← Back to community benchmarks

Ornstein-Hermes-3.6-27B-SABER-8bit-MTPLX-Optimized-Speed

M4 Max (40c) · 128 GB · 8bit · 2026-05-14
Performance
1k
tokens
235.4
PP tok/s
16.3
TG tok/s
4350
TTFT (ms)
28.3
Peak mem (GB)
Hardware
Chip M4 Max (40c)
Memory 128 GB
GPU Cores 40
Software
oMLX v0.3.9.dev2
macOS macOS 26.5
Context 1,024
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 235.4 16.3 28.3 GB current
4k 244.2 16.1 29.8 GB view
Batching Results
Batch Size TG tok/s Speedup
16.3 1.00×
31.4 1.93×
45.2 2.77×