OpenAI has announced that an internal, unreleased version of its next major model — called Astra — solved or made significant progress on 10 long-standing open problems in mathematics and theoretical computer science.
According to Livemint, the problems had seen little or no progress for at least a decade. SiliconANGLE reports that OpenAI published the proofs alongside the claim, which matters: unlike a chatbot answer you have to take on faith, a mathematical proof can be checked line by line by human experts. Digit makes a similar point, arguing the release deserves "a closer look than just the small paper where it is published."
The way OpenAI made the announcement drew nearly as much attention as the result. Gizmodo says the company "smuggled" the reveal of Astra, its next AI model, into a blog post about math — a low-key rollout for what is billed as a next-generation system. BleepingComputer framed it as OpenAI teasing Astra off the back of the math results.
Rival Anthropic responded quickly. Moneycontrol reports that Anthropic claims its publicly available Claude Fable model independently solved five of the problems — a pointed contrast, since Astra remains unreleased while Fable is already shipping.
Moneycontrol also notes the announcement has reignited debate over where AI is headed, with the results held up as evidence of growing potential for tackling complex scientific challenges.
A few caveats are worth keeping in view. Astra is not available to the public, the claims come from the companies themselves, and coverage varies on whether the model fully "solved" the problems or merely advanced them — The American Bazaar and The Indian Express describe the work as advances and breakthroughs rather than clean solutions.
Why it matters: if AI can genuinely produce new mathematics that experts couldn't, it shifts these systems from tools that summarize existing human knowledge to ones that add to it — and the published proofs mean, for once, the claim can actually be verified.