On February 20, 2026, OpenAI announced something that doesn't fit neatly into any existing category: a Lean 4 formal proof of global regularity for the 3D incompressible Navier-Stokes equations. That sentence requires unpacking, which is what this article is for.
The Navier-Stokes existence and smoothness problem is one of the seven Millennium Prize Problems. It has resisted attack since the equations were written down in the 19th century. Claude-Louis Navier and George Gabriel Stokes described how viscous fluids move; nobody has proven that those equations always produce smooth, unique solutions in three dimensions. They might blow up. Nobody knows.
OpenAI's release claims to close that gap. It's roughly 200,000 lines of Lean 4 code, machine-verified in under 30 minutes, released publicly on GitHub under an open-source license. It introduces a technique called "energy cascade suppression." And it has not been peer-reviewed.
Here are seven things you need to know about what was released, what it actually proves, and what happens next.
The short version: A formal proof, a paper, and a codebase.
OpenAI released a Lean 4 formal proof of global regularity for the 3D incompressible Navier-Stokes equations with viscosity. The claim is specific: for any smooth initial data with finite energy, a unique smooth solution exists for all time. No blow-up. No loss of regularity. The solution behaves.
The release includes three components:
The proof was downloaded over 10,000 times in its first week. That's not a measure of correctness, but it is a measure of interest. Mathematicians, computer scientists, and curious onlookers all pulled the repository.
What makes this release unusual isn't just the result. It's the form. Mathematical proofs are typically written in natural language, checked by human referees, and published in journals. This proof is written in a programming language that a computer can verify. That changes the verification story in ways that matter.
Key Takeaway: OpenAI released a Lean 4 formal proof, a paper, and a verified certificate for the 3D Navier-Stokes global regularity problem. Everything is public and open-source.
Formal verification is not peer review. It's something else entirely.
Lean 4 is a functional programming language and proof assistant. You write mathematical statements and proofs in it, and the Lean kernel checks every logical step. If the proof compiles, the logic is correct — assuming the definitions and lemmas are correct.
This is different from a traditional proof. A traditional proof is checked by humans who read it, understand it, and decide whether the reasoning holds. That process is slow, error-prone, and subject to disagreement. Formal verification is mechanical. The computer doesn't care about your reputation or your argument's elegance. It checks the steps.
The Lean 4 proof was verified in under 30 minutes on a standard server. The formalization used over 1,000 definitions and 5,000 lemmas from mathlib, the community-maintained library of formalized mathematics. That's a substantial dependency base, and it means the proof rests on a large body of prior formal work.
Here's the catch: Lean 4 verifies that the proof is correct according to the formalized definitions. It does not verify that those definitions faithfully represent the original mathematical problem. If the formalization of "Navier-Stokes equations" or "global regularity" is subtly wrong, the proof could be internally consistent and still not address the intended problem. That's why human review of the formalization itself matters.
Key Takeaway: Lean 4 provides machine-checked correctness of the logical steps. It does not verify that the formalization matches the intended mathematics. Both checks are necessary.
This is the part mathematicians will scrutinize most closely.
The proof relies on a novel technique called "energy cascade suppression." In fluid dynamics, energy cascades from large scales to small scales. This cascade is what makes turbulence hard to control and what potentially allows solutions to blow up. If energy concentrates at small scales, the solution can lose smoothness.
Energy cascade suppression is a method for controlling the nonlinear terms in the Navier-Stokes equations. The paper explains how it works, and the Lean 4 formalization encodes the technique. If the technique is sound, it could apply to other nonlinear partial differential equations where similar cascade problems arise.
That "if" is doing a lot of work. The technique is new. It hasn't been independently verified. Mathematicians will need to check whether the suppression argument is valid, whether it generalizes, and whether it introduces hidden assumptions. The Lean 4 proof shows the technique is internally consistent. It doesn't show the technique is correct in the sense that matters to mathematicians.
This is where the real intellectual work will happen. The formalization is a tool. The mathematics is the substance.
Key Takeaway: Energy cascade suppression is the novel mathematical idea behind the proof. It could have applications beyond Navier-Stokes, but it requires independent scrutiny.
This was not a fully automated process.
OpenAI's internal model generated the proof. Then it was translated into Lean 4. That translation involved mathematicians from several universities who collaborated with OpenAI on the formalization. The result demonstrates the potential of AI in generating complex mathematical proofs, but it also highlights the limits.
The model didn't write Lean 4 code directly. It produced mathematical arguments that humans then formalized. That's a meaningful distinction. The AI contributed the mathematical insight — or at least a candidate insight — and humans did the work of making it machine-checkable.
This pattern is likely to repeat. AI generates candidate proofs or proof sketches. Humans formalize them, verify them, and assess their significance. The collaboration is real, and it's not symmetric. The AI is a tool for generating ideas. The humans are responsible for validation.
OpenAI's release doesn't claim the model solved the problem autonomously. It claims the model generated a proof that, after human formalization, Lean 4 verified. That's a significant achievement. It's also a reminder that "AI solved a Millennium Prize Problem" is a headline, not a description.
Key Takeaway: The proof was generated by an OpenAI model and formalized with human assistance. The collaboration was essential, not incidental.
Global regularity is not the same as winning the Millennium Prize.
The proof claims to establish global regularity for the 3D incompressible Navier-Stokes equations. That means: given smooth initial data with finite energy, a unique smooth solution exists for all time. This is the positive resolution of the existence and smoothness problem.
