New benchmark tests AI's ability to predict post-cutoff discoveries

View on Reddit

Opus 4.7 scored 17.75% partial credit on 'The Singularity Gate' benchmark; all models got 0% fully correct. The benchmark evaluates frontier AI's capacity to anticipate paradigm-breaking scientific findings published after training cutoff.