OpenAI's Astra Solves 3 More Erdős Problems in 10 New Math Advances
Updated
Updated · Quanta Magazine · Aug 3
OpenAI's Astra Solves 3 More Erdős Problems in 10 New Math Advances
3 articles · Updated · Quanta Magazine · Aug 3
Summary
OpenAI said on Aug. 1 that its unreleased Astra model produced 10 additional mathematical advances, including solutions to three more problems posed by Paul Erdős.
Those results build on OpenAI's May 20 announcement that an internal model found a counterexample to Erdős' 1946 unit distance conjecture — a result mathematicians called the first historically significant AI-generated proof.
Researchers say the new wave of results is spreading beyond one-off wins: techniques tied to the unit-distance work were quickly adapted to other problems, and Astra's latest advances have reinforced talk of a phase change in AI's math ability.
The breakthroughs are reshaping a community centered on erdosproblems.com, where hobbyists, students and top mathematicians have already used public models to solve open questions and verify proofs.
That momentum is also shifting the profession itself, with some mathematicians saying AI now outperforms them on many problems and Fields Medalist Jacob Tsimerman leaving academia for OpenAI in July 2026.
If an AI can autonomously crack elite mathematical conjectures for a few hundred dollars, is the age of human-led discovery over?
With AI solving decades-old Erdős puzzles using unexpected theories, what happens when machines generate proofs too complex for humans to understand?
Astra’s $2,000 Proofs: How OpenAI’s Multi-Agent AI Solved Ten Decade-Old Math Problems and Changed the Future of Mathematical Discovery
Overview
In August 2026, OpenAI's Astra model made headlines by solving ten major open problems in mathematics and theoretical computer science. Every proof was formalized and verified in Lean 4, allowing anyone to instantly check correctness and removing the need for blind trust. Astra’s multi-agent system tirelessly explored vast search spaces, constructing complex, counterintuitive objects that had eluded human intuition for decades. This leap shifted the bottleneck in mathematics from finding answers to verifying them, collapsing traditional peer review to a simple file download. However, Astra’s breakthroughs also sparked debate about attribution, proof integrity, and the future role of human mathematicians.