OpenAI Astra Solves 10 Mathematical Problems Amid Anthropic Challenge

OpenAI Inc. introduced Astra, a new model family designed for long-running tasks, by publishing proofs for 10 open problems in pure mathematics and theoretical computer science. The results include solving a 27-year-old question regarding the existence of non-sofic groups, alongside advances in lattice cryptography, von Neumann algebra, and quantum complexity. The core mathematical arguments were generated by Astra at an estimated compute cost of $2,000, while humans collaborated with the system to produce manuscripts and machine-checkable certificates using the Lean programming language.

While the company described the achievement as a major step for scientific reasoning, it noted that Astra did not solve any Millennium Prize Problems. CEO Sam Altman has since demoed Astra's capabilities to federal officials in Washington, D.C. OpenAI Inc. also emphasized that claiming human authorship for these AI-generated proofs would misrepresent the nature of the intellectual work.

Following the announcement, Anthropic PBC researcher Levent Alpoge claimed that the company's Claude Fable model independently solved five of the same problems, including the closest vector problem and arithmetic circuit complexity, within 24 hours using generic prompts. OpenAI Inc. has not publicly responded to these claims, and Anthropic PBC has not yet published a formal research paper to verify its results.