An OpenAI Model Just Solved an 80-Year-Old Math Problem That Stumped Mathematicians

Quick Answer

In May 2026, OpenAI announced that one of its general-purpose reasoning models autonomously disproved a central conjecture in discrete geometry — the Erdős unit distance problem, which had been open since 1946. The proof was verified by external mathematicians including Fields Medalist Tim Gowers, who called it “a milestone in AI mathematics.” This is the first time an AI has independently solved a prominent open problem in mathematics.

Introduction

In 1946, the mathematician Paul Erdős asked a deceptively simple question. If you place n points in a plane, what is the maximum number of pairs that can be exactly one unit apart? The unit distance problem became one of the most famous open questions in combinatorial geometry. Erdős offered a monetary prize for its solution. Generations of mathematicians tried and failed to crack it.

Eighty years later, the answer came from an unexpected source. An AI model, built by OpenAI as a general-purpose reasoning system, found a proof that disproves the longstanding conjecture about this problem. The result has been verified by a panel of external mathematicians who confirmed the proof is correct. Fields Medalist Tim Gowers called it “a milestone in AI mathematics.”

This is not a toy problem. This is a central question in a mathematical subfield that had resisted solution for nearly a century. And an AI solved it autonomously.

What Is the Unit Distance Problem?

The unit distance problem is about how many pairs of points in a plane can be exactly one unit apart. It is easy to state and extremely difficult to solve.

Here is the intuitive version. Take a handful of points. Drop them on a flat surface. Measure the distance between every pair. Count how many pairs are exactly one unit apart. Now what is the maximum possible count for a given number of points?

Erdős showed that the answer grows roughly linearly with the number of points. The best known constructions, based on rescaled square grids, achieved growth slightly faster than linear — but the improvement was vanishingly small for large numbers of points. For decades, mathematicians believed these grid constructions were essentially optimal. No one could prove it, but no one could improve on them either.

The problem was listed as one of the most important open questions in the field. The 2005 book “Research Problems in Discrete Geometry” called it “possibly the best known (and simplest to explain) problem in combinatorial geometry.” Noga Alon, a leading combinatorialist at Princeton, described it as “one of Erdős’ favorite problems.”

How the AI Solved It

What makes this result remarkable is not just that an AI solved the problem, but how it solved it. The model was not specifically trained for mathematics. It was not scaffolded to search through proof strategies. It was not even targeted at the unit distance problem in particular.

OpenAI tested a new general-purpose reasoning model on a collection of Erdős problems as part of a broader evaluation. The model produced a proof for the unit distance problem that brought unexpected ideas from algebraic number theory to bear on an elementary geometric question. It found an infinite family of constructions that disproved the prevailing conjecture, providing a polynomial improvement over the square grid approach that had been considered optimal for decades.

The proof was then checked by external mathematicians, who confirmed its correctness. They also wrote a companion paper explaining the argument and providing context for the significance of the result.

According to leading number theorist Arul Shankar, “In my opinion this paper demonstrates that current AI models go beyond just helpers to human mathematicians — they are capable of having original ingenious ideas, and then carrying them out to fruition.”

What This Means for AI and Mathematics

This is the first time an AI has autonomously solved a prominent open problem that was central to a subfield of mathematics. Previous AI math achievements have been impressive but limited. AlphaGo found novel strategies in Go. AlphaFold solved protein folding. But those systems were trained specifically for their domains. This OpenAI model is a general-purpose reasoning system that happened to solve a problem that had stumped humans for eighty years.

The implications are significant. Mathematics has always been considered a uniquely human endeavor — the purest form of reasoning, resistant to automation. If general-purpose AI can contribute original proofs to open problems, the role of mathematicians may shift from being the primary discoverers to being collaborators who guide, verify, and interpret AI-generated mathematics.

OpenAI’s approach was notable. The model produced a proof, but human mathematicians reviewed and validated it. This human-in-the-loop verification model is likely how AI-assisted mathematics will work for the foreseeable future. AI generates candidate proofs. Humans check them. The best results emerge from the collaboration.

The Wider Context: OpenAI’s Spring 2026 Momentum

The unit distance breakthrough is part of a broader wave of OpenAI product and research announcements in Spring 2026. The company also released a personal finance experience in ChatGPT that connects to 12,000+ financial institutions via Plaid, powered by GPT-5.5 reasoning. It published the Frontier Governance Framework aligning with emerging AI regulations. It demonstrated Codex-powered agents that self-improve on tax preparation, processing 7,000 returns with 97% accuracy. It partnered with Dell to bring Codex to hybrid and on-premises enterprise environments.

Each of these announcements tells a different story. The finance feature shows practical consumer application. The Codex agents show enterprise automation. The math proof shows frontier reasoning capability. Together, they paint a picture of a company pushing on every front simultaneously.

What Critics and Skeptics Are Saying

The result is not without debate. Some mathematicians point out that the proof was checked by humans, not independently verified by another AI. The companion paper by external mathematicians provides confidence, but the field will need time to fully absorb and validate the result.

Others note that the model was tested on a collection of Erdős problems, and this was the one it solved. The success rate is not 100%. But even one solved problem from a general-purpose model represents a milestone. The bar for AI has moved from “can it do math” to “can it do math that humans cannot.”

What Comes Next

The proof is publicly available, along with the model’s chain of thought and the companion paper by external mathematicians. This transparency is important — it allows the mathematical community to study the proof, learn from the AI’s approach, and apply similar techniques to other problems.

For AI researchers, the result validates a direction: general reasoning models can contribute to frontier science without domain-specific training. The next step is scaling. If a model can solve one Erdős problem, how many more can it solve with more compute, better training, or more sophisticated search strategies?

For mathematicians, the message is clear. AI is becoming a collaborator, not just a calculator. The relationship between human and machine intelligence in mathematics is about to get much more interesting.

Frequently Asked Questions

Q: Did an AI really solve an 80-year-old math problem?

A: Yes. An OpenAI general-purpose reasoning model disproved the central conjecture of the Erdős unit distance problem, which had been open since 1946. The proof was verified by external mathematicians including Fields Medalist Tim Gowers.

Q: Was the AI trained specifically for mathematics?

A: No. The model was a new general-purpose reasoning system, not a math-specialized system. It was tested on a collection of Erdős problems and produced a valid proof for this one.

Q: Is this the first time AI has solved an open math problem?

A: Yes, this is the first time a prominent open problem central to a subfield of mathematics has been solved autonomously by AI.

Q: Where can I see the proof?

A: OpenAI published the full proof, the companion paper by external mathematicians, and an abridged chain of thought from the model. Links are available in OpenAI’s blog post about the result.

Q: Does this mean AI will replace mathematicians?

A: No. It means AI will become a collaborator in mathematical research. Human mathematicians will guide, verify, and interpret AI-generated work. The best results will come from human-AI collaboration.

Conclusion

The unit distance problem was a wall that mathematicians had been running into for eighty years. An AI could not see the wall because it had no preconceptions about what was possible. It brought tools from algebraic number theory that no one had thought to apply. It constructed a proof that disproved the consensus view.

This is not science fiction. This is May 2026. And it will not be the last time an AI solves a problem that humans could not.