Autonomous Disproof of the Erdős Unit Distance Conjecture by a GPT-5.5 Pro Agent

OpenAI's recent disproof of the Erdős unit distance conjecture represents a milestone in using artificial intelligence to solve important open mathematical problems. However, this achievement relied on an internal model with undisclosed computational costs, limiting independent reproduction and analysis by the broader research community. In this paper, we present a simple autonomous agent built upon the publicly available GPT-5.5 Pro. Using a problem-agnostic, three-stage prompting pipeline—brainstorm, execute, and review—the agent consistently disproved the conjecture, generating a correct proof in 8 out of 8 independent trials. These 8 proofs differ from existing AI-generated proofs. The system operates efficiently, consuming an average of approximately 107k reasoning tokens per trial. We open-source our code and outputs, providing the first publicly accessible AI agent to autonomously resolve this conjecture.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-06-24
DOI
https://doi.org/10.5281/zenodo.20822771
Primary Topic
Polynomial and algebraic computation
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
preprint

Autonomous Disproof of the Erdős Unit Distance Conjecture by a GPT-5.5 Pro Agent

Yichen Huang
Zenodo (CERN European Organization for Nuclear Research)
Polynomial and algebraic computation
preprint

Autonomous Disproof of the Erdős Unit Distance Conjecture by a GPT-5.5 Pro Agent

Yichen Huang
preprint en

Abstract

OpenAI's recent disproof of the Erdős unit distance conjecture represents a milestone in using artificial intelligence to solve important open mathematical problems. However, this achievement relied on an internal model with undisclosed computational costs, limiting independent reproduction and analysis by the broader research community. In this paper, we present a simple autonomous agent built upon the publicly available GPT-5.5 Pro. Using a problem-agnostic, three-stage prompting pipeline—brainstorm, execute, and review—the agent consistently disproved the conjecture, generating a correct proof in 8 out of 8 independent trials. These 8 proofs differ from existing AI-generated proofs. The system operates efficiently, consuming an average of approximately 107k reasoning tokens per trial. We open-source our code and outputs, providing the first publicly accessible AI agent to autonomously resolve this conjecture.

Zenodo (CERN European Organization for Nuclear Research)
Polynomial and algebraic computation
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.