The Singularity Gate

A benchmark for AI-driven paradigm-shifting scientific discovery

No model fully predicts a discovery · partial credit scores below

Benchmark tasks are built around real scientific findings published after the newest model’s training cutoff, so contamination is not a concern. Passing the Singularity Gate is necessary, though not sufficient, for autonomous AI-driven discovery. A model that can predict paradigm-breaking discoveries is not necessarily Einstein-level, but one that cannot is certainly not.

Score Progression: Overall

Mean score: 0–100%

Reported scores are partial credit; no model has fully matched a finding on this corpus.

Reasoning effort: Opus 4.8 / Opus 4.7 / Sonnet 5 / Fable 5 / GPT-5.6 Sol = max · Gemini 3.1 Pro = high · GPT-5.5 = xhigh.

Native harness with tool use, web search disabled: Claude Code (Claude & Fable models) · Gemini CLI (Gemini 3.1 Pro) · Codex (GPT-5.5 / GPT-5.6 Sol).

v1.1. Findings with a public trace on or before GPT-5.6 Sol’s training cutoff (2026-02-16) were retired, so this release scores every model on the contamination-cleaned set.

Read the paper → Methodology

Leaderboard

* Claude Fable 5 answered 45% of the corpus (original run) and 39% (latest run). Its safety classifier refuses all life-sciences tasks plus much of Earth & Planetary and Chemistry, and the latest run refuses more than the original. The score shown is over the items both Fable runs answered; head-to-head there, original 23.7 vs latest 22.5. Every other model is scored over the full set. Per-field figures for Fable use its full answered set.

v1.1. Findings with a public trace on or before GPT-5.6 Sol’s training cutoff (2026-02-16) were retired, so this release scores every model on the contamination-cleaned set.

Per-field breakdown

Fable 5 leads four of five fields on the items it answered; GPT-5.6 Sol leads Life Sciences, which Fable did not attempt. Highlighted cell = field leader. Fable 5 figures use its full answered set.

Reasoning effort: Opus 4.8 / Opus 4.7 / Sonnet 5 / Fable 5 / GPT-5.6 Sol = max · Gemini 3.1 Pro = high · GPT-5.5 = xhigh.