OpenAI names Astra as next major model family; internal version solved 10 open math problems

OpenAI announced on August 1, 2026 that its next major model family is called Astra, built to handle long-running tasks and complex problems by coordinating multiple agents working together over extended periods. An internal version of Astra solved ten open problems in mathematics and theoretical computer science—problems mathematicians had made no progress on for at least a decade, and much longer in most cases. OpenAI published formal proofs in Lean for each solution and released a walkthrough of the model's reasoning process for each problem.

The model consumed about $2,000 in tokens at GPT-5.6 Sol API rates to generate all ten solutions. Astra is not yet publicly released and has no announced pricing or availability date. CEO Sam Altman has showcased Astra to politicians and regulators in Washington, D.C. The models are undergoing testing and will be the first to go through a planned U.S. government review process that requires official approval before public release. OpenAI has not confirmed whether Astra will be branded as GPT-6 or as another GPT-5 point release.

For practitioners, Astra signals a shift in OpenAI's model strategy: from released products to long-running agentic systems that reason over hours or days and can solve previously intractable problems. The mathematical breakthrough demonstrates test-time compute can be leveraged far beyond current released models. The government review process indicates frontier AI model deployment may now face regulatory gatekeeping before public availability, a structural change in how frontier labs ship.