SAIR's Open Math Model Initiative

Announces that SAIR is starting an open model initiative for the mathematical community and is openly collecting expressions of interest from partners. Previously SAIR's activities consisted primarily of podcasts, short events, and competitions with limited resources; this initiative extends the effort toward building open weight models and open source tooling, and is being announced earlier than planned. This is an organizational announcement and invitation, with a contact email and a Zulip discussion channel; it is a call for interest rather than completed models or evaluation results.

Proposes that models be shaped by the mathematical community, with participation open across institutions, regions, and career stages. Researchers are to determine how models are trained, what they are evaluated on, and how they serve research and education, with the community needing models it can inspect, improve, and run independently. Stated as initiative principles; no implemented governance structure or participation data is provided.

Specifies that the first phase focuses on everyday mathematical work and measures value through reliable assistance, verifiable results, and the cost of sustained use. Concretizes the scope as understanding difficult arguments, checking references, exploring examples, writing code, and formalizing proofs. A planning description; the text gives no benchmarks, metric values, or pilot results.

Sets out a governance framework for open development, data ownership, and shared intellectual property. Requires open licensed weights and code, published training methods, reproducible evaluations, documented data sources and permissions, and reporting failures and limitations alongside successes; data use requires explicit consent, outputs are shared under open licenses such as Apache 2.0, MIT, or CC BY 4.0, and researchers retain ownership of their own independent and prior work. Given as principles and commitments, with specific licensing and attribution arrangements to be agreed with contributors in advance.

The initiative addresses mathematical researchers, educators, and institutions and funders supporting open science, with the first phase limited to everyday mathematical work such as understanding difficult arguments, checking references, exploring examples, writing code, and formalizing proofs; outputs are intended to be released under open licenses, and partnerships with academic and industry partners are planned to provide compute while preserving research independence.

The announcement states that some key planning is still underway and that partner announcements and further details will follow shortly; therefore model scale, compute sources, evaluation plans, final license choices, and how governance rules will operate in practice remain to be specified. The loaded text is the blog post body and comments, without attachments, tables, or technical details, so actual model capabilities or costs cannot be assessed.