How many more Millennium Prize Problems will AI solve in 2026?
This event counts how many of five named Millennium Prize Problems an eligible AI lab announces it has solved by the end of 2026. Each outcome market covers one count, from 0 to 5. The key rule is that the announcement alone is enough: no confirmation from the Clay Math Institute or anyone else is needed, but the lab must claim the work as its own.
Leading right now: 1 68%, 0 20%, 2 10%, 3 3%.
| Market | Chance of yes | Deadline |
|---|---|---|
| Will AI solve 1 more Millennium Prize Problem in 2026? | 68.5% | Jan 1, 2027, 4:59 UTC |
| Will AI solve 0 more Millennium Prize Problems in 2026? | 19.5% | Jan 1, 2027, 4:59 UTC |
| Will AI solve 2 more Millennium Prize Problems in 2026? | 10.2% | Jan 1, 2027, 4:59 UTC |
| Will AI solve 3 more Millennium Prize Problems in 2026? | 2.9% | Jan 1, 2027, 4:59 UTC |
| Will AI solve 4 more Millennium Prize Problems in 2026? | 0.9% | Jan 1, 2027, 4:59 UTC |
| Will AI solve 5 more Millennium Prize Problems in 2026? | 0.9% | Jan 1, 2027, 4:59 UTC |
Odds read from Polymarket at 22:18 UTC on Nov 21, 2026. $690K traded so far, $98K in the last 24 hours.
How this market resolves
Yes: Each market resolves YES if the number of distinct qualifying problems announced as solved by an eligible AI lab by the deadline equals the number in its title. For example, the '1 more' market is YES if exactly one qualifying problem is announced, and the '0 more' market is YES if none are.
No: A market resolves NO if the final count of distinct qualifying problems differs from its number. The '0 more' market is NO if at least one qualifying announcement is made. The 'N more' markets are NO if the count is higher or lower than N.
When it really ends
December 31, 2026, 11:59 PM ET (Eastern Time). That is 2027-01-01 04:59 UTC, assuming standard time (UTC-5) applies on that date. The endDate field in the data (2027-01-01T04:59:00Z) matches this moment, so the two do not differ.
What counts
- An announcement by an eligible AI lab or its official representative that states one of the five listed problems has been solved.
- An announcement that presents the solution as the work of the lab, its researchers or its models, alone or jointly with outside researchers.
- Several labs announcing the same problem, which still counts as one problem.
- Any company with a model on the MMLU-Pro leaderboard at the time of the announcement, in addition to the named labs.
- Subsidiaries and research divisions of the listed labs.
What does not count
- Announcements about the Navier-Stokes existence and smoothness problem, which is explicitly excluded.
- Partial results or progress toward a solution.
- An announcement that only credits or congratulates outside researchers, even if they used the lab's models, compute or research credits.
- Claims by companies, organizations or individuals that are not eligible labs.
- Repeat announcements for the same problem, which do not raise the count.
Fine print people miss
- The word 'solution' is judged by what the announcement says. No confirmation from Clay Math Institute or any other body is required, so an unverified claim could count.
- The rules do not say what happens if a lab announces a solution and later retracts it or it is shown to be wrong. They are silent on this.
- The primary source is official information from the labs, but a consensus of credible reporting may also be used.
- The eligible list is open-ended: any company with a model on the MMLU-Pro leaderboard at the time of the announcement qualifies.
- The five qualifying problems are the Birch and Swinnerton-Dyer Conjecture, the Hodge Conjecture, P versus NP, the Riemann Hypothesis, and Yang-Mills existence and mass gap.
- The title says 'more', but the rules count announcements made by the deadline and do not mention a start date. They do not say whether earlier announcements would count.
What moves the price
- A major lab publicly claiming a full proof of a listed problem would push the count up, and so favor higher-number markets over the '0 more' market.
- A lab saying a result is only partial or progress would not change the count under the rules.
- Experts or the lab itself finding an error in a claimed proof might lead to a retraction, though the rules do not say how that would be treated.
- Announcements where outside mathematicians are credited for a result would not count, so they would not push the count up.
- The passing of time with no claims makes the '0 more' outcome more likely, as the deadline gets closer.
Put a bot on this market
A PROWL bot reads the news around the clock, judges each post against these exact rules, and trades inside the limits you set. It starts in paper mode.
Questions people ask
Which Millennium Prize Problems count in this market?
Five count: Birch and Swinnerton-Dyer, Hodge, P versus NP, Riemann Hypothesis, and Yang-Mills existence and mass gap. Navier-Stokes is explicitly excluded.
Does the Clay Math Institute have to confirm the solution?
No. The rules say no confirmation from Clay Math Institute or any other organization is required. An eligible lab's announcement that the problem is solved is enough.
When does the market close?
At 11:59 PM Eastern Time on December 31, 2026. That is 04:59 UTC on January 1, 2027.
Does it count if an AI helps a mathematician who announces the proof?
Not if the announcement only credits the outside researchers, even if they used the lab's models, compute or credits. The lab must present the solution as its own work, alone or jointly.
Do multiple labs solving the same problem count twice?
No. Each problem is counted once, however many labs or announcements there are.
Which companies count as eligible AI labs?
A long named list, including Anthropic, Google, OpenAI, Meta, Nvidia and DeepSeek, plus their subsidiaries and research divisions. Any other company with a model on the MMLU-Pro leaderboard at the time of the announcement also counts.
More Science markets explained
Written from the market's own rules text on Nov 16, 2026. If the rules on Polymarket differ from this page, the rules on Polymarket govern.