OpenAI Releases AI Math Manuscripts Solving Hundreds of Open Problems
OpenAI published 722 manuscripts covering 372 result families generated by an unreleased frontier AI model, detailing solutions to hundreds of previously unsolved mathematical problems. The repository includes reasoning process summaries, compute expenditure estimates, and problem-solving attempt statistics. This represents a major milestone for automated reasoning and AI for Science, proving that frontier AI models can solve long-standing open research problems. However, it also triggers serious debates regarding research ethics, academic publishing standards, and how credit is attributed to human mathematicians whose work underpins these systems. OpenAI reported that generating a single average result consumed compute roughly equivalent to ChatGPT Pro thinking continuously for three hours. The release follows recommendations from the newly established Advisory Group on Mathematics and Artificial Intelligence (AGMAI), which urged AI labs to publish through established academic channels and disclose model parameters, prompts, and compute costs.
## BACKGROUND
Automated theorem proving and mathematical reasoning have been long-standing goals in computer science, but AI systems historically had little impact on core mathematical research. The Advisory Group on Mathematics and Artificial Intelligence (AGMAI) was established at the Institute for Advanced Study to help the mathematical community navigate ethical norms, credit attribution, and publication standards as AI models rapidly advance in pure mathematics.