OpenAI Says Its AI Disproved an 80‑Year‑Old Erdős Geometry Conjecture

Despite its straightforward wording, the question became one of the most studied problems in combinatorial and discrete geometry. For decades, mathematicians investigated possible constructions of points that maximize the number of unit‑distance pairs, as well as upper bounds that limit how large that number can be.

A widely held belief was that the most efficient arrangements of points would resemble square‑grid‑like patterns, similar to points placed on a lattice.

According to reports about OpenAI’s new work, the AI-generated proof challenges this assumption by showing that the conjecture underlying that belief is false.

What OpenAI Claims Its Model Did

OpenAI says a general‑purpose reasoning model produced an original proof that disproves a central conjecture connected to the Erdős problem.

Key elements of the claim include:

• The model generated a new mathematical argument, rather than reproducing known results.
• The proof reportedly uses techniques involving algebraic number theory to construct configurations that outperform traditional square‑grid approaches.
• The discovery contradicts long‑standing expectations about how optimal point arrangements should look.

If correct, the result changes the theoretical picture around one of discrete geometry’s best‑known open questions.

How This Differs From the Earlier GPT‑5 Claim

The announcement comes after a previous controversy involving GPT‑5.

In that earlier case, OpenAI representatives said the model had solved several Erdős problems. Later analysis showed that the model had rediscovered solutions already present in the mathematical literature, meaning it had not produced genuinely new results.

The new claim is different in two important ways:

• The proof is described as original, not a rediscovery of known work.
• External mathematicians reportedly reviewed the argument and expressed support for its validity.

Because mathematics requires rigorous verification, the ultimate status of the result depends on detailed peer review and publication.

What Mathematicians Have Said So Far

Reports indicate that several prominent mathematicians examined the proof and offered supportive comments. Those cited include researchers such as Noga Alon, Melanie Wood, and Thomas Bloom, who have expertise in combinatorics and number theory.

Observers have suggested the result is unusually strong for an AI‑generated proof. Some commentary described it as far beyond previous AI attempts at producing novel mathematical results.

At the same time, the broader mathematical community typically requires extensive verification before accepting a proof of this scale. Full validation usually involves detailed scrutiny and, eventually, publication in a peer‑reviewed journal.

Why Researchers Say It Matters

Beyond the specific geometry result, researchers see a broader implication: AI may be starting to handle long chains of reasoning required for advanced research.

Many difficult problems in science require hundreds or thousands of logical steps connecting ideas across fields. If AI systems can reliably construct and verify such chains, they could contribute to discovery in areas such as:

• theoretical physics
• biology and medicine
• engineering and materials science

Some researchers say the development hints that AI may evolve from a tool that assists scientists to one that occasionally generates new theoretical insights itself.

The Bottom Line

OpenAI’s claim is that a reasoning model produced a genuinely new proof disproving a conjecture tied to the Erdős planar unit distance problem—something mathematicians have studied since 1946.

That makes the announcement fundamentally different from earlier AI math claims that merely reproduced existing results.

However, in mathematics the final verdict comes only after detailed peer review. If the proof holds up under that scrutiny, it could mark one of the first major examples of AI contributing an original solution to a long‑standing open problem in pure mathematics.