← Back to community benchmarks

Qwen3.5-35B-A3B-Heretic

M3 Ultra (80c) · 512 GB · 4bit · 2026-03-15
Performance
1k
tokens
2,637
PP tok/s
96.9
TG tok/s
388
TTFT (ms)
20.1
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.2.12
macOS macOS 26.3.1
Context 1,024
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 2,637 96.9 20.1 GB current
4k 2,683 93.5 20.2 GB view
8k 2,670 89.3 20.4 GB view
16k 2,465 81.7 20.8 GB view
32k 2,127 67.8 21.6 GB view
64k 1,625 54.0 23.3 GB view
128k 1,059 38.1 26.7 GB view
195k 756.5 29.3 30.3 GB view
Batching Results
Batch Size TG tok/s Speedup
96.9 1.00×
169.7 1.75×
280.8 2.90×
426.7 4.40×