OpenAI says an internal version of Astra, the company’s next major model, was used to produce solutions to ten previously open problems in mathematics and theoretical computer science. OpenAI emphasizes that these fields allow straightforward verification of solutions, making the results amenable to independent checking.
Which problems were solved
According to OpenAI, the solved problems cover high‑dimensional geometry, coding theory, arithmetic circuit complexity, group theory, operator algebras, quantum complexity, lattice cryptography and extremal combinatorics. The company states that each of these problems is of substantial interest within its respective mathematical community, and several are of broad interest across mathematics.
Reactions from the field
Several experts have acknowledged the significance of the proofs. Henry Yuen, an associate professor of computer science at Columbia University, highlighted that the questions involved—such as new circuit lower bounds, an easy‑to‑describe non‑sofic group, or hardness of approximation for CVP without relying on a unique‑games‑type conjecture—are central to the area. Yuen wrote that he feels the importance of these questions personally and cares deeply about their answers.
Why this matters
The announcement can be interpreted in two ways. One interpretation is that the results indicate AI systems are beginning to exhibit the kind of creative intuition that helps solve research problems. The other is that AI systems have become capable of reliably solving extremely complex, open problems where humans have already defined the research direction and structure. Both readings imply that large AI models are increasingly powerful scientific tools in their own right.
An open question remains whether AI systems can autonomously formulate new research questions that expand the frontier of human knowledge, rather than primarily following human‑defined directions. Nevertheless, OpenAI’s results suggest we are moving closer to a situation where AI may act not only as an assistant but as an active research partner.
Further reading
OpenAI published more details in a blog post titled “Ten advances in mathematics and theoretical computer science,” which outlines the specific problems and the role of their models.
