About MathDB

MathDB is a collaboration between Caltech mathematics faculty, graduate students and undergraduates, building an open database of the unsolved problems and conjectures of research mathematics: what each one says, what is known about it, and who has attempted it.

Problems catalogued

66,809

With a recorded AI attempt

211

Claimed solved with AI

184

Why now

In the past few months AI has started producing results in pure mathematics that working mathematicians take seriously.

• Riemann hypothesis: Claude Fable 5 improved a bound on the problem, open since 1859.
• Jacobian conjecture: a claimed counterexample settles it negatively in three or more dimensions, leaving the plane case open.
• Erdős unit-distance conjecture: disproved in 2026 by a construction packing more unit distances into the plane than the conjecture allowed, credited to an internal OpenAI model working with Noga Alon, W. T. Gowers and seven other mathematicians.

Results are arriving faster than the literature can absorb them, and most are claims rather than settled mathematics: not refereed, not independently replicated. That is the case for a public record, and it is why this database exists.

Every problem page states its verdict, its evidence, and whether anyone has checked it. The AI benchmark and the human board aggregate them.

Taking part

Anyone can read MathDB. A signed-in account can post a problem, record progress, or challenge a claimed solution. Corrections are especially welcome: a wrong statement of a conjecture is worse than no entry at all. Start from the problem list, browse by area, or find us on Discord.