How many more Millennium Prize Problems will AI solve in 2026?

The market question is narrowly about qualifying announcements on the five listed Millennium Prize Problems other than Navier-Stokes, and that distinction matters a lot here. The recent OpenAI claim about Navier-Stokes shows that leading labs are willing to make bold public claims, but that specific problem is explicitly excluded from resolution, so it does not directly move the needle toward a Yes outcome. The separate report that OpenAI described substantial progress on another Millennium Prize Problem is interesting, but progress is not a solved announcement, and the market requires a lab to state that it has solved a qualifying problem by year-end. On the facts provided, there is still no confirmed qualifying announcement for any of the five counted problems.

The strongest argument for No is that these problems are among the hardest open problems in mathematics, and even a breakthrough by a top AI lab would need to clear a very high bar in a short window. The market is not asking whether AI makes advances, formalizes partial results, or produces plausible proofs; it asks whether an eligible lab publicly announces a solved status for one of the listed problems. That means a number of things have to go right simultaneously: the model has to produce a correct proof, the lab has to validate it enough to make a public announcement, and the announcement has to be about one of the counted problems rather than Navier-Stokes. With only a few months left in the year, the time constraint is a major obstacle.

The main argument for Yes is that the recent reporting suggests the frontier may have shifted faster than expected. If an internal system can plausibly claim a major result on Navier-Stokes and another unnamed Millennium problem is already in a state described as substantial progress, then there is at least some chance that one of the five qualifying problems has been solved internally and could be announced before December 31. The market price still assigns meaningful probability to zero solutions, but it is no longer a slam dunk because the combination of advanced reasoning systems, large-scale agentic search, and formal verification tools makes a first breakthrough less implausible than it was earlier in the year. Even so, the evidence so far points more toward imminent claims of progress than toward a verified qualifying announcement.

Overall, I lean toward 0 more qualifying AI solutions in 2026, but not strongly enough to be extreme. The recent news increases the chance that something dramatic could happen, yet the exclusion of Navier-Stokes, the lack of any identified solved claim for the five counted problems, and the practical difficulty of turning a tentative result into a public solved announcement keep the No side ahead.