OpenAI Claims 3 Major Scientific Breakthroughs In Math, Physics, And Computer Science

OpenAI has announced three new scientific discoveries reportedly produced by its internal AI models, spanning mathematics, computer science, and physics. The findings could challenge longstanding scientific assumptions, with independent researchers now being brought in to examine and explain the results.
 

TL;DR

 

• OpenAI claims its AI models produced three significant scientific results across mathematics, computer science, and physics.
• The company says its models resolved major mathematical conjectures, disproved the Exponential Time Hypothesis, and produced rigorous proofs involving quantum field theory.
• OpenAI is working with external academics to independently examine the findings, with some AI-generated materials already publicly available.

OpenAI published "Working with academic communities to build scientific understanding," detailing three research results involving harmonic analysis, complexity theory, and quantum field theory.

Unlike the usual AI announcements centered on new models, features, or improvements in reasoning capabilities, OpenAI's latest claims concern original scientific research into problems that have challenged experts for decades.

According to the company, its models have generated mathematical arguments that could establish new findings across multiple fields. However, OpenAI is treating the resulting manuscripts as research material that requires further scientific examination.

Source

"In our view, these manuscripts are best understood as a form of raw research data," OpenAI said in its announcement.

The company is also experimenting with different approaches to releasing its findings, allowing external academic communities to investigate the claims and develop their own explanations.

So, what exactly has OpenAI claimed to achieve?

OpenAI Claims Its AI Models Solved Major Mathematical Conjectures

The first reported breakthrough concerns the local smoothing conjecture, a longstanding problem in harmonic analysis originally proposed by Christopher Sogge in 1991.

The conjecture addresses how waves behave when they become concentrated, particularly the idea that extreme focusing should persist only for short periods.

In 2019, mathematicians Larry Guth, Hong Wang, and Ruixiang Zhang established the conjecture in two spatial dimensions. More recently, in 2025, Hong Wang and Joshua Zahl proved the three-dimensional Kakeya conjecture.

Now, OpenAI claims its models have produced a proof of the local smoothing conjecture in every dimension.

According to the company, the result would also resolve the Kakeya conjecture and Stein's restriction conjecture across all dimensions, potentially extending mathematical understanding beyond previously established results.

OpenAI said its models generated the mathematical argument on October 18, 2026, before the proof was formalized in Lean on October 19.

Lean is a formal proof system that allows mathematical statements and logical reasoning to be checked by software.

The company has published both the AI-generated natural-language explanation and its Lean formalization through its public GitHub repository.

OpenAI Says Its AI Disproved The Exponential Time Hypothesis

The second reported breakthrough could have significant implications for theoretical computer science.

OpenAI claims its AI system disproved the Exponential Time Hypothesis (ETH), a longstanding assumption about the computational difficulty of solving certain Boolean satisfiability problems.

ETH has played an important role in complexity theory, with researchers using it to establish conditional limits on the efficiency of algorithms.

The hypothesis also relates to broader questions in computer science, including the famous P versus NP problem, although disproving ETH would not automatically settle P versus NP.

According to OpenAI, its system generated a natural-language disproof on October 16, 2026, which was subsequently formalized in Lean on October 17.

The company said the result could encourage researchers to revisit computational hardness assumptions used across graph theory, computational biology, databases, and computational geometry.

However, the practical implications remain uncertain. Even if the proof withstands independent scrutiny, developing faster algorithms for real-world applications would require further research.

To investigate the findings, OpenAI is collaborating with a group led by Russell Impagliazzo and Ramamohan Paturi of the University of California San Diego, alongside Ryan Williams of the Massachusetts Institute of Technology (MIT).

Interestingly, Impagliazzo and Paturi were among the original researchers who introduced the Exponential Time Hypothesis, making their involvement particularly relevant to the evaluation.

OpenAI has agreed to make the underlying research materials publicly available and support a conference dedicated to examining the result and its wider implications.

AI Models Also Produce Physics Proofs Involving Quantum Field Theory

The third development concerns theoretical physics, where OpenAI claims its models have produced mathematically rigorous existence proofs involving certain interacting quantum field theories in four dimensions.

These include quantum chromodynamics (QCD), which describes the strong force responsible for interactions between quarks and gluons, and a unified completion of the Standard Model of particle physics.

The research also includes an existence proof for pure Yang–Mills theory, which is connected to one of the Clay Mathematics Institute's Millennium Prize Problems.

OpenAI said its models first generated a natural-language proof concerning Yang–Mills existence on September 9, 2026, followed by a proof involving QCD on September 18.

The company completed a Lean formalization of the Yang–Mills existence result on October 16.

However, the physics research has not been made publicly available in the same manner as the mathematics and computer science results.

Instead, an independent academic group led by Nobel Prize-winning physicist David Gross will examine the material and prepare its own explanation before determining how the work should be released.

The group also includes Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov.

Importantly, the reported Yang–Mills existence proof should not be interpreted as confirmation that the complete Millennium Prize problem has been solved. That problem also includes requirements concerning the existence of a mass gap.

OpenAI Turns To Academic Researchers For Independent Verification

As reported by Unite.AI, OpenAI is adopting three different approaches to sharing the research rather than releasing all the findings under a single process.

The mathematics result is publicly available for researchers to investigate independently, while the computer science result is being examined through an external collaboration with publicly accessible proof materials.

Meanwhile, the physics group has decided to delay the release of AI-generated materials while its members prepare their own account of the findings.

 

OpenAI said these collaborations will follow four principles: transparency, no academic credit claimed by OpenAI for model-generated discoveries, recognition for researchers who develop scientific understanding, and independence for participating academics.

Participation will be voluntary and unpaid, with researchers free to publish their conclusions without representing OpenAI.

As highlighted by YourStory, the findings still face academic scrutiny before they can be treated as established scientific breakthroughs.

A computer-checked proof can strengthen confidence in a mathematical argument, but researchers must also determine whether its formal statements capture the intended scientific claims, assess their significance, and examine any broader implications.

OpenAI has also highlighted its previously announced $5 million contribution to the Simons Foundation, which will support academic workshops, conferences, and programs exploring AI-generated research results.

The company believes collaboration between AI systems and human researchers could help accelerate the discovery and understanding of difficult scientific problems.

For now, the three results represent substantial scientific claims supported by different levels of publicly available evidence and academic examination. Their wider significance will depend on how the research community evaluates the proofs and builds on the findings.