Model Benchmark — Leaderboard

Overall Leaderboard

Scores are normalized (higher is better) Updated 2h ago

Model Alpha Model Beta Model Gamma

Arena Elo

1,325

↑ 18 vs last week

MMLU (5-shot)

89.6%

↑ 1.3%

HumanEval Pass@1

72.4%

↑ 2.8%

GPQA (Diamond)

56.7%

↓ 0.6%

Model	Arena Elo	MMLU	HumanEval	GPQA
1 Atlas 3 Pro Proprietary	1,325	89.6%	72.4%	56.7%
2 Vega 70B 70B	1,281	87.1%	69.2%	53.9%
3 Nova Flash Preview	1,217	84.0%	66.1%	51.2%

Scores aggregate multiple datasets and are subject to change as evaluation suites are refreshed.