OpenAI's Astra Solves 10 Decades-Old Math Problems

OpenAI dropped a strange claim yesterday: an internal, unreleased model called Astra just solved 10 open problems in math and theoretical computer science, some of which had sat unsolved for decades. One dates back nearly 30 years. The list spans geometry, group theory, and quantum complexity, territory where progress usually comes from specialists spending years on a single proof, not a chatbot spitting out an answer overnight.

What makes this different from the usual benchmark bragging is the price tag. OpenAI says the cost of running Astra on these problems has dropped low enough that mathematicians could plausibly throw entire backlogs of open questions at it, not just cherry-picked showcase cases. That changes the math here from "look what our model can do" to "here's a tool you could actually use."

Details on Astra itself are thin. It's described as an internal version of OpenAI's next major model family, not something available to the public yet. But the specific problems it cracked, including a proof involving non-sofic groups, give a clearer picture of what it's actually capable of than any leaderboard score would.