← Back to community benchmarks

Llama-3.2-3B-Instruct

M3 Pro (14c) · 18 GB · 4bit · 2026-03-11

Performance

4k

tokens

619.2

PP tok/s

51.4

TG tok/s

6617

TTFT (ms)

2.8

Peak mem (GB)

Hardware

Chip M3 Pro (14c)

Memory 18 GB

GPU Cores 14

Software

oMLX v0.2.6

macOS macOS 26.3.1

Context 4,096

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	639.4	61.9	2.5 GB	view
4k	619.2	51.4	2.8 GB	current

Batching Results

Batch Size	TG tok/s	Speedup
1×	61.9	1.00×
2×	110.8	1.79×
4×	113.1	1.83×
8×	138.2	2.23×