- OpenAI's unreleased AI model Astra solved 10 complex problems in math and quantum science
- Astra generated proofs independently, verified using the Lean proof assistant tool
- Each mathematical result cost about $2,000 in AI computing power at current API prices
OpenAI has revealed that its upcoming, unreleased AI model, named Astra, has successfully solved 10 long-standing problems in the fields of mathematics, quantum complexity, and theoretical computer science.
If these claims hold up under close inspection by the global scientific community, the achievement will indeed mark a historic shift. It just proves how much AI is moving away from being a basic digital assistant that simply helps humans with tasks, to becoming a genuine contributor to high-level scientific research.
How the AI Solved the Problems
According to an official blog post by OpenAI, the AI model generated the complex mathematical proofs entirely on its own. Following this, researchers stepped in to verify the work. They converted the AI's generated proofs into a formal, computer-verifiable format using a specialized proof assistant tool called 'Lean'. This process allowed scientists to independently check every single logical step of the AI's work to ensure there were no errors.
OpenAI also provided a glimpse into the financial cost of this advanced computing. The company stated that at current API prices, it cost roughly $2,000 in AI computing power to produce each successful mathematical result.
What Did the AI Fix?
The problems solved by the unreleased Astra model involve deeply complex academic theories. According to OpenAI, the advances include creating better bounds for sphere packing and error-correcting codes, as well as the claimed construction of the very first explicit "non-sofic group". A "non-sofic group" is a highly theoretical mathematical structure made of infinite symmetries that cannot be approximated or modeled by simpler, finite systems.
Additionally, the model managed a proof of the quantum parallel repetition theorem for general two-player entangled games. It also delivered fresh results on two famous mathematical theories: Ehrhart's volume conjecture and Connes' rigidity conjecture.
A Growing Trend in AI Research
This is not the first time OpenAI has claimed success in the field of mathematics. Just a few months ago, the company behind ChatGPT announced that one of its reasoning models had made a genuine, original advance on the 80-year-old planar unit distance problem, which was first posed by the legendary mathematician Paul Erdos. That specific breakthrough was later independently examined and verified by researchers within the global mathematics community.
Astra's new findings are expected to face the exact same level of intense scrutiny. As is standard with any major scientific claim, independent verification and strict peer review will ultimately determine just how significant these 10 new solutions are.
However, the announcement adds to a growing mountain of evidence that advanced AI systems are now confidently tackling research problems that were once thought to be completely out of reach for machines.
