OpenAI's largest math release tackles 4,000 problems with Lean proofs
The release includes 722 manuscripts, 372 research families and Lean-checked proofs, while OpenAI says the model attempted about 4,000 problems.
- On Tuesday, OpenAI released 722 AI-generated mathematical manuscripts in a public GitHub repository, organized into 372 research families spanning number theory and theoretical computer science.
- OpenAI developed this release process with advice from the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, seeking to "directly empower scientists with state-of-the-art capabilities."
- The repository includes formal proofs in Lean, allowing computers to verify logical steps, while OpenAI provided reasoning summaries for 10 families showing how the model approached selected problems.
- While the Advisory Group urged labs to refrain from treating results as marketing vehicles, OpenAI disclosed only average compute times; OpenAI spokesperson Lindsay McCallum said the company remains unbound by these recommendations.
- The release adds to a growing body of AI-generated results the mathematical community is still processing, though mathematician Terence Tao has previously criticized the "insane" pace of such breakthroughs.
89 Articles
89 Articles
OpenAI has published 722 mathematical research papers at once, proving and disproving a whole range of problems. For scientists, such a scale is both impressive and disconcerting, but the shock that hit mathematicians should also sober the public, finds Kristjan Port in R2's tech commentary.
OpenAI has released 722 papers that solved 372 difficult mathematical problems using undisclosed AI models. In response, the mathematics community is concerned that the indiscriminate generation of correct answers by AI will undermine logical understanding, which is the core of the discipline, and is urging a return to human-centered research and the cessation of cooperation with AI companies.
OpenAI says it has solved 377 maths equations that have stumped experts for years
Mathematicians have lashed out after the AI giant released hundreds of new findings.
OpenAI releases findings on hundreds of math problems
Unreleased AI model attempts approximately 4,000 research problems during testing, according to OpenAI
Coverage Details
Bias Distribution
- 49% of the sources lean Left
Factuality
To view factuality data please Upgrade to Premium

































