← Back to community benchmarks

Llama-3.2-3B-Instruct

M4 (10c) · 32 GB · 4bit · 2026-03-11
Performance
64k
tokens
181.9
PP tok/s
10.1
TG tok/s
360312
TTFT (ms)
9.6
Peak mem (GB)
Hardware
Chip M4 (10c)
Memory 32 GB
GPU Cores 10
Software
oMLX v0.2.7
macOS macOS 26.3.1
Context 65,536
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 558.9 49.2 2.3 GB view
4k 520.3 41.1 2.6 GB view
8k 470.4 33.9 3.0 GB view
16k 391.2 25.2 3.9 GB view
32k 276.9 16.9 5.8 GB view
64k 181.9 10.1 9.6 GB current
Batching Results
Batch Size TG tok/s Speedup
49.2 1.00×
94.4 1.92×
114.7 2.33×