[ Front page ] [ The edge ] [ Finished ] [ About ] [ Links ]
THIS ONE IS FINISHED. It stopped being a frontier on 2026-09-24: superseded. The page stays, because what it recorded still happened.
Epoch AI stopped running this set on 2026-06-08. Every FrontierMath set of that generation stops within two days of it, and a second-version tier four opens on 2026-06-09 and has been run sixty-four times since, most recently six days before this closure. The definition here names the set of 2025-07-01, so 47.9 per cent is a final score rather than a frontier: nothing can move it because nobody is taking the test. The successor is not opened in its place today, because the best score recorded on it is 97.6 per cent, which is past the bar that closed #0012 and #0015.
47.9 per cent
Held by AI Co-Mathematician. Set 2026-05-08. Written down here 2026-09-07.
FrontierMath is a collection of mathematics problems written by research mathematicians specifically to be hard for machines, and kept private so that no model can have been trained on the answers. It is graded in tiers. The lower tiers are finished - models score above ninety per cent and the test has stopped distinguishing anything, which is why they sit on the closed page. The top tier is not finished. It stands below half, and a single problem there takes a specialist hours or days. This is the number to watch, because when it saturates as well the field will need a harder examination than research mathematicians currently know how to write, and nobody has one ready.
That description was written when this frontier was opened, on 2026-09-07, and is never rewritten. Where it and the number above it disagree, the number is the reading and the description is history.
Where to check it: https://epoch.ai/data/ai-benchmarking-dashboard
The program reads it from https://epoch.ai/data/benchmarks.csv, which is the same figure in a form a script can parse.
Previous: Unsolved integer sequences worked out Next: Real software issues resolved
[ Front page ] [ The edge ] [ Finished ] [ About ] [ Links ]