Compare AI models

Specifications
Parameters
16B—
Context window
8k8k
API pricing
Cheapest input
—$30.00Azure
Cheapest output
—$60.00Azure
—$9,000Azure
Benchmarks
HumanEval
48.1%67%
WinoGrande
74.6%87.5%
MMLU
70%86.4%
GSM8K
77.4%92%
Overview
CompanyMoonshot AIOpenAI
Release dateFeb 22 2025Mar 14 2023
AccessOpen WeightClosed
Model detailsView modelView model

Frequently asked questions

GPT-4 leads Moonlight on 4 of the 4 benchmarks they both report (HumanEval, WinoGrande, MMLU, GSM8K). GPT-4 shipped 711 days before Moonlight, so benchmark comparisons should account for the intervening progress.

Context windows are 8k (Moonlight) vs 8k (GPT-4). Moonlight is open weight, while GPT-4 is closed.

On HumanEval, GPT-4 leads at 67% vs Moonlight at 48.1%. On WinoGrande, GPT-4 leads at 87.5% vs Moonlight at 74.6%. On MMLU, GPT-4 leads at 86.4% vs Moonlight at 70%. On GSM8K, GPT-4 leads at 92% vs Moonlight at 77.4%.