Mathematicians spent decades building one of the largest proofs in history. A swarm of AI agents just checked the whole thing in 24 days.
Axiom Math, an AI startup based in Palo Alto, announced on October 26, 2026 that its AxiomProver system completed the first full computer-checked formalization of the Classification of Finite Simple Groups, a result so sprawling it earned the nickname the “Enormous Theorem.”
What Axiom actually did
The original proof runs to approximately 15,000 pages. It drew on contributions from over 100 mathematicians spread across more than 500 papers.
Axiom’s agents translated the argument into Lean, a language in which a computer verifies every logical step.
The output was approximately 43.1 million lines of Lean code. According to the research findings, that makes it the largest Lean codebase ever created.
The work took 24 days from start to finish.
The errors hiding in plain sight
Axiom’s process identified and corrected over 2,000 errors in the existing literature. Most were apparently minor, but 34 were classified as notable deviations.
One was more serious. The team found a significant gap in the Unbalanced Group Theorem, a piece of the larger argument, and had to develop a new segment of proof to close it.
The review process was overseen by expert mathematicians, according to the findings.
Why the Enormous Theorem matters
Finite simple groups are often described as the atoms of symmetry. Every finite group can be broken down into them, much like molecules break down into elements.
The classification says every finite simple group falls into one of four buckets. It is either cyclic of prime order, an alternating group, a member of one of 16 families of groups of Lie type, or one of 26 sporadic groups.
The sporadic groups are the oddballs that refuse to fit any pattern. The most famous is the Monster group.
Axiom’s formalization confirms that this list is complete.
John G. Thompson is among the historical contributors to the original proof.
How the agent swarm works
AxiomProver is not one model grinding through pages. It is a multi-agent system in which different agents handle different jobs.
Those roles include decomposition, search, proof-writing, and review.
The company is led by founding CEO Carina Hong. Mathematician Ken Ono is also named as a key figure connected to the effort.
Earlier in the year, the system reportedly achieved a perfect score on the Putnam exam, the notoriously difficult undergraduate math competition, and solved open conjectures.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.