Caltech Mathathon: First Hackathon for Research-Level Mathematics
Original: Caltech Mathathon – first hackathon ever devoted to research level mathematics
Why This Matters
Signals a shift in how frontier AI is being integrated into formal mathematical research and peer validation.
Caltech is hosting the world's first research-level mathematics hackathon on October 30–November 1, 2026. One hundred teams will use frontier AI models to tackle open mathematical conjectures over 40 hours, with over $2M in AI credits as prizes, judged by leading mathematicians.
The Caltech Mathathon, scheduled for October 30 to November 1, 2026, at the California Institute of Technology, is billed as the first hackathon ever devoted to research-level mathematics. One hundred teams will be given access to frontier AI models and 40 hours to solve open conjectures and develop new mathematical theories. Prize pools include over $2M in AI credits, awarded in two rounds: first at the event itself, then after the broader math community has had time to verify submitted results.
The event is motivated by a series of recent AI breakthroughs in pure mathematics: the disproof of Erdős's planar unit-distance conjecture (open for 80 years) on May 20, 2026; the construction of the first explicit non-sofic group (open 27 years) on August 1, 2026; and an unverified claim that the six-sphere admits a complex structure (open 78 years) on August 23, 2026.
Organizers frame the event around two central questions: how much AI can accelerate the path from ideation to peer-reviewed publication, and what role mathematicians play when AI can solve longstanding conjectures at speed. Participants will defend their results before leading mathematicians, who will assess both the validity and the team's understanding of the work. The event includes commitments to responsible AI use.