- In May 2026, OpenAI announced that one of its reasoning models had disproved a decades-old conjecture in discrete geometry connected to the famous unit distance problem first posed by mathematician Paul Erdős in 1946.
- The AI-generated proof introduced a new infinite family of geometric constructions that went beyond previously accepted bounds.
- Leading mathematicians like Tim Gowers and Noga Alon called the result potentially historic, though they said that the proof needs to survive long-term peer review and independent verification.
Artificial intelligence has jumped into one of academia’s toughest areas with a claim that could change how people think about mathematical discovery. OpenAI recently announced that one of its reasoning-focused AI models produced a proof that seems to disprove a long-standing conjecture in discrete geometry which is a branch of math about arranging points, lines, and shapes.
This centers on the unit distance problem, originally posed by Hungarian mathematician Paul Erdős in 1946. For nearly 80 years, mathematicians tried to figure out how many pairs of points can stand exactly one unit apart when placing ‘n’ points on a flat plane. Researchers slowly improved their estimates, but most thought the main constructions were already very close to the best answer.
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.
— OpenAI (@OpenAI) May 20, 2026
For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.
An OpenAI model has now disproved that… pic.twitter.com/j2g3Ze0zEG
The Geometry Problem That Confused Mathematicians for Nearly 80 Years
At first, the unit distance problem sounds too easy. Imagine dropping dots on a sheet of paper. The idea is to arrange them so the biggest number of pairs sit exactly one unit apart.
Here’s how the puzzle works:
- Pick some number of points: ‘n’.
- Arrange them on a flat surface.
- Count how many pairs of points are exactly a distance of 1 apart.
- Find the arrangement that gives you the highest possible number.
As ‘n’ grows, the difficulty increases. Even tiny changes in arrangement can shift the total count, so the whole thing becomes a tough geometric optimization challenge.
For decades, mathematicians used lattice-based constructions, mainly square grids and triangular grids. These patterns helped set upper and lower bounds for the problem and were considered some of the best-known approaches But no one could point to the truly optimal arrangement.
OpenAI claims that its reasoning model came up with a totally new family of geometric constructions that beat previous approaches by a polynomial factor. In practical terms, the AI found new ways to arrange points that produced noticeably more unit-distance pairs than experts thought was possible.
Instead of sticking to traditional geometric reasoning, the AI used ideas from algebraic number theory which is an area of math more about equations and numerical structure than geometric arrangements. This crossover seemed extremely unexpected since the two fields aren’t usually mixed for the unit distance problem.
The model tested several configurations, uncovered hidden mathematical relationships, and generated a proof explaining why the construction works
Unlike earlier AI systems built mainly for calculation or symbolic theorem-proving, OpenAI framed this model as a general reasoning engine that can invent original mathematical strategies.
Why Researchers Believe This Could Change the Future of AI Research
The impact of the announcement goes way beyond geometry. Artificial intelligence has already changed businesses like software engineering, image generation, and language processing. But advanced math has always been one of the hardest tests for real reasoning since proofs demand strict logical consistency and not just pattern matching.
That’s why mathematicians are both cautious and serious. Fields Medalist Tim Gowers called the development a potentially huge milestone, as long as the proof survives detailed observation. Noga Alon, a top expert in combinatorics and discrete math, also recognized the significance.
Their comments show that experts are starting to see AI systems as more than just research assistants. But, mathematicians plan to check the OpenAI proof line by line before anyone declares the conjecture dead.
Also read: OpenAI’s Codex Can Now Control Your Entire Desktop With “Computer Use”
Conclusion
Regardless of whether the proof survives every layer of examination, OpenAI’s announcement already marks a turning point for artificial intelligence and scientific discovery. The idea that a machine found a new mathematical construction for a problem that has puzzled researchers since 1946 changes how people see AI’s abilities. If the proof holds up, it could become one of the first big examples of AI making an original leap in pure mathematics, not just helping humans out.
