Stanford University's Center for Research on Foundation Models
Interdisciplinary research initiative within Stanford HAI, launched in 2021 and directed by Percy Liang. It maintains the HELM evaluation framework and the Foundation Model Transparency Index.
Industry: Artificial intelligence research | Location: Stanford, United States | Founded: 2021
About Stanford University's Center for Research on Foundation Models
The Center for Research on Foundation Models (CRFM) is an interdisciplinary research initiative at Stanford University in Stanford, California. It sits within the Stanford Institute for Human-Centered Artificial Intelligence (HAI). Its website says it brings together faculty, students, researchers and engineers from more than ten departments to study the development and deployment of foundation models, which the site describes as resource-intensive models that have become the dominant approach in AI in the 2020s.
Stanford HAI announced the centre on 18 August 2021 and named Percy Liang, an associate professor of computer science, as its director. HAI said at launch that more than 175 researchers across over ten departments were involved. The centre's founding publication, On the Opportunities and Risks of Foundation Models, is a roughly 200-page report with 114 listed authors, Rishi Bommasani first and Liang last. HAI's interview marking the first anniversary said the first year also produced an agenda-setting workshop and an open-source library for training foundation models, released on GitHub as Mistral.
The website groups the centre's work into technical foundations, beneficial applications, societal impact and AI policy, and links projects including HELM, Levanter, the Foundation Model Transparency Index, Openness and Ecosystem Graphs.
HAI announced the Foundation Model Transparency Index on 18 October 2023. Built by the centre with a team from Stanford, MIT and Princeton, it scored 10 foundation model developers on 100 indicators, with top scores between 47 and 54. The centre's blog lists later evaluation work, including HELM Safety in November 2024, HELM Capabilities in March 2025 and HELM Long Context in September 2025.
In 2026 the centre worked with Arabic.AI, a Dubai-based enterprise AI company, on Arabic-language evaluation. Wamda reported on 28 January 2026 that the two had announced a benchmark for Arabic language models built by extending HELM, with an Arabic leaderboard as the first phase. The centre's blog lists HELM Arabic on 18 December 2025 and HELM Arabic Enterprise on 26 May 2026. TechScoop reported on 10 June 2026 that the enterprise benchmark covers six tasks, including financial reasoning and legal question answering.
Leadership
- Percy Liang, Director
- Rishi Bommasani, Society Lead