Terence Tao Launches SAIR Foundation's Open Math Model Initiative

On September 18, 2026, Fields Medalist Terence Tao announced the launch of the Open Math Model initiative on behalf of SAIR Foundation. According to the announcement, the program is meant to unite academia and industry in building open-weight models and companion open-source tooling for mathematics and scientific research.

The first phase focuses on the everyday work of mathematics: understanding difficult arguments, checking the literature, exploring examples, writing code, and formalizing proofs. The foundation says the value of these models will be measured by reliable assistance, verifiable results, and the ongoing cost of daily use, not by one-off demonstrations.

On development, the initiative commits to releasing weights and code under open licenses, publishing training methods, and providing reproducible evaluations. Training data will carry traceable provenance and compatible licensing, with sharing boundaries clearly labeled. The announcement also states that model releases should report failures and limitations alongside successes, and that independent teams should be able to reproduce the work and adapt models to their own mathematical needs.

Data and governance form another pillar. The mathematical community owns its data and decides how it is used; models will only be trained or improved on user data with explicit consent and pre-agreed terms, and researchers decide what to share and for what purpose. Governance rules and decisions will be public, members will have clear channels to propose ideas, join decisions, and hold leaders accountable, and industrial partners supplying resources such as compute must preserve the community's research independence.

On intellectual property, models, code, and tools developed under the program will be shared with the community under open licenses such as Apache 2.0, MIT, or CC BY 4.0 as appropriate. Researchers retain ownership of their independent prior work, and contributions of mathematical ideas and results shared with the initiative earn attribution, with ownership and credit for jointly developed output agreed with contributors from the outset.

Tao also explained the accelerated timeline. The foundation had originally planned to move step by step, starting with several pilot projects over the coming months, but given the current moment and the strong demand for open models, it decided to speed up and bring the first public announcement forward even though some key planning is still in progress.

The program is still in its recruitment phase. SAIR is now collecting expressions of interest from partners who can contribute funding, compute, expertise, or community building, with registration at the bottom of the announcement page. Tao said more updates will follow shortly, including a partner list and plans for folding the initiative into the foundation's existing podcasts, events, and competitions. SAIR Foundation was founded in 2025 by Tao and is guided by Nobel laureates and leading scientists.

The signal to watch is whether a nonprofit, community-governed open-weight route can carve out space between closed commercial frontier models and purely academic projects. Partner names and pilot deployments will show whether the effort becomes shared infrastructure for mathematics, or remains a statement of intent.