Meta AI solves five open math problems across six papers
Math and money, both from the same wave: Meta says its models closed five previously open research questions, and a 3D-generation startup says the general-model era grew its business instead of killing it.
Meta published six papers on October 2 that it says answer five previously open research questions — spanning probability, PDEs, group theory, optimization, arithmetic physics and non-associative algebra. The setup matters more than the count: mathematicians co-authored the work, a second group of mathematicians reviewed it, and Meta says the papers were produced in an ordinary meta.ai chat session, with no custom scaffolding built for the task. The caveats are real. Two of the six are disproofs of recent, narrow conjectures — including a 384-element counterexample to a 2024 conjecture, built from a search program Muse Spark wrote — three of the papers credit independent concurrent solutions found by other teams, and none of it has passed journal peer review yet. The "AlphaGo moment for mathematics has arrived this year" line circulating in Chinese media is Shuchu Bi, head of Meta Superintelligence Labs, speaking about his own lab; treat it as a quote, not a verdict. Why it matters: labs have spent a year demonstrating theorem-proving, and this is the first batch where the output is new answers to questions working mathematicians actually posed — which sharpens a credit problem we flagged when an AI closed a real conjecture two weeks ago — React's creator closed Conway's 50-year-old conjecture with AI.
Meshy says its annual recurring revenue went from $1 million to $100 million in under two years — a growth curve arguing that the GPT-6 Astra wave is feeding specialized 3D generation rather than replacing it. The number is company-issued (a September 30 announcement), but the trajectory holds up: 36kr reported roughly $60 million when the startup closed its roughly $400 million Series B in July, and QbitAI's analysis makes the sharp point — Astra's viral Blender demos handle the geometric, code-buildable parts (walls, staircases, parametric CAD), while production-grade characters — hair, cloth folds, engraved shield patterns — still fall to purpose-built models, so the more projects agents can start, the more good assets cost money. Named customers in the coverage include 37 Interactive, NetEase, Nexon and printer maker Bambu Lab. The headline number remains unaudited and unconfirmed by any top-tier English outlet — but if it holds, Meshy beat HeyGen's 29-month trip from $1 million to $100 million and landed inside what Bessemer calls "supernova" territory.
What to watch: whether Meta's six papers survive journal peer review — and which of the concurrent outside solutions reaches print first.
If Meta's five answers hold up, does an AI-assisted proof deserve the same credit as a human one? Tell us in the comments.
