← Back to community benchmarks

Llama-3.2-3B-Instruct

M4 Pro (16c) · 24 GB · 8bit · 2026-03-10

Performance

1k

tokens

818.6

PP tok/s

60.0

TG tok/s

1252

TTFT (ms)

3.9

Peak mem (GB)

Hardware

Chip M4 Pro (16c)

Memory 24 GB

GPU Cores 16

Software

oMLX v0.2.6

macOS macOS 26.4

Context 1,024

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	818.6	60.0	3.9 GB	current
4k	795.9	52.0	4.3 GB	view

Batching Results

Batch Size	TG tok/s	Speedup
1×	60.0	1.00×
2×	113.7	1.90×
4×	130.9	2.18×