Status as of October 2026. Despite an extraordinary surge in AI mathematical capability, none of the nine target problems has a full, accepted resolution. Recent major breakthroughs all fall short of the required full statements: the Navier–Stokes proof covers only the forced case (Clay statements C and D) On the Navier–Stokes Millennium Prize Problem Existence and Smoothness of the Navier–Stokes Equation, the quasi-Riemann zero-free region and improved critical-line results leave the full Riemann Hypothesis distant AI Proved the Quasi-Riemann Hypothesis. What Happens ... Learning more about Claude's mathematical capabilities AI Proved the Quasi-Riemann Hypothesis. What Happens ..., and the Yang–Mills existence proofs explicitly exclude the crucial mass gap Working with academic communities to build scientific ... Secret Unified Physics Slop | Not Even Wrong Secret Unified Physics Slop | Not Even Wrong. Similarly, the Exponential Time Hypothesis refutation leaves P vs NP open OpenAI says its AI model has disproved the Exponential ... The Exponential Time Hypothesis is False, and the released Hodge and Birch–Swinnerton-Dyer results are restricted cases The most exciting claims from OpenAI's heap of new proofs Sharing AI progress in mathematics OpenAI's 722 Math Manuscripts: The Results, Proofs, ....

Drivers of near-term timelines. The pace of recent capability gains strongly supports a near-term breakthrough, anchoring the early percentiles of this estimate. In just a few months, AI systems saturated benchmarks like FrontierMath Tier 4 and First Proof Third Batch Press Release FrontierMath Erdős, claimed a proof of local smoothing in all dimensions Working with academic communities to build scientific ..., and demonstrated massive parallelization by deploying 10,000 concurrent agents for 88 hours On the Navier–Stokes Millennium Prize Problem. Furthermore, any of these nine problems could fall to a disproof (such as a Hodge counterexample or unforced Navier–Stokes blowup), which is highly amenable to concentrated agentic search. With compute costs falling roughly 47% per quarter and some internal lab results currently being withheld for write-ups, a qualifying public release in 2027 is highly plausible. These drivers support a 10th percentile of April 1, 2027, and a 25th percentile of October 1, 2027.

Barriers establishing a mid-2028 median. Extrapolating benchmark success directly to Millennium Prize problems risks underestimating their historical intractability. Solving partial cases does not necessarily bridge the conceptual gaps required for the full conjectures; for instance, the techniques used to advance the zeta-zero critical-line fraction are not expected to prove the full Riemann Hypothesis Learning more about Claude's mathematical capabilities AI Proved the Quasi-Riemann Hypothesis. What Happens .... Additionally, early AI claims are often contested. A Lean formalization mismatch was already noted in the forced Navier–Stokes proof Navier-Stokes lost in translation: Why Lean verification of ... Navier–Stokes priority controversy, and an apparent Collatz disproof earlier in 2026 exploited a Lean kernel bug Postmortem for Kernel Soundness Bug #14576. Because the resolution criteria demand a full proof that eventually gains expert acceptance, flawed early claims will fail, shifting the timeline of the first valid release later. Balancing these barriers against rapid capability gains puts the median at August 1, 2028.

The long right tail. The distribution extends into 2040 to account for the risk of a capability plateau. While continued exponential progress is the baseline, there is a meaningful probability that these specific deep problems—characterized by known hurdles like the parity barrier or the mass gap—will resist current AI paradigms for a decade or more. If scaling current AI systems encounters a conceptual wall, solving these remaining problems may require genuinely new mathematical ideas The Fate of the Riemann Hypothesis: Revisited rather than brute-forced search, significantly delaying a complete and accepted resolution. I place the 75th percentile at January 1, 2031, but retain a 10% chance that none of the nine problems is resolved before 2040, consistent with the related probability estimate . That substantial residual risk puts the 90th percentile at January 1, 2040, rather than treating a breakthrough during the 2030s as nearly assured.