A study titled “The Gold Rush in AI4Math: Where Are We Now?” analyzed 32,944 mathematics papers submitted to arXiv between March 1 and August 20, 2026. Researchers Jiashun Jin, Zheng Tracy Ke, and Bingcheng Sui found that the share of preprints acknowledging AI use rose from less than 5% in March to 24.14% by August 20.
The analysis highlights a shift toward substantive AI contributions, where models assist in proof construction rather than just formatting. This metric grew from 1.39% in March to 14.09% by August 20. Out of 3,575 papers confirming AI usage, 1,712 involved such substantial contributions.
OpenAI products are used in over 60% of papers with confirmed AI involvement, followed by Anthropic tools. Combinatorics leads in the volume of substantial AI use, while Metric Geometry shows the highest adoption rate relative to its size. Authors from the US and China account for approximately two-thirds of AI-assisted math papers.
Researchers identified 717 open-problem records in papers with substantive AI contributions, with 71% of these problems reported as fully resolved. The study authors emphasize that these figures reflect voluntary disclosures, suggesting the actual prevalence of AI-assisted research is likely higher.
