← Back to community benchmarks

Qwen3.5-35B-A3B-5bit

M1 Ultra (48c) · 64 GB · 5bit · 2026-03-11
Performance
64k
tokens
559.4
PP tok/s
20.6
TG tok/s
117155
TTFT (ms)
27.4
Peak mem (GB)
Hardware
Chip M1 Ultra (48c)
Memory 64 GB
GPU Cores 48
Software
oMLX v0.2.7
macOS macOS 26.3.1
Context 65,536
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 697.4 53.7 24.1 GB view
4k 808.4 51.4 24.2 GB view
8k 812.9 46.8 24.4 GB view
16k 768.4 40.6 24.9 GB view
32k 703.9 32.2 25.7 GB view
64k 559.4 20.6 27.4 GB current
Batching Results
Batch Size TG tok/s Speedup
53.7 1.00×
84.4 1.57×
124.1 2.31×
159.6 2.97×