← Back to community benchmarks

Qwen3-8B

M4 Max (40c) · 48 GB · 4bit · 2026-04-17

Performance

16k

tokens

661.9

PP tok/s

52.1

TG tok/s

24752

TTFT (ms)

7.1

Peak mem (GB)

Hardware

Chip M4 Max (40c)

Memory 48 GB

GPU Cores 40

Software

oMLX v0.3.5

macOS macOS 26.4.1

Context 16,384

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	829.8	88.0	5.0 GB	view
4k	853.0	79.0	5.5 GB	view
8k	782.6	67.5	6.0 GB	view
16k	661.9	52.1	7.1 GB	current
32k	527.2	37.1	9.4 GB	view
64k	366.2	21.3	14.1 GB	view