I just released a new benchmark called The Singularity Gate. Tests whether frontier AI can predict paradigm-breaking scientific discoveries published after their training cutoff. Top score: 17.75% (partial credit, Opus 4.7). Fully-correct outcome rate: 0% across all respondents. Passing the Singularity Gate is necessary, though not sufficient, for autonomous AI-driven discovery. A model that can predict paradigm-breaking discoveries isn't necessarily Einstein-level. But a model that can't is definitely not. https://preview.redd.it/fjf5jz0wow3h1.png?width=900&format=png&auto=webp&s=465df48dd9959f190285ee250266e109e59b4cca Claude Opus 4.7 (max) - 17.75% GPT-5.5 (xhigh) - 16.08% Claude Opus 4.6 (max) - 15.11% Gemini 3.1 Pro (high) - 14.42% Claude Sonnet 4.6 (max) - 13.67% These are partial-credit scores. No model fully predicts a discovery. Happy to discuss methodology, related work, or the framing in the comments.…