← Back to community benchmarks

Qwen3.6-35B-A3B

M5 Max (40c) · 128 GB · 8bit · 2026-07-24
User commentmlx
Performance
16k
tokens
2,723
PP tok/s
92.3
TG tok/s
6018
TTFT (ms)
36.5
Peak mem (GB)
Hardware
Chip M5 Max (40c)
Memory 128 GB
GPU Cores 40
Software
oMLX v0.5.3
macOS macOS 26.5.2
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 3,489 100.1 35.5 GB view
4k 4,322 98.2 36.4 GB view
8k 4,077 95.9 36.7 GB view
16k 2,723 92.3 36.5 GB current
32k 1,888 85.7 36.8 GB view
64k 1,200 74.4 37.4 GB view
128k 726.6 60.7 38.7 GB view
195k 505.8 50.6 40.1 GB view
Batching Results
Batch Size TG tok/s Speedup
100.1 1.00×
208.6 2.08×
418.1 4.18×
839.4 8.39×