Compare AI models

Specifications
Parameters
—16B
Context window
200k8k
Benchmarks
HumanEval
75.9%48.1%
GPQA Diamond
33.3%35.2%
MMLU
76.7%70%
GSM8K
88.9%77.4%
Overview
CompanyAnthropicMoonshot AI
Release dateMar 4 2024Feb 22 2025
AccessClosedOpen Weight
Model detailsView modelView model

Frequently asked questions

Claude 3 Haiku leads Moonlight on 3 of the 4 benchmarks they both report (HumanEval, GPQA Diamond, MMLU, GSM8K). Claude 3 Haiku shipped 355 days before Moonlight, so benchmark comparisons should account for the intervening progress.

Context windows are 200k (Claude 3 Haiku) vs 8k (Moonlight). Claude 3 Haiku is closed, while Moonlight is open weight.

On HumanEval, Claude 3 Haiku leads at 75.9% vs Moonlight at 48.1%. On GPQA Diamond, Moonlight leads at 35.2% vs Claude 3 Haiku at 33.3%. On MMLU, Claude 3 Haiku leads at 76.7% vs Moonlight at 70%. On GSM8K, Claude 3 Haiku leads at 88.9% vs Moonlight at 77.4%.