Compare AI models

Specifications
Parameters
16B—
Context window
8k200k
Benchmarks
HumanEval
48.1%99.3%
GPQA Diamond
35.2%81.3%
SimpleQA
5.6%19.3%
MMLU
70%90.3%
Overview
CompanyMoonshot AIOpenAI
Release dateFeb 22 2025Apr 16 2025
AccessOpen WeightClosed
Model detailsView modelView model

Frequently asked questions

o4-mini-high leads Moonlight on 4 of the 4 benchmarks they both report (HumanEval, GPQA Diamond, SimpleQA, MMLU). Moonlight shipped 53 days before o4-mini-high, so benchmark comparisons should account for the intervening progress.

Context windows are 8k (Moonlight) vs 200k (o4-mini-high). Moonlight is open weight, while o4-mini-high is closed.

On HumanEval, o4-mini-high leads at 99.3% vs Moonlight at 48.1%. On GPQA Diamond, o4-mini-high leads at 81.3% vs Moonlight at 35.2%. On SimpleQA, o4-mini-high leads at 19.3% vs Moonlight at 5.6%. On MMLU, o4-mini-high leads at 90.3% vs Moonlight at 70%.