← Back to community benchmarks

Llama-3.2-3B-Instruct

M4 (10c) · 16 GB · 4bit · 2026-03-11

Performance

1k

tokens

478.1

PP tok/s

48.4

TG tok/s

2144

TTFT (ms)

2.3

Peak mem (GB)

Hardware

Chip M4 (10c)

Memory 16 GB

GPU Cores 10

Software

oMLX v0.2.8b1

macOS macOS 26.2

Context 1,024

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	478.1	48.4	2.3 GB	current
4k	433.9	36.9	2.6 GB	view

Batching Results

Batch Size	TG tok/s	Speedup
1×	48.4	1.00×
2×	91.7	1.89×
4×	102.9	2.13×