OpenAI AI models solve three major scientific problems in 2026

A trifecta of scientific breakthroughs

OpenAI recently announced that its AI models successfully solved three major scientific problems across different disciplines. These results include a Millennium Prize problem in physics, a foundational hypothesis in computer science, and a 35-year-old conjecture in mathematics. Each discovery addresses a deep foundational issue that human experts had struggled to resolve for decades.

What distinguishes these results is that they were all formally verified in Lean. Lean is a proof assistant, which is a type of software that checks every single logical step of a mathematical argument by a machine. Because Lean verifies the logic, the proofs are considered valid without the need for traditional peer review to catch human errors.

The announcements represent a concentrated burst of AI-generated scientific discovery. While humans previously used AI as a tool to speed up work, these results show the technology operating independently at the frontier of science. The AI produced the natural-language proofs and the Lean code for the computer science and mathematics results, while the physics result remains under expert review.

Rewriting computer science logic

In the field of computer science, OpenAI's model disproved the Exponential Time Hypothesis, or ETH. For years, ETH served as a load-bearing pillar for the field, leading researchers to assume certain problems were fundamentally too slow to solve efficiently. This assumption influenced hundreds of papers in areas like graph theory, computational biology, and databases.

By proving ETH is false, the AI has effectively rewritten the bottom layer of theoretical computer science logic. Many problems that were previously considered computationally impossible may now be solvable. This could lead to the development of core algorithms that run significantly faster than anyone previously believed was possible.

The verification of this disproof was led by researchers from UC San Diego and MIT, who were the original proposers of the hypothesis. OpenAI has shared the natural-language disproof and the Lean code on GitHub. To facilitate discussion, the company also pledged to pay for a global meeting where the community can review the findings.

Solving a Millennium Prize problem

In physics, the AI produced a rigorous mathematical proof for the existence of interacting four-dimensional quantum field theories. This work addresses a Millennium Prize Problem, which is one of seven major mathematical challenges carrying a $1 million reward. Specifically, it proves the existence of the mathematical framework underlying the strong nuclear force, known as Yang-Mills theory.

The proof covers quantum chromodynamics and provides a completion of the Standard Model of particle physics. This is significant because quantum field theory is the bedrock of modern particle physics but has historically lacked a rigorous mathematical foundation. The AI generated the specific Yang-Mills existence proof on September 9, 2026, and the broader QCD proof on September 18.

Due to the extraordinary complexity of the argument, the physics team has chosen a different disclosure strategy. They will not release the raw AI-generated files publicly. Instead, a panel of experts, including a Nobel Prize winner, will write interpretive papers to explain the results to the scientific community.

Cracking the local smoothing conjecture

In the realm of harmonic analysis, the AI solved the local smoothing conjecture, which was first proposed in 1991. This conjecture describes how waves behave when they focus, suggesting that extreme focusing must be brief. Solving this problem was a major goal because it directly implies two other famous open problems: the Kakeya conjecture and the Stein restriction conjecture.

While humans had made some progress on these problems in recent years, the AI found a new auxiliary partial differential equation to solve the conjecture in all dimensions at once. This effectively resolved the Kakeya and Stein restriction conjectures completely. The AI produced the natural-language proof on October 18 and completed the Lean verification the following day.

OpenAI has taken a fully open approach to this specific mathematical result. The company made both the proof document and the Lean code available on GitHub for mathematicians to study. This reflects the culture of mathematics, where proofs are intended to be read, challenged, and understood by the broader community freely.

New rules for AI discovery

OpenAI is using different disclosure strategies for each result based on the nature of the field. For computer science and math, they are providing full transparency with public code. For physics, they are using expert mediation because the results require deep domain interpretation that goes beyond what a machine can verify. This shows a nuanced approach to how AI-generated science is shared with the world.

The company outlined four principles for collaboration on these results: transparency, no authorship claims, respect for scholarly contribution, and independence. OpenAI will not claim academic credit for the discoveries, and the funds for workshops and research grants are being managed by the Simons Foundation. This ensures that the human scholars who write the interpretive papers retain full ownership of their work.

The ultimate takeaway is that AI is now producing results that humans must work to understand. While Lean verifies that the logic is valid, it does not determine the significance of the discovery. The scientific community now faces the task of digesting these new truths and determining how they will reshape the future of their respective fields.