Stanford HAI / CRFM

The Stanford Institute for Human-Centered Artificial Intelligence (Stanford HAI) is an interdisciplinary AI research institute at Stanford University, launched in March 2019. The Center for Research on Foundation Models (CRFM) is a research center within HAI, established in August 2021 to study and shape the development of foundation models. Together, Stanford HAI and CRFM are among the most consequential academic AI research organizations globally, with research outputs spanning AI evaluation (the HELM framework), AI policy, foundation-model research (the Alpaca instruction-tuning project, Stanford CRFM's Levanter and Marin training infrastructure), and other areas. Stanford HAI is co-directed by Fei-Fei Li, the Sequoia Professor of Computer Science and a foundational figure in computer-vision research, and John Etchemendy, Stanford Provost Emeritus and Patrick Suppes Family Professor of Humanities and Sciences. CRFM is directed by Percy Liang, Associate Professor of Computer Science.

At a glance

• Founded: Stanford HAI launched March 2019 by Stanford University. CRFM launched August 2021 as a research center within Stanford HAI.
• Status: Academic research institute. Operates as part of Stanford University with faculty affiliation across Stanford's School of Engineering, School of Humanities and Sciences, and other schools.
• Funding: Stanford University institutional support, federal research grants (NSF, DARPA, NIH, and other agencies), corporate-sponsorship support, and individual donor contributions. Stanford HAI launched with a $1 billion fundraising goal as part of Stanford's broader AI investment strategy.
• CEO: Co-Director leadership structure. Fei-Fei Li (Denning Co-Director, HAI), John Etchemendy (Denning Co-Director, HAI). Percy Liang (Director, CRFM).
• Other notable leadership: Russ Altman (Senior Fellow, HAI), James Landay (Faculty Co-Director of HAI Education), and senior research-program leadership across HAI's research portfolio.
• Open weights: Yes. Stanford research outputs are released openly through Hugging Face, GitHub, and academic-paper publication. The Alpaca instruction-tuning project (March 2023) was a foundational open-research release.
• Flagship outputs: Alpaca (March 2023; instruction-tuned LLaMA fine-tune), HELM (Holistic Evaluation of Language Models, the principal academic foundation-model benchmark suite), Levanter and Marin (open foundation-model training infrastructure), and the Stanford AI Index annual report.

Origins

Stanford HAI was launched in March 2019 by Stanford University as the principal interdisciplinary AI research institute on campus. The institute's founding mission emphasized human-centered AI: AI research informed by humanities, social-sciences, and ethical considerations alongside the engineering and computer-science research that dominated contemporary AI work. Fei-Fei Li, who returned to Stanford after a period at Google Cloud, and John Etchemendy, the Stanford provost, took on the founding Co-Director roles.

The 2019 to 2021 period built HAI's research portfolio across foundation-model research, AI for healthcare, AI for education, AI policy, AI ethics, and other application areas. Russ Altman, James Landay, and other senior Stanford faculty took on senior research-program leadership roles. The institute coordinates research across more than 200 affiliated Stanford faculty.

The Center for Research on Foundation Models (CRFM) launched in August 2021 within HAI, with Percy Liang as founding Director. The CRFM founding paper, "On the Opportunities and Risks of Foundation Models," was a academic-research artifact that established the term "foundation models" as the principal framing for large pretrained AI systems. The paper had hundreds of authors across Stanford and other institutions and shaped subsequent academic-research and industry discussion of the foundation-model paradigm.

The 2022 to 2023 period saw CRFM produce influential research outputs including HELM (Holistic Evaluation of Language Models), a benchmark framework that has analyzed dozens of foundation models with substantially more comprehensive evaluation methodology than commercial benchmarks; Alpaca (March 2023), an instruction-tuned LLaMA fine-tune that demonstrated low-cost instruction-tuning methodology; and other research on AI evaluation, AI safety, and AI for application domains.

The 2024 to 2026 period has continued the research output across HAI and CRFM. Recent work includes the Marin training infrastructure (an open foundation-model training pipeline), the Levanter library (JAX-based training infrastructure), continued HELM evaluation expansions, and continued AI policy contribution. Stanford HAI has been particularly active in AI policy through 2024 to 2026, with senior fellows contributing to US federal AI policy discussions, the AI Index annual report, and other public-engagement activities.

Mission and strategy

Stanford HAI's stated mission is to advance AI research, education, policy, and practice to improve the human condition. The "human-centered" framing has been remarkably consistent since the founding period, with HAI serving as the principal academic locus for interdisciplinary AI research that integrates humanities, social-sciences, and ethical considerations alongside core engineering and computer-science research.

