Llama 3.2 3B Instruct

Meta llm

A lightweight 3B-parameter instruction-tuned model created via pruning and distillation from Llama 3.1 8B, designed for on-device and edge deployment.

overall ⌄

—

nothing published

price · $/M tokens

self-hosted

open weights, no first-party API

to run it yourself ⌄

2 GB

minimum · 7 GB recommended

HF ↗ Ollama ↗ HuggingFace ↗Blog ↗

Benchmarks

What its maker published, and what anyone else measured. Every figure links the document it came from.

benchmark	published	measured
MMLU-Pro	—	34.7
GPQA	—	22.4
AIME 2025	—	3.3
HLE	—	5.2

Lineage

↑ distill Llama 3.1 8B

view in full graph →

Metadata

org	Meta
released	2024-10-24
params	3B
license	llama-3.2
hf id	meta-llama/Llama-3.2-3B-Instruct