Thread · 7 stories · Mar 3 – Sep 8
Knuth's conjecture, an Erdős disproof, 44 OEIS conjectures, Fermat's Last Theorem in Lean and a claimed Navier–Stokes proof, all inside six months.
Jump to timeline ↓In March 2026 Donald Knuth confirmed Claude had settled a Hamiltonian-cycle conjecture from his own drafts. By September the claims had escalated to a Millennium problem: OpenAI says 10,000 concurrent agents produced a 166-page Navier–Stokes proof in 88 hours.
In between came an Erdős conjecture disproved, DeepMind's AlphaProof Nexus clearing open Erdős and OEIS problems, and Claude formalising Fermat's Last Theorem in 13 million lines of Lean. The running question is no longer whether models can do research mathematics, but who checks the output — and what mathematicians do next.
