A 70B-parameter instruction-tuned model that Meta positions as approaching Llama 3.1 405B's performance on several benchmarks at a fraction of the inference cost.

Benchmarks

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

benchmark	published	measured
MMLU-Pro	68.9	—
GPQA	50.5	—

Lineage

↑ finetune Llama 3.1 70B Instruct

view in full graph →

Metadata

org	Meta
released	2024-12-06
params	70B
license	llama-3.3
hf id	meta-llama/Llama-3.3-70B-Instruct