Caltech Mathathon 2026: First Hackathon Focused on Research‑Level Mathematics

Quick Take

The Caltech Mathathon (Oct 30–Nov 1 2026) will run a 40‑hour, $2 M+ AI‑credit sprint where up to 100 teams attempt to solve open research‑level math problems with frontier language models, then defend their results before expert mathematicians.

What the Mathathon Is

Format: A three‑day hackathon hosted at the California Institute of Technology. Teams receive access to large‑scale AI models and a budget of AI compute credits (over $2 million worth) to explore conjectures.
Goal: Produce well‑explained, verifiable solutions to open problems, then present the work to a panel of leading mathematicians for evaluation.
Prizes: Awards for the most promising and clearly communicated results, with a second round of prizes after community verification.
Eligibility: Open to any university or independent team; organizers are undergraduate students at Caltech (not officially representing the institute or its sponsors).
Responsible AI: The organizers publish a FAQ outlining commitments to ethical AI use and transparent reporting.

Why It Matters

Recent AI breakthroughs—disproving Erdős’s planar unit‑distance conjecture, constructing the first explicit non‑sofic group, and (unverified) showing a complex structure on the six‑sphere—demonstrate that large language models can contribute to deep pure‑math research. The Mathathon tests how quickly AI can move a conjecture from idea to peer‑reviewed result and probes the evolving role of human mathematicians when AI can generate proofs faster than traditional workflows.

Community Reactions on Hacker News

Organizers Clarify Their Position

"We are a team of undergrads at Caltech. We don't represent Caltech, any Caltech departments, or any of our sponsors. We don't receive monetary compensation. All the funding raised goes toward paying our judges and participants. Our goal is to promote responsible AI use." – brian-bfz (organizer)

Skepticism About Hackathon Timing

"Waiting on the output of an LLM for 40 hours feels antithetical to what makes classic hackathons appealing. LLM math progress usually involves weeks of model runs with intermittent correction, not a short sprint." – Semkas

Concerns Over Labor and Motivation

"Is this big‑lab push to move maths forward, or a way to use professional mathematicians as cheap labour for validating LLM outputs?" – youoy

Institutional Context

"Caltech’s CS department has struggled to recruit top talent; this hackathon offers a way for students to gain ML recognition that the school currently lacks." – corinthia (recent Caltech graduate)

Technical Perspective on Model Harnessing

"Current generic harnesses waste reasoning tokens; math requires deep reasoning, so a focused hackathon could serve as a testbed for more efficient model usage." – chaoxu

Historical Counterpoint

"This isn’t the first math‑focused hackathon; William Stein ran SageMath hackathons on the BSD conjecture two decades ago." – boothby

Practical Resources Shared by Participants

"Open problems in computational geometry (Erik Demaine et al., 2024) and broader lists of unsolved problems are useful starting points for teams." – nill0

Expected Outcomes

Benchmarking AI‑Assisted Proof Generation – By measuring how many conjectures are resolved within 40 hours, the Mathathon will provide concrete data on AI’s speed advantage over traditional research cycles.
Human‑AI Collaboration Models – Teams must not only generate results but also explain them clearly, revealing how mathematicians can add value through insight, verification, and exposition.
Community Verification Pipeline – The two‑stage prize structure encourages rigorous post‑event peer review, helping the broader math community assess the reliability of AI‑produced proofs.
Tooling Improvements – Feedback from participants on token efficiency, prompting strategies, and model harnesses will inform the next generation of math‑focused AI platforms.

How to Get Involved

Apply: Teams can submit applications via the official site (apply link).
Prepare: Review the provided lists of open problems (e.g., computational geometry challenges) and consider how to structure prompts for maximal reasoning depth.
Engage: Follow the organizers on the event’s FAQ page for updates on hardware support, model access, and responsible‑AI guidelines.

Conclusion

The Caltech Mathathon represents a bold experiment at the intersection of competitive hackathons and frontier mathematical research. By allocating massive AI compute resources to a short, intensive sprint, it will test whether large language models can accelerate the discovery pipeline and redefine the mathematician’s role in an AI‑augmented future.