OpenAI and Anthropic invest $15 million to advance AI-powered mathematical research

Partnership and funding

OpenAI has pledged $5 million to a Simons Foundation initiative aimed at helping mathematicians analyze, understand, and build upon AI-generated results. Anthropic, the creator of the Claude model, is also part of this collaboration. Together, the three organizations have committed $15 million in financial and computational resources, including OpenAI’s contribution.

Purpose of the initiative

The program will be managed by the Simons Foundation, which will distribute funds through open calls for proposals. The funding will support workshops, evaluations of AI-generated results, training in computational mathematics tools, and discussions on criteria for adopting these technologies in research.

How mathematicians will be supported

In addition to funding, OpenAI and Anthropic will provide AI tools and computational resources to selected researchers. The support will not be limited to mathematics; other areas of basic science will also have access to these resources.

“The community should lead discussions about how these tools are evaluated, adopted, and used,” said David Spergel, president of the Simons Foundation.

Context and challenges

Producing a mathematical proof and explaining its meaning are two distinct tasks. The initiative aims to fund the work needed to study AI-generated results, discuss their implications, and transform these findings into knowledge that can be reused by other scientists.

OpenAI’s recent history in mathematics

In October, OpenAI published mathematical results from an internal frontier model in a public GitHub repository, including procedures for review and citation. Some of the demonstrations were formalized in Lean, a language that enables computerized verification of proofs step by step.

Formal verification checks whether a conclusion follows from the definitions and hypotheses, but understanding the proof strategy and its applications requires human analysis. OpenAI plans to expand the available formalizations and has consulted an independent advisory group on mathematics and AI at the Institute for Advanced Study to guide the disclosure.

In 2025, models from OpenAI and Google DeepMind achieved performance comparable to gold medalists in the International Mathematical Olympiad by solving competition problems. The new initiative aims to go beyond problem-solving, fostering the creation of new mathematical knowledge and peer review.