66×
13.5 1.30×