Introduction

In 2000, the Clay Mathematics Institute designated seven Millennium Prize Problems, each carrying a US$1 million reward, as the deepest open questions at the frontier of mathematics []. They are the P versus NP problem, the Hodge conjecture, the Riemann hypothesis, the Yang-Mills existence and mass gap problem, the Navier-Stokes existence and smoothness problem, the Birch and Swinnerton-Dyer conjecture, and the Poincare´ conjecture. Of these, only the Poincare´ conjecture has previously been resolved, proved by Grigori Perelman between 2002 and 2003 using Richard Hamilton’s Ricci flow techniques []. The remaining six problems, including Navier-Stokes, had stood unresolved for a quarter of a century.

On 8 September 2026, OpenAI announced that an internal, unreleased AI system had constructed an exact, finite-time singular solution to the three-dimensional Navier-Stokes equations, the foundational mathemati-cal description of fluid motion, and specifically whether their solutions can develop a ”singularity,” a point at which an infinitesimally small region of fluid begins to move at infinite speed []. The company published a 166-page manuscript alongside a machine-checkable formalization in the Lean proof assistant [].

The Equations

The incompressible Navier-Stokes equations, formalized by Navier (1822) and Stokes (1845), describe fluid velocity u(x, t) and pressure p(x, t) under viscosity ν and forcing f :

∂u + (u · ∇)u = −∇p + ν∇2u + f ,∇ · u = 0.(1)

∂t

Their frictionless simplification, the Euler equations (Euler, 1757), drops the viscous term:

∂u

  • (u · ∇)u = −∇p + f ,∇ · u = 0.(2)

∂t

The Millennium Prize question, as formally posed by Fefferman [] on behalf of the Clay Institute, asks whether, in unbounded three-dimensional space with smooth forcing and finite kinetic energy, solutions must stay smooth forever or can develop a singularity, a point where velocity becomes unbounded in finite time.

The Problem and its Significance

Formulated in the mid-nineteenth century, the Navier–Stokes equations apply Newton’s second law of mo-tion to fluids and underpin disciplines ranging from aeronautical engineering to meteorology and cardiovas-cular medicine []. The Clay Institute’s version of the problem asks whether, in three-dimensional space ex-tending infinitely in all directions, and under smoothly behaved external forces, solutions to these equations can ever ”blow up.” According to Charles Fefferman of Princeton University, who authored the Institute’s official problem statement, this question had remained open for over 160 years [].

Nearly Two Centuries of Struggle

Direct analysis of the equations’ nonlinearity has resisted proof for generations. Jean Leray proved in 1934 only that weak, non-smooth solutions to the Navier-Stokes system must exist globally in time, without settling whether those solutions remain smooth; a foundational result with no newer equivalent []. For decades afterward, the field’s dominant strategy was numerical rather than purely analytic. In 2025, Jiajie Chen and Thomas Y. Hou provided a rigorous, computer-assisted proof of finite-time blow-up for the 3D Euler equations with smooth initial data in a bounded domain, building on numerical evidence for a potential singularity that Guo Luo and Hou had first identified in 2014 []. Their method still relied on computer-assisted estimates to control the stability of a numerically constructed blow-up profile, illustrating how heavily the field continued to lean on computation even in its most rigorous results. A rarer purely analytic exception came from Diego Co´rdoba, Luis Mart´ınez-Zoroa, and Fan Zheng, who in 2023 proved finite-time blow-up for the Euler equations using a constructive method, an infinite sequence of disjoint, smooth vorticity profiles, but only for a velocity field with limited (C1,α) smoothness away from the origin, short of the full regularity the Millennium Prize requires [].

The AI-Generated Proof

OpenAI’s effort began on 1 September 2026 after rumors that Anthropic’s Levent Alpo¨ge and NYU’s Tristan Buckmaster were close to a related result []. Roughly 100 agents first resolved a simplified Euler regularity question in about 50 hours; a network that grew to around 10,000 concurrent agents then attacked Navier-Stokes directly, reaching a result after 88 hours, with a further 17 hours spent on Lean formalization []. The effort generated 2.7 million inter-agent messages and roughly 130 billion output tokens []. OpenAI has stated it is not claiming the associated $1 million prize [].

