OpenAI’s Astra just cracked 10 long-standing math problems. The new multi-agent model solved a set of difficult open mathematical problems that had remained unproven for decades — then formalized the proofs in Lean. Sam Altman presented Astra in Washington this week. It is set to become the first model to go through a new government approval process before release. The most striking detail: solving all 10 problems used less than $2,000 worth of tokens at Sol’s pricing. OpenAI has tried other hard problems too, with some still unsolved — but the scaling potential is enormous. Could this mark a new era for AI-assisted mathematical discovery?