About Arbiter
A public works project for the hardest questions we know.
Arbiter is a leaderboard and a funding rail. On one side: the open problems in mathematics, physics, and quantitative science that still resist us. On the other: donors who want their dollars turned into focused AI compute — and a permanent, public record when something finally yields.
What you are funding
Every problem on Arbiter has its own purse. When you donate $10, $50, $100, or a custom amount, that money is earmarked for AI compute on that problem alone. Coordinated models attempt the work — exploring lemmas, testing approaches, discarding dead ends — for as long as the purse can pay for the run.
If the balance reaches zero before a solution, the attempt pauses. It does not quietly continue on someone else’s dime, and it does not redirect leftover funds. The next donor restarts the clock. Progress is continuous only when support is.
When a problem is solved, it moves to the solved leaderboard. We publish the full solution and the thinking patterns that produced it — including the work that did not pan out — so researchers and curious readers can audit, learn from, and extend the result.
Impact scores & ranking
Not every open problem weighs the same. Each entry carries an impact score from 1 to 100 — a judgment of how much solving it would move science, technology, and human understanding. You can rank the board by impact, by total donations, by number of donors, or filter between open and solved. The goal is simple: make scientific weight and community priority legible in the same place.
Unsolved pages
Explain why the problem matters and what discoveries or capabilities a solution would unlock downstream.
Solved pages
List related problems newly in reach and describe the longer arc of impact the result enables.
Living purses
Funding totals, donor counts, and progress toward the compute goal update as support arrives.
Open traces
Successful runs publish reasoning in full — not a press release, a working record.
Who this is for
Researchers — champion the conjecture that would unblock your field, and read every intermediate step when coordinated models make headway.
People excited about the future of AI — put frontier systems to work on problems that have resisted human attack for decades, with transparent accounting of what the compute actually produced.
Anyone who wants to advance research — no lab affiliation required. Pick a problem, give what you can, and join a public ledger of progress.
Ready to fund a run?
Open the leaderboard, sort by what you value, and put compute behind a problem worthy of the effort.
Browse open problemsStay in touch
Questions about a problem, a partnership, or listing a new entry? Send a note — we read every message.