Proof Strategy (summarized)

The following outline reflects OpenAI’s public description of its construction:

  1. Start at rest. The construction begins with a fluid at zero initial velocity.
  2. Apply smooth, localized forcing. A forcing function f , smooth and compactly supported in space and time is introduced, satisfying the Prize’s regularity requirement [].
  3. Generate a contracting vortex. The resulting flow forms a vortex tube whose radius narrows faster than its length, spiraling inward while stretching along its axis, the classical vortex-stretching mecha-nism long suspected of driving blow-up [].
  4. Layer an infinite cascade. Following the layered-profile strategy pioneered by Co´rdoba, Mart´ınez-Zoroa, and Zheng, infinitely many individually smooth solutions are combined so that the combination

remains singular while the forcing stays smooth [].

  1. Bound the energy. Analytic estimates confirm the flow’s kinetic energy stays finite throughout, even as velocity becomes unbounded, satisfying the Prize’s finite-energy condition [].
  2. Formalize in Lean. Each analytic step is translated into a machine-checkable statement, closing logical gaps that human review alone might miss [].

The full argument is available in OpenAI’s public manuscript and Lean repository for direct inspection [].

Implications

The result marks a shift from a decade of approximation-heavy, computer-assisted proof toward AI-generated exact analytic proof, compressing work that might otherwise take human mathematicians years into days []. For STEM audiences in Eswatini, it illustrates AI’s growing reach into domains once considered uniquely human. The manuscript and Lean formalization are public, and formal verification by the wider mathemat-ical community is now underway [].

Conclusion

An exact singular solution to the Navier-Stokes equations resolves a question open since the equations were written down nearly two centuries ago, a genuine milestone arrived at through unprecedented computational scale.

References

  1. Charles L. Fefferman. Existence and smoothness of the Navier–Stokes equation. Technical report, Clay Mathematics Institute, 2000. URL .
  2. Grigori Perelman. The entropy formula for the Ricci flow and its geometric applications, 2002. URL

.

  1. OpenAI. On the Navier–Stokes Millennium Prize Problem. , September 2026. Accessed 9 September 2026.
  2. Kingy.ai. OpenAI’s Navier–Stokes Proof Claim: Evidence and Dispute. Kingy.ai, September 2026. URL . Ac-cessed 9 September 2026.
  3. Konstantin Kakaes. AI has solved one of Math’s $1 Million Millennium Prize Problems.

QuantaMagazine,September2026.URL

. Accessed 9 September 2026.

  1. Jean Leray. Sur le mouvement d’un liquide visqueux emplissant l’espace. Acta Mathematica, 63(1): 193–248, 1934. doi: 10.1007/BF02547354.
  2. Jiajie Chen and Thomas Y. Hou. Singularity formation in 3D Euler equations with smooth initial data and boundary. Proceedings of the National Academy of Sciences, 122(26), 2025. doi: 10.1073/pnas. 2500940122.
  3. Diego Co´rdoba, Luis Mart´ınez-Zoroa, and Fan Zheng. Finite time singularities to the 3D incompress-ible Euler equations for solutions in c∞(R3 \ {0}) ∩ c1,α ∩ l2, 2023. URL .
  4. Time News.OpenAI AI Model Solves Navier-Stokes Millennium Prize Math Problem.Time News, September 2026.URL . Accessed 9 September 2026.
  5. Unite.AI.OpenAIsaysInternalAISystemResolvedtheNavier–Stokes Problem.Unite.AI,September2026.URL . Accessed 9 September 2026.
  6. Runtimewire. OpenAI says 10,000 AI Agents Solved Navier-Stokes in 88 Hours. Runtimewire, September 2026.URL . Accessed 9 September 2026.
  7. Interesting Engineering. 10,000 OpenAI Agents Crack 90-year-old Navier-Stokes Mystery in just 88 Hours. Interesting Engineering, September 2026. URL . Accessed 9 September 2026.