What it does not do is constitute a complete solution to the Millennium Prize Problem as defined by the Clay Institute. The Clay Institute requires publication in a refereed journal and acceptance by the mathematical community. Neither has happened. The proof is a preprint and a formalization, released for community scrutiny.
There's also a subtlety about what "global regularity" means in the formalization. The Lean 4 proof encodes specific definitions. If those definitions match the standard mathematical definitions, the proof addresses the standard problem. If they diverge, it addresses a related but distinct problem. This is not a trivial concern. Formalization requires precision, and precision can reveal ambiguities that natural-language mathematics glosses over.
The proof is a major step. It is not the final word.
Key Takeaway: The proof claims global regularity for 3D incompressible Navier-Stokes. It does not yet satisfy the Millennium Prize requirements, which include peer review and journal publication.
Machine verification is powerful. It's also limited.
Lean 4 verified that the proof is logically correct according to the formalized definitions and lemmas. It includes a verified certificate that can be checked independently. If you have the Lean toolchain, you can run the proof yourself and confirm the verification. That's a level of confidence that traditional proofs rarely achieve.
But Lean 4 did not verify that the formalization faithfully represents the original mathematical problem. It did not verify that the definitions of "Navier-Stokes equations," "global regularity," or "smooth solution" match the standard mathematical usage. It did not verify the novelty or significance of the result. It did not verify that the proof addresses the problem the Clay Institute posed.
These are not minor caveats. They're the core of what the mathematical community will need to assess. The formalization is a translation. Translations can be faithful or unfaithful. The Lean 4 proof shows the translated argument is valid. It doesn't show the translation is correct.
This is why peer review still matters. Mathematicians will read the paper, examine the formalization, and decide whether the definitions and lemmas capture the intended problem. That process will take time. It may uncover issues. It may require clarifications. It may ultimately confirm the result. But it hasn't happened yet.
Key Takeaway: Lean 4 verified the logical correctness of the proof relative to the formalized definitions. It did not verify that the formalization matches the intended mathematical problem.
The mathematical community has already started digging in.
Discussions began on February 22, 2026, two days after the release. Mathematicians are reading the paper, examining the Lean 4 code, and assessing the energy cascade suppression technique. Full peer review will take time — months, possibly years. It may uncover issues or require clarifications. It may lead to revisions. It may confirm the result.
If the proof is accepted, it would be a major breakthrough and a strong candidate for the Millennium Prize. The Clay Institute would need to evaluate it according to its criteria. That process is separate from peer review and has its own requirements.
Beyond the prize, the techniques could inspire new approaches in fluid dynamics and PDE theory. Energy cascade suppression, if sound, could apply to other nonlinear equations. The formalization methods could become standard practice. The collaboration model — AI generates, humans formalize, Lean verifies — could spread.
The release also raises questions about the future of AI in mathematics. If AI can generate proofs of this complexity, what role do mathematicians play? The answer, for now, is: the same role they've always played, plus new ones. Mathematicians assess significance, verify translations, and decide what matters. AI generates candidates. That division of labor may shift, but it hasn't shifted yet.
Key Takeaway: Peer review is underway. If accepted, the proof could win the Millennium Prize and reshape both fluid dynamics and the practice of formal mathematics.
OpenAI's release is a landmark moment for AI and formal mathematics. The Lean 4 proof provides a high level of confidence in the logical correctness of the argument. It showcases the potential of AI-human collaboration in solving complex problems. It has already advanced the conversation.
But it is not the finish line. The mathematical community must still assess the correctness of the formalization, the soundness of the energy cascade suppression technique, and the significance of the result. The world will be watching as mathematicians dig into 200,000 lines of code and a new mathematical idea.
Whether or not it wins the Millennium Prize, it has already changed something. The question is what.
What did OpenAI release regarding Navier-Stokes? A Lean 4 formal proof of global regularity for the 3D incompressible Navier-Stokes equations, along with a paper and a verified certificate. The proof is approximately 200,000 lines of code and was released on GitHub under an open-source license.
What is Lean 4? Lean 4 is a functional programming language and proof assistant. It allows mathematicians to write proofs in a formal language that a computer can verify. The Lean kernel checks every logical step.
Has the proof been peer-reviewed? No. The proof is a preprint and formalization released for community scrutiny. Peer review is ongoing and will take time.
Is the proof open source? Yes. It is available on GitHub under an open-source license and can be downloaded and checked by anyone with the Lean toolchain.
What does the proof claim to show? It claims to establish global regularity for the 3D incompressible Navier-Stokes equations: for any smooth initial data with finite energy, a unique smooth solution exists for all time.
How was the proof generated? An internal OpenAI model generated the proof, which was then translated into Lean 4 with assistance from mathematicians at several universities.
What is the significance of this result? If correct, it resolves a 200-year-old problem and would be a strong candidate for the Millennium Prize. It also demonstrates the potential of AI in generating complex mathematical proofs.
Can I verify the proof myself? Yes. You can run the Lean 4 proof assistant on the provided code. The proof includes a verified certificate that can be checked independently.
What are the next steps? Peer review, assessment of the formalization, and evaluation of the energy cascade suppression technique. If accepted, the Clay Institute would evaluate it for the Millennium Prize.
Did OpenAI win the Millennium Prize? No. The Millennium Prize requires publication in a refereed journal and acceptance by the mathematical community. Neither has happened yet.
Dive into the Lean 4 proof yourself on GitHub, explore the accompanying paper, and join the mathematical community in scrutinizing this result. The future of AI in mathematics is being written right now — don't just watch, verify.