Leaderboard
PreposterouslyDifficultMath
Overall results and scores on each mathematical-research ladder.
No results yet
Results for will appear here. Scores for untested models remain unknown.
| Rank | Model / system | Overall score | Milestone | Source / review | Run / release date |
|---|
✨ Reading the scores
The overall score is the sum of a model’s best score on each of the five ladders. A total is calculated when all five ladder scores are available. A public release covering fewer ladders is shown as an unranked lower bound (≥). Scores for missing tracks remain unknown.
Track scores record progress on the published milestone ladders. Difficulty calibration is subjective. Higher scores mark further progress within a track.
Zero means no scored advance beyond the ladder’s existing rigorous frontier. Scores for untested models remain unknown.
Self-reported scores and token totals were supplied by Thomas. GPT-6 Astra used 240 million tokens in total across the run. Run dates and benchmark revisions remain unknown; public transcripts are unavailable here.
The OpenAI release reaches Kakeya rung 4 once, through either H4 or M3. Its other four tracks are unreported. Scores map the public manuscript claims to the ladders and remain provisional; no formal proof of these endpoints was identified in the release. Sources and cross-reference
The abc / Vojta track requires a machine-checkable formal proof for scores of 2 or higher. Adjudication rules
An asterisk marks an endpoint whose eventual provability is uncertain in the source ladder.
A calibrated Mahler ladder is under development.