Overview
- OpenAI announced on Tuesday that an unreleased internal model, coordinated across roughly 10,000 autonomous agents, produced a proposed proof that three-dimensional Navier–Stokes equations can develop a finite-time singularity after about 88 hours of work and 17 additional hours to formalize the argument in Lean.
- The company published a technical paper and a machine-checkable Lean proof but said it will not claim the $1 million Clay Mathematics Institute prize and acknowledged that the Clay Institute and independent mathematicians must still validate the result through peer review and community acceptance.
- Two mathematicians, Tristan Buckmaster and Levent Alpöge, say they were working on related research and have questioned whether information about their progress reached OpenAI; OpenAI denies direct access to their unpublished work while admitting it cannot completely rule out indirect improvement from de-identified user-derived data.
- OpenAI reported the effort exchanged roughly 2.7 million messages, generated about 130 billion output tokens across its agents, and cost what company statements and reporting estimate as multi‑million dollars in compute, illustrating a new, large-scale agent-based research method.
- If the proof survives careful independent checking, it would mark a major moment for AI-assisted mathematics, but acceptance will depend on traditional peer review and broader community verification and it raises urgent questions about data governance, authorship and how researchers use closed lab tools.