OpenAI Reasoning Model Disproves 80 Year Old Erdős Math Conjecture

OpenAIOpenAI

· Updated

An internal OpenAI reasoning model autonomously solved the planar unit distance problem, disproving a 1946 conjecture by Paul Erdős. Researchers Alex Wei, Hongxun Wu, and @wjmzbmr1 detailed the breakthrough and the future of human-AI collaboration on the OpenAI Podcast.

OpenAI researchers Alex Wei, Hongxun Wu, and @wjmzbmr1 shared how an internal reasoning model (AI trained to perform explicit step-by-step reasoning before answering) disproved a central conjecture in discrete geometry posed by Paul Erdős in 1946. The model solved the planar unit distance problem by discovering new constructions.
Problem
Planar unit distance problem
Conjecture origin
Paul Erdős (1946)
New lower bound exponent
1.014 (refined)
Mathematical fields
Discrete geometry, algebraic number theory
Verification
External (Fields Medalist Tim Gowers and others)
Researchers
Alex Wei, Hongxun Wu, and @wjmzbmr1

This milestone validates the shift toward reasoning models that use test-time compute (allocating extra processing power during inference) to solve frontier research problems. Unlike specialized systems, this model autonomously connected concepts from algebraic number theory to resolve a geometric question. It follows the GPT-5.5 feature set.

OpenAI is using these results to test how models can contribute to biology and materials science. This capability echoes the GPT-Rosalind launch, pointing to reasoning models acting as autonomous research partners. The researchers detailed the discovery and human-AI collaboration on the OpenAI Podcast.

View the full update on openai.com
OpenAI
OpenAI
@OpenAI
X

What happened when one of our models found a counterexample to an 80-year-old Erdős conjecture? Researchers @alexwei_, @HongxunWu, and @wjmzbmr1 shared the story on the OpenAI Podcast with @AndrewMayne and explained how mathematicians and models can work together to make new discoveries.

26retweets416likes
View on X

Still wondering? A few quick answers below.

The planar unit distance problem is a famous question in discrete geometry first posed by Paul Erdős in 1946. It asks for the maximum number of pairs of points that can be exactly one unit apart in a set of n points. For decades, mathematicians believed square grid constructions were the most efficient way to maximize these pairs.

The model discovered an entirely new family of point constructions that performs better than the previously accepted square grid model. By applying sophisticated ideas from algebraic number theory, the model proved that the number of unit distance pairs grows faster than the nearly linear rate mathematicians had assumed was the upper limit for eighty years.

No, the breakthrough came from a general-purpose reasoning model rather than a system built specifically for mathematics or this particular geometry problem. OpenAI evaluated the model on a collection of open math problems to test if advanced reasoning systems could contribute to frontier research, and the model autonomously produced the proof without specialized human direction.

Yes, the proof has been checked and verified by a group of prominent external mathematicians, including Fields Medalist Tim Gowers and leading number theorist Arul Shankar. These experts confirmed the validity of the argument and published a companion paper explaining the significance of the result and the sophisticated algebraic number theory tools the model used.

This milestone demonstrates that reasoning models can hold together long, difficult chains of logic and connect ideas across distant fields of knowledge. OpenAI intends to apply these same autonomous reasoning capabilities to accelerate research in biology, physics, and medicine, moving AI from a simple assistant to a partner capable of surfacing original ideas.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

OpenAI

See all AI news & updates from OpenAI →

Share this update