← Back to community benchmarks

Qwen3.5-35B-A3B-Hereticmxfp8

M5 Pro (20c) · 48 GB · 8bit · 2026-03-29
Performance
16k
tokens
1,936
PP tok/s
42.0
TG tok/s
8465
TTFT (ms)
36.8
Peak mem (GB)
Hardware
Chip M5 Pro (20c)
Memory 48 GB
GPU Cores 20
Software
oMLX v0.2.21
macOS macOS 26.3.1
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,749 50.7 34.9 GB view
4k 2,143 48.1 35.9 GB view
8k 1,998 45.5 36.2 GB view
16k 1,936 42.0 36.8 GB current
Batching Results
Batch Size TG tok/s Speedup
50.7 1.00×
84.3 1.66×
114.6 2.26×