AI Models Conquer Legendary Erdős Math Problems in Breakthrough Year
On May 20, 2026, OpenAI announces that an internal AI model finds a counterexample to the "unit distance" problem, a conjecture posed by legendary Hungarian mathematician Paul Erdős in 1946. The result marks the first historically significant mathematical proof to emerge from an AI model. Though human mathematicians quickly refine the work, the AI introduces ideas from a distant branch of mathematics that no one has previously applied to the problem.
Within days, related techniques derived from the AI's approach help solve other important mathematical problems. Then on August 1, OpenAI reveals that an unreleased model named Astra makes 10 additional mathematical advances, including solutions to three more Erdős problems. Many mathematicians view these breakthroughs as a dramatic turning point in how mathematical research is conducted.
Erdős, one of history's most prolific mathematicians, poses thousands of questions over his career that are known for being both simple to state and mathematically deep. Princeton University's Noga Alon, who solves dozens of Erdős problems over his decades-long career, says these AI models are fundamentally changing mathematical research. Mathematicians are now examining what makes Erdős problems so amenable to AI in hopes of understanding how the technology might transform the broader landscape of mathematics.