OpenAI Claims Blowup — $1M Problem Rocked

Kyiv,,Ukraine,-,March,17,,2024,Openai,Logo,On,Iphone — OPENAI BOMBSHELL

OpenAI says an internal AI proved that a three-dimensional fluid can “blow up” in finite time, striking the heart of a 90-year puzzle.

Story Snapshot

OpenAI announced a solution to the Navier–Stokes existence and smoothness problem, showing finite-time singularity.
An unreleased model used about 10,000 AI agents and ran for roughly 88 hours to produce the proof.
OpenAI released a 165-page write-up and a Lean formalization package for machine checking.
The claim focuses on a forced, three-dimensional incompressible fluid case with singularity formation.

What OpenAI Claims The AI Proved

OpenAI stated its in-house system produced a proof resolving the Navier–Stokes existence and smoothness problem by showing a fluid governed by these equations can reach a singularity in finite time. That means velocity or its derivatives become unbounded in a real, finite time window.

The company framed this as establishing the “finite-time blowup” side of the problem for a forced three-dimensional incompressible flow. OpenAI published a write-up and Lean files to present the argument and its structure.

OpenAI said the result came from a next-generation model that coordinated about 10,000 agents over roughly 88 hours. The company described the model as more capable than its prior named systems.

Outlets summarized the effort as a large-scale, parallel search guided by proof tactics, producing both an analytical manuscript and corresponding machine-checkable artifacts.

The reported package includes the core Navier–Stokes claim along with a related Euler equation blowup result, organized for independent review.

Why A Singularity Matters For The Big Question

Mathematicians and physicists use the Navier–Stokes equations to model weather, ocean waves, airflow over planes, and blood in arteries.

The central Millennium Prize question asks whether smooth solutions always exist for all time in three dimensions, or whether they can break. Showing a finite-time singularity answers that by pointing to a breakdown.

OpenAI’s description ties the breakdown to a physically reasonable, smooth start and forcing, not a contrived corner case, which makes the claim more striking if it holds.

Lean, the proof assistant named in the release, matters because it checks every logical step. Formal proof tools remove guesswork and help catch hidden gaps.

In recent years, researchers have treated formal verification as a sanity filter for ambitious claims: if a proof compiles in Lean, the core logic matches the stated theorem. OpenAI’s choice to release Lean artifacts aligns with that best practice and signals confidence in the proof’s internal consistency.

One of the trickiest problems in mathematics has now fallen to AI after just days of work. The groundbreaking result was announced amid rumour after similar, but less complete work, also created with AI, was announced just hours before https://t.co/I1fNSBLjTe

— New Scientist (@newscientist) September 8, 2026

How The AI Workforce Tackled The Problem

OpenAI described a swarm of agents splitting tasks, exploring proof paths, generating lemmas, and stitching arguments into a single, coherent line.

That approach echoes current work in neural theorem proving, where a language model proposes steps and a symbolic engine checks them.

Teams iterate: try subgoals, learn from failures, refine search, and grow a library of tools. The loop continues until a stable proof emerges and can be compiled by a trusted checker.

OpenAI: "we solved Navier-Stokes with 10,000 AI agents."

NYU mathematician Tristan Buckmaster: "interesting. I was working on exactly that using your Codex tool."

OpenAI: 🙂

the closed-model advantage is that nobody can audit what they borrowed.

— zeroemployees (@zeroemployees) September 9, 2026

Earlier efforts showed promise on contest problems, but this claim targets a marquee equation. The move from competition-level theorems to a Millennium Prize-class result marks a jump in scope.

The reported addition of a full Lean formalization suggests the system did more than write persuasive prose. It built a proof with a backbone strong enough for a machine to verify, which is where modern math and modern AI meet in a productive handshake.

What This Could Mean Next

A finite-time blowup claim for a forced, three-dimensional incompressible fluid would reshape the roadmap for analysis and simulation. Engineers could better flag where models fail. Mathematicians could aim at sharper thresholds and stability zones.

If this approach generalizes, similar agent swarms may attack other deep problems, from turbulence structure to partial differential equation regularity. The math community has learned that lasting acceptance comes through careful checking and clear mapping to the official statements.