PredictPedia tracks two related prediction-market contracts that ask whether artificial intelligence labs will solve Millennium Prize Problems. The Clay Mathematics Institute lists seven Millennium Prize Problems, and these contracts focus on whether an AI lab can be credited with a solution to one of them. The contracts do not ask whether anyone wins the Clay prize. The Kalshi rules state that an award by the Clay Mathematics Institute is not required [1]. The contracts differ mainly in how they define a solution. The Kalshi event, titled "When will an AI lab solve another Millennium Prize Problem?", requires a complete proof, disproof, or counterexample that addresses the problem as originally formulated. That work must be published in a reputable peer-reviewed mathematics or scientific journal, or the Clay Mathematics Institute or another recognized academic body must officially state that the problem has been resolved by reference to a specific written work. Alternatively, at least two named Source Agencies must report that the problem has been resolved by reference to a publicly available written work [1]. Under these rules, a preprint, announcement, blog post, conference presentation, or machine-checked formalization does not by itself count, although it may be the written work relied upon for official recognition [1]. The Kalshi rules also exclude partial results, special cases, results that depend on unproven assumptions, and retracted or withdrawn claims [1]. The Kalshi rules also require attribution. A solution counts for a lab when Source Agencies report that the lab, its personnel, or an AI system it developed and operated produced the solution in whole or substantial part, and the lab or system is credited as an author, contributor, or producer. Collaborative work qualifies, but independent third-party use of a lab's public model without participation or credit does not [1]. The two Polymarket events use a simpler standard. The event "AI lab announces another Millennium Prize solution by...?" resolves Yes if an eligible AI lab, or an official representative of that lab, announces that the lab has solved one of the qualifying problems by the deadline in the title. The qualifying problems are the Riemann Hypothesis, P versus NP, the Yang-Mills existence and mass gap problem, the Hodge Conjecture, and the Birch and Swinnerton-Dyer Conjecture. The Navier-Stokes problem is excluded. The announcement must express that the problem has been solved, and partial results or progress do not qualify. No further confirmation from the Clay Mathematics Institute or any other organization is required [2]. The companion event "How many more Millennium Prize Problems will AI solve in 2026?" counts distinct qualifying problems for which an eligible lab announces a solution by December 31, 2026, with each problem counted once no matter how many labs or announcements address it [3]. The listed eligible labs in the Polymarket rules include a broad set of companies, among them Anthropic, Google, OpenAI, Meta, and Microsoft, together with other companies that have a model on a named leaderboard at the time of an announcement [2][3]. Announcements from companies outside that definition do not qualify [2]. For the market snapshots, the Kalshi event includes five contracts on whether any AI lab solves a problem before several dated cutoffs, and the two Polymarket events include eight contracts on announcements or on counts for 2026 [1][2][3]. Each event's full set of saved markets is included in the sample, but the providers do not establish that they list every possible outcome. Snapshot prices reflect the moment the data was observed. They are not executable quotes or verified probabilities, and volume is not a measure of liquidity or the number of traders.

Historical Context

Millennium Prize Problems are a well-known list of unsolved problems. A Wikipedia article on unsolved problems in mathematics notes that prizes are often awarded for the solution of long-standing problems and that lists such as the Millennium Prize Problems receive considerable attention [5]. On 7 October 2026, OpenAI released 722 manuscripts that claimed to have solved a large number of open and unsolved major mathematical problems, according to a Wikipedia article retrieved on 26 October 2026 [4]. The same article reports that these included partial results on several of the qualifying problems, and that mathematicians raised concerns about the manuscripts' validity and exposition style and about whether releasing them advanced human understanding [4]. The retrieved source describes these events as background and does not state how they affect any contract or whether any contract has been or will be resolved in light of them.

Frequently Asked Questions