← Back to community benchmarks

Qwen3.5-4B9bit

M3 Ultra (60c) · 96 GB · 8bit · 2026-03-09
Performance
32k
tokens
1,585
PP tok/s
70.9
TG tok/s
20669
TTFT (ms)
7.5
Peak mem (GB)
Hardware
Chip M3 Ultra (60c)
Memory 96 GB
GPU Cores 60
Software
oMLX v0.2.6
macOS macOS 26.3.1
Context 32,768
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,354 107.3 5.6 GB view
4k 1,762 101.7 5.7 GB view
8k 1,810 97.4 6.1 GB view
16k 1,752 87.2 6.9 GB view
32k 1,585 70.9 7.5 GB current
Batching Results
Batch Size TG tok/s Speedup
107.3 1.00×
178.7 1.67×
238.4 2.22×