Recent Mathematical Breakthroughs
On August 1, 2026, OpenAI announced ten significant advancements in mathematics and theoretical computer science, achieved through an internal version of their upcoming model, Astra. These results address long-standing open problems across various mathematical fields.
Key Results from Astra
The advancements include:
- New upper bounds on high-dimensional sphere-packing density, reaching down to the Cohn–Elkies threshold.
- Exponentially improved bounds on the maximum size of binary codes at any prescribed minimum distance, with similar results for high-dimensional spherical codes.
- A construction that establishes the existence of non-sofic groups, addressing a central question in group theory.
- A disproof of Connes’s rigidity conjecture, which posited that certain groups are uniquely determined by their von Neumann algebras.
- New lower bounds in arithmetic circuit complexity for computing the permanent, including a lower bound of order n4/log n.
- An exponential parallel repetition theorem for general two-player quantum games, extending classical complexity theory principles.
- Polynomial-factor hardness of approximation for the closest vector problem, relevant to post-quantum cryptography.
- A resolution of Ehrhart’s volume conjecture regarding convex bodies with a single interior lattice point.
- A superexponential lower bound for multicolor triangle Ramsey numbers, resolving Erdős problem 183.
- Results on compactness and degeneracy conjectures, addressing Erdős problems 146 and 180.
Process and Community Engagement
OpenAI noted that the token sequences required to derive these solutions would have cost approximately $2,000 at Sol API rates. For each solution, the company also released a narration of the model’s thinking process.
In May, OpenAI had previously shared an AI-generated disproof of the Erdős unit-distance conjecture, discovered during the evaluation of an unreleased model.
OpenAI emphasizes its commitment to the mathematical community, acknowledging the importance of proper attribution. The company stated that it prepared the manuscripts and formalized the proofs in Lean, taking responsibility for their correctness while attributing the mathematical arguments to the model. The manuscripts, Lean certificates, and model narrations were published alongside the August 1 announcement.
Original source: OpenAI News