← Back to community benchmarks

Qwen3-VL-8B-NSFW-Caption-V4.5-affine

M5 Pro (20c) · 64 GB · 4bit · 2026-07-21
Performance
32k
tokens
980.7
PP tok/s
28.2
TG tok/s
33412
TTFT (ms)
13.1
Peak mem (GB)
Hardware
Chip M5 Pro (20c)
Memory 64 GB
GPU Cores 20
Software
oMLX v0.5.1
macOS macOS 26.5.2
Context 32,768
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,770 58.5 6.2 GB view
4k 1,720 53.0 6.8 GB view
8k 1,559 46.9 8.2 GB view
16k 1,367 38.3 9.9 GB view
32k 980.7 28.2 13.1 GB current
64k 665.9 17.5 19.9 GB view
Batching Results
Batch Size TG tok/s Speedup
58.5 1.00×
104.3 1.78×
147.2 2.52×
283.8 4.85×