What's this about?
Noam Brown of OpenAI revealed that an internal version of Astra — OpenAI's next major model family — has solved 10 significant open problems spanning mathematics, quantum complexity, and theoretical computer science. OpenAI published the results in a post titled "Ten advances in mathematics," calling it a major step for scientific reasoning. Crucially, this isn't a benchmark score: these are problems no human had solved yet.
What exactly is an "open problem"?
An open problem is a question mathematicians have formally posed but nobody — human or machine — has been able to answer. It's the difference between acing an exam (where answers exist) and pushing the boundary of what's known. Until now, AI models mostly re-solved things humans already knew; solving genuinely open problems means the model produced new mathematics that experts then verified. That's why the skeptics' bar was "wait for the verdict of a real mathematician."
What's the community saying?
- The moment that flipped the mood: number theorist Daniel Litt (@littmath), one of the most level-headed AI-skeptical mathematicians, replied simply "It's a big deal" — Ethan Mollick and others said they were waiting precisely for that signal.
- A running joke about the "littmath test" — even the person he replied to said he opened X after a bike ride just to check whether littmath had commented yet.
- Thariq (Claude Code dev) framed it as Jevons paradox for math: as math gets cheaper to produce, demand for mathematicians goes up, not down.
- Astra-watchers note this is a pre-announcement play: OpenAI is establishing "scientific reasoning" as the headline capability of its next family before any product ships.