OpenAI AI Model Cracks Decades-Old Erdős Math Conjecture

An unreleased OpenAI reasoning model has produced a verified solution to the Paul Erdős unit distance problem, a conjecture in combinatorial geometry that has remained unsolved since 1946. Mathematicians and computer scientists have described the result as the most significant AI achievement in formal mathematics to date.

What Happened

OpenAI announced that one of its experimental AI reasoning models generated a complete proof for the Erdős unit distance conjecture, a problem first posed by Hungarian mathematician Paul Erdős more than 80 years ago. The company said the result demonstrates a meaningful advance in AI reasoning capability beyond pattern recognition or retrieval.

According to reporting by Scientific American, the proof is of a quality that mathematicians say would likely be accepted for publication in a top-tier mathematics journal had it been produced by human researchers. New Scientist described the moment as one that has "stunned" working mathematicians.

The Guardian reported that OpenAI characterized the work as evidence of progress in AI reasoning, stating the model was not simply retrieving or recombining known results but constructing novel mathematical arguments.

Background

The unit distance problem is a question in discrete geometry. Erdős asked, in 1946, how many times a single distance can appear among a set of points in a two-dimensional plane. More precisely, given a set of points, the conjecture concerns the maximum number of pairs of those points that can be exactly one unit apart. Despite its accessible formulation, the problem resisted resolution for decades and is considered a benchmark problem in combinatorics.

Paul Erdős, who died in 1996, is one of the most prolific mathematicians in history, having authored or co-authored more than 1,500 papers. Many of his open problems carry formal cash prizes and are treated as long-term challenges for the mathematics community.

OpenAI, founded in 2015 and headquartered in San Francisco, develops large-scale AI systems including the GPT series of language models and the o-series of reasoning models. The company has not publicly identified which specific model produced the Erdős proof, describing it only as an unreleased reasoning model under active development.

What the Proof Involves

The Indian Express reported that OpenAI shared details of the result with selected mathematicians prior to the public announcement. Those mathematicians confirmed the proof's validity, according to coverage from multiple outlets.

Scientific American noted that the result is the first AI-generated mathematical proof that independent experts believe meets the standard for publication in mathematics' most selective journals. Prior AI contributions to mathematics, including work by DeepMind's AlphaProof system in 2024, addressed competition-level problems but did not reach the threshold of original research-grade proofs on open conjectures of this standing.

The Guardian cited OpenAI as saying the work on the Erdős planar unit distance problem is part of a broader research effort to develop AI systems capable of performing and verifying complex multi-step reasoning tasks.

Scope and Limitations

OpenAI has not released the model that produced the proof, and no timeline for a public release has been announced. The company has not stated whether the model can be applied systematically to other open problems or whether the Erdős result required targeted fine-tuning or scaffolding.

Mathematicians cited across multiple reports indicated they are continuing to review the proof in full, and broader peer scrutiny is ongoing. Independent verification by the wider mathematics community, beyond the initial group of reviewers, has not yet been completed as of the time of publication.

OpenAI has said it plans to publish further details about the model and its methodology, though no specific date has been provided for that disclosure.