← Back to community benchmarks

Qwen3.5-4B

M3 Ultra (80c) · 256 GB · 4bit · 2026-03-10
Performance
8k
tokens
2,541
PP tok/s
137.2
TG tok/s
3224
TTFT (ms)
4.6
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 256 GB
GPU Cores 80
Software
oMLX v0.2.7
macOS macOS 26.2
Context 8,192
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 2,579 161.4 4.0 GB view
4k 2,614 151.0 4.2 GB view
8k 2,541 137.2 4.6 GB current
16k 2,385 118.5 5.3 GB view
32k 2,111 92.0 6.9 GB view
64k 1,706 64.4 8.6 GB view
Batching Results
Batch Size TG tok/s Speedup
161.4 1.00×
276.7 1.71×
382.6 2.37×
514.6 3.19×