OpenAI’s Astra Model Solves 10 Long-Standing Mathematical Problems

OpenAI has announced that an internal version of its next major model, Astra, has solved ten open problems in mathematics and theoretical computer science. These specific challenges had seen no progress on their primary results for at least a decade, marking a significant milestone in AI-driven discovery.

To ensure the validity of these breakthroughs, OpenAI formalized each argument into a Lean certificate, making the proofs available on GitHub alongside the model's narration of its thinking process. The solved problems span a wide array of complex fields, including high-dimensional geometry, coding theory, arithmetic circuit complexity, group theory, operator algebras, quantum complexity, lattice cryptography, and extremal combinatorics. Notable achievements include the resolution of Erdős problem 183 via a superexponential lower bound for multicolor triangle Ramsey numbers and the disproof of Connes’s rigidity conjecture regarding groups and von Neumann algebras. According to OpenAI, the total token cost to reach these solutions was approximately $2,000 at Sol API rates.

The Path to Formalization

This announcement follows a previous milestone in May, when OpenAI released an AI-generated disproof of the Erdős unit-distance conjecture. The current push toward formal verification via Lean is part of a broader strategy to move beyond "black box" AI outputs. By providing machine-checkable proofs, OpenAI is attempting to bridge the gap between generative AI and the rigorous standards of mathematical verification.

Implications for Research

The ability of a model to resolve problems that have stumped human mathematicians for years suggests a paradigm shift in theoretical research. This capability raises critical questions about the future of intellectual discovery and attribution. OpenAI explicitly stated that claiming human authorship for a proof generated entirely by an AI system would misrepresent both the system’s contribution and the nature of genuine human intellectual work.

Expanding Academic Access

As part of its effort to empower the scientific community, OpenAI has launched the "ChatGPT for Academic Researchers" initiative. This program provides 100,000 researchers with free access to the company's top models to accelerate peer-led discovery. Observers will now watch to see how the broader mathematical community validates these Lean certificates and whether Astra can be applied to other unsolved conjectures in the hard sciences.