The strategy combines four threads. First, foundation-model research through CRFM, contributing to academic AI research output and producing open-research artifacts (Alpaca, HELM, Levanter, Marin, foundation-model research papers). Second, AI for application domains including healthcare, education, climate, and other areas where HAI's research outputs target societal value. Third, AI policy engagement through the HAI Policy Initiative, the AI Index annual report, and senior-fellow contribution to US and international AI policy discussions. Fourth, AI education and ecosystem development through Stanford's graduate-and-undergraduate programs and other educational initiatives.

The competitive premise is that academic AI research, particularly with the resource scale Stanford has assembled and the interdisciplinary breadth HAI provides, can produce contributions that complement and balance the commercially-driven AI research at frontier labs. Stanford's positioning as the academic hub of Silicon Valley provides the institute with continuous engagement with commercial AI organizations.

Models and products

Stanford HAI is a research institute rather than a model-development organization in the conventional commercial sense. The institute's outputs include:

• Alpaca. Released March 2023. Instruction-tuned LLaMA-7B fine-tune using the self-instruct method. The release demonstrated low-cost instruction-tuning methodology and was widely adopted by the research community.
• HELM (Holistic Evaluation of Language Models). Foundation-model benchmark framework. Has evaluated over 30 foundation models with comprehensive evaluation methodology covering accuracy, calibration, robustness, fairness, bias, toxicity, and other dimensions.
• Levanter. JAX-based training infrastructure for open foundation-model research.
• Marin. Open foundation-model training pipeline including data curation, preprocessing, and training-recipe components.
• Foundation-model research papers. Annual academic-publication output across foundation-model research, AI evaluation, AI policy, and other areas.
• AI Index annual report. The principal annual industry-and-research report on AI, with comprehensive coverage of capabilities, deployment, investment, and policy.
• Stanford CRFM models. Selected research-purpose model releases including the Mistral 7B research artifacts (distinct from the Mistral AI commercial company despite the name), Marin-released models, and other research-only model artifacts.

The principal distribution channel is academic-paper publication, GitHub for training code and infrastructure, and Hugging Face for model and dataset releases under the stanford-crfm organization.

Benchmarks and standing

Stanford HAI and CRFM produce evaluation infrastructure rather than competing on capability benchmarks. HELM is widely regarded as the most comprehensive academic foundation-model evaluation framework and is regularly cited in academic-research and industry-coverage discussion of foundation-model capability.

The Alpaca release in March 2023 was particularly influential, with the demonstration that instruction-tuning could be performed at substantially lower cost than commercial alternatives shaping the broader open-source AI ecosystem's instruction-tuning methodology.

Stanford HAI's standing in the global AI research community is anchored on the foundation-model paper, HELM, the Alpaca release, the AI Index annual report, the senior-faculty cohort, and the institute's role in AI policy discussions. Industry coverage frequently characterizes Stanford HAI as the principal academic AI research institute globally, with Stanford's broader Silicon Valley engagement amplifying the institute's research impact.

Leadership

As of April 2026, Stanford HAI's senior leadership includes:

• Fei-Fei Li, Denning Co-Director, HAI. Sequoia Professor of Computer Science. Foundational figure in computer-vision research, particularly through ImageNet and the broader visual-recognition research line. Founder of World Labs, the spatial-AI Insurgent. Public face for Stanford HAI on AI policy, AI ethics, and broader AI research positioning.
• John Etchemendy, Denning Co-Director, HAI. Patrick Suppes Family Professor of Humanities and Sciences and Stanford Provost Emeritus. Senior policy and institutional leadership for HAI, particularly on humanities-integration and AI ethics.
• Percy Liang, Director, Center for Research on Foundation Models. Associate Professor of Computer Science. Lead author of the foundation-models research paper and senior figure in foundation-model evaluation research.
• Russ Altman, Senior Fellow at HAI. Stanford Bioengineering, Genetics, and Medicine professor. Senior research-program leadership.
• James Landay, Faculty Co-Director of HAI Education. Stanford Computer Science professor.

Stanford HAI's structure includes more than 200 affiliated Stanford faculty across departments, with senior research-program leaders across the institute's principal research lines.

Funding and backers

Stanford HAI's capital structure is the academic-research-institute model funded through Stanford University institutional support, federal research grants, corporate-sponsorship support, and individual donor contributions. The institute's launch was supported by a $1 billion fundraising goal, with the institute building endowment-and-gift support through the launch period.

Federal research grants from NSF, DARPA, NIH, and other agencies provide research-program-specific funding