Atlas

Benchmarks

← All benchmarks

GDP.pdf (Mean Criteria Pass Rate)

Professional Work · 2026-07-24

GDP.pdf scored as mean criteria pass rate: the fraction of rubric conditions satisfied across tasks, averaged over completed runs, rather than the strict per-prompt accuracy the Surge AI leaderboard reports. Anthropic reports this metric from an internal harness, so its scores are far higher than leaderboard accuracy and are not comparable with it.

Top models (higher is better)

ModelScore
Claude Mythos 587.3
Claude Opus 585.5
Opus 4.884.8
Sonnet 567.5
Loading Atlas data…