OpenAI publishes hundreds of AI-generated solutions to unsolved math problems

By Splitfeed EditorialLast updated Oct 7, 2026, 21:00 UTC

On October 7, 2026, OpenAI published a large set of AI-generated mathematical results covering problems that had not previously been solved. The exact count differs across outlets: Ynetnews and Fortune report figures in the range of 370 to 377 previously unsolved problems now addressed, while SiliconANGLE puts the total at 722 AI-generated mathematical discoveries. The difference likely reflects whether a given outlet counted only the unsolved problems that received solutions or tallied all mathematical findings in the release, including results that may not have been formally classified as open problems.

The significance of the release, if the figures hold up to scrutiny, lies in the scale and the nature of the output. Unsolved mathematics problems are not uniform in difficulty or importance, and a count of hundreds carries very different weight depending on which problems are included. Fields ranging from number theory to combinatorics to geometry each have their own catalogues of open questions, some decades or centuries old, others more recent and technical. Whether OpenAI's AI systems produced formally verified proofs, plausible solution sketches, or something in between is a distinction that matters enormously in mathematics, where community acceptance of a proof requires checking by qualified mathematicians. None of the outlets whose summaries are available have reported on the verification status of these results.

None of the three camps, Pro-AI, Anti-AI, or Middle Ground, have published reactions to this story yet. The Pro-AI camp would typically treat a release like this as evidence that AI is accelerating scientific discovery and moving beyond narrow benchmarks into genuine research contributions. The Anti-AI camp would likely raise questions about whether the solutions are formally rigorous, who has independently verified them, and whether framing AI-generated output as mathematical breakthroughs overstates what the systems actually did. The Middle Ground camp would probably welcome the results cautiously while emphasizing the need for peer review and noting that the value of the work depends entirely on its mathematical validity.

The clearest next development to watch is whether the mathematical community, through journals, preprint servers, or formal verification systems, accepts or challenges the released solutions. Any independent review that confirms or disputes specific results would settle the question of whether this release represents genuine mathematical progress or a more limited demonstration of AI-assisted problem exploration.