OpenAI's Latest Models Crack 377 Unsolved Math Problems Through Computational Brute Force

OpenAI has released solutions to 377 previously unsolved mathematical problems using an advanced internal model, marking a significant milestone in artificial intelligence's capacity for frontier-level research and raising complex questions about the future of mathematics as a human discipline. The company announced on October 6, 2026, that its unreleased model had tackled 722 manuscripts organized into 372 result families, with each solution requiring an average of three hours of computing time. The problems span multiple domains including algebra, number theory, theoretical computer science, mathematical logic, and topology, with notably ambitious claims including partial solutions to the Riemann hypothesis and work on three of the seven Millennium Prize Problems. The results follow OpenAI's September 2026 release of GPT-6 Astra, a frontier model that achieved 98 percent accuracy on the FrontierMath Tier 4 benchmark—a test comprising 50 problems that typically require human mathematicians several weeks to solve individually—and solved ten open problems in mathematics and theoretical computer science, complete with machine-checkable Lean 4 proofs published on GitHub.