← Back to community benchmarks

Qwen2.5-3B-Instruct

M4 (10c) · 24 GB · 4bit · 2026-03-10

Performance

4k

tokens

478.7

PP tok/s

39.3

TG tok/s

8557

TTFT (ms)

2.4

Peak mem (GB)

Hardware

Chip M4 (10c)

Memory 24 GB

GPU Cores 10

Software

oMLX v0.2.7

macOS macOS 26.3.1

Context 4,096

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	496.9	45.4	2.2 GB	view
4k	478.7	39.3	2.4 GB	current

Batching Results

Batch Size	TG tok/s	Speedup
1×	45.4	1.00×
2×	88.6	1.95×
4×	114.7	2.53×