OpenAI's AI model solves ten long-standing math problems, sparking debate among researchers
Just the facts

OpenAI's AI model solves ten long-standing math problems, sparking debate among researchers

Summary

OpenAI announced that its unreleased Astra model produced solutions to ten decades-old mathematical problems, prompting both excitement and concern within the academic community about credit, cost and the future of the discipline.

OpenAI disclosed that its internal Astra model resolved ten open problems across various branches of mathematics, including sphere-packing in high dimensions, error-correcting codes, network structure thresholds, quantum game theory and the existence of non-sofic groups. The company released more than 250 pages of technical papers and verification code, and later updated its announcement to acknowledge prior work by researchers Andreas Thom and Gábor Kun, after criticism that the original wording downplayed their contributions.

Mathematicians such as Oxford Fields Medalist James Maynard said the results represent questions that experts have pursued for years without success. Some scholars, like Yang-Hui He of the London Institute for Mathematical Sciences, described the recent pace of AI breakthroughs as a “phase transition” in the field. Others expressed uncertainty about the long-term impact, noting that many problems solved by AI are those typically tackled by graduate students and could affect training and assessment.

Cost and access were also raised as concerns. OpenAI estimated the token cost of generating the ten solutions at about $2,000, but researchers warned that the true expense, including trial runs, could be higher and potentially limit participation by less-funded institutions. The proprietary nature of leading AI models contrasts with the traditionally open-source tools used in mathematics, prompting calls for open-weight alternatives.

The debate extends to how credit is assigned. Kun described the original announcement as “rather comical” and “sloppy,” while Cambridge mathematician Francesco Fournier-Facio warned that simplified headlines might lead policymakers and journalists to overlook the human contributions underlying the AI-generated results. A recently published Leiden Declaration, endorsed by the International Mathematical Union and signed by thousands of mathematicians, urges responsible communication about AI capabilities to avoid overstating their impact.

While some, such as He, see potential benefits—including broader participation and freeing researchers from routine work—others remain cautious. Maynard noted that the solved problems, though impressive, do not yet appear to constitute fundamentally new mathematical ideas worthy of the highest honors. The community awaits further evaluation of the proofs and their influence on future research directions.

Keep reading — for free

Create a free account to follow the news. No card required.