Compare AI models

Specifications
Parameters
1B/4B/12B/27B16B
Context window
—8k
API pricing
Cheapest input
$0.04DeepInfra—
Cheapest output
$0.08DeepInfra—
$15.00DeepInfra · bf16—
Benchmarks
LiveCodeBench
29.7%14.3%
HumanEval
87.8%48.1%
MATH
89%45.3%
GPQA Diamond
42.4%35.2%
SimpleQA
10%5.6%
MMLU-Pro
67.5%42.4%
MMLU
76.9%70%
GSM8K
95.9%77.4%
Overview
CompanyGoogleMoonshot AI
Release dateMar 12 2025Feb 22 2025
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

Gemma 3 leads Moonlight on 8 of the 8 benchmarks they both report. Moonlight shipped 18 days before Gemma 3, so benchmark comparisons should account for the intervening progress.

Gemma 3 has 1B/4B/12B/27B parameters, while Moonlight has 16B.

On LiveCodeBench, Gemma 3 leads at 29.7% vs Moonlight at 14.3%. On HumanEval, Gemma 3 leads at 87.8% vs Moonlight at 48.1%. On MATH, Gemma 3 leads at 89% vs Moonlight at 45.3%. On GPQA Diamond, Gemma 3 leads at 42.4% vs Moonlight at 35.2%. On SimpleQA, Gemma 3 leads at 10% vs Moonlight at 5.6%. On MMLU-Pro, Gemma 3 leads at 67.5% vs Moonlight at 42.4%. On MMLU, Gemma 3 leads at 76.9% vs Moonlight at 70%. On GSM8K, Gemma 3 leads at 95.9% vs Moonlight at 77.4%.