Compare AI models

Specifications
Parameters
7B/13B/33B/65B16B
Context window
2k8k
Benchmarks
HumanEval
23.7%48.1%
MATH
10.6%45.3%
WinoGrande
77%74.6%
MMLU
63.4%70%
GSM8K
50.9%77.4%
Overview
CompanyMetaMoonshot AI
Release dateFeb 24 2023Feb 22 2025
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

Moonlight leads LLaMA 1 on 4 of the 5 benchmarks they both report. LLaMA 1 shipped 729 days before Moonlight, so benchmark comparisons should account for the intervening progress.

LLaMA 1 has 7B/13B/33B/65B parameters, while Moonlight has 16B. Context windows are 2k (LLaMA 1) vs 8k (Moonlight).

On HumanEval, Moonlight leads at 48.1% vs LLaMA 1 at 23.7%. On MATH, Moonlight leads at 45.3% vs LLaMA 1 at 10.6%. On WinoGrande, LLaMA 1 leads at 77% vs Moonlight at 74.6%. On MMLU, Moonlight leads at 70% vs LLaMA 1 at 63.4%. On GSM8K, Moonlight leads at 77.4% vs LLaMA 1 at 50.9%.