← Back to community benchmarks

Qwen3-VL-30B-A3B-Instruct

M5 Pro (20c) · 48 GB · 4bit · 2026-03-18

Performance

1k

tokens

1,673

PP tok/s

80.5

TG tok/s

612

TTFT (ms)

17.6

Peak mem (GB)

Hardware

Chip M5 Pro (20c)

Memory 48 GB

GPU Cores 20

Software

oMLX v0.2.18

macOS macOS 26.3.1

Context 1,024

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	1,673	80.5	17.6 GB	current
4k	2,370	69.0	18.1 GB	view
8k	2,121	59.4	18.4 GB	view

Batching Results

Batch Size	TG tok/s	Speedup
1×	80.5	1.00×
2×	106.6	1.32×
4×	37.8	0.47×