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

Why we chose it

This market exists because on September 8 OpenAI claimed its internal model — "significantly more capable than GPT-6 Astra" — orchestrated roughly 10,000 agents over 88 hours (~$15M of compute) to produce a Lean-verified proof of finite-time singularity in Navier-Stokes, complete with a priority soap opera: OpenAI started after hearing a rumor tied to NYU's Tristan Buckmaster and Anthropic's Levent Alpöge, whose secret work turned out to be on forced Euler, not Navier-Stokes.

So this market asks: does lightning strike the other five? Crucially, OpenAI says it already launched the model at the remaining problems — the swarm is loose, and resolution only needs a lab *announcement*, not Clay's blessing.

The case for 0: Navier-Stokes was uniquely gettable — a checkable counterexample sitting atop years of DeepMind-adjacent singularity work, whereas P vs NP, Hodge and Yang-Mills have no "almost there" partial result to extend; Riemann got a real jump (an Anthropic bound Anthropic itself says won't scale to a proof). The case for 1+: a sister Polymarket board prices BSD at 46% to fall next, Hodge at 30.5%, with only 13.5% on nothing through 2027 — comedy angle: traders think elliptic curves crack before Christmas. Board sums to 98%, so no longshot juice to mock.