← Back to community benchmarks

Qwen3-4B-Instruct-2507

M5 Pro (16c) · 48 GB · 4bit · 2026-04-17

Performance

1k

tokens

1,992

PP tok/s

95.3

TG tok/s

514

TTFT (ms)

2.9

Peak mem (GB)

Hardware

Chip M5 Pro (16c)

Memory 48 GB

GPU Cores 16

Software

oMLX v0.3.6

macOS macOS 26.4.1

Context 1,024

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	1,992	95.3	2.9 GB	current
4k	2,235	76.4	3.4 GB	view

Batching Results

Batch Size	TG tok/s	Speedup
1×	95.3	1.00×
2×	152.8	1.60×
4×	186.5	1.96×