ProteinGym Hard: leaderboard

Hard ProteinGym subset reported in Anthropic's Claude Opus 4.8 system card.

Metric: Rank correlation (self-reported). Source: benchmarklist.com. Status: saturation imminent. 10 models tracked.

Top models

#ModelScore
1Claude Opus 547.7
2Claude Mythos 545.8
3Claude Mythos Preview43.1
4Claude Opus 4.840
5Claude Opus 4.737.7
6Gemini 3.1 Pro (Preview)37
7Claude Sonnet 536.6
8GPT-5.6 Sol35.5
9Claude Sonnet 4.635

Interactive version: theaggregate.ai/benchmark?slug=proteingym-hard · How It Works · Data refreshed daily, snapshot 2026-09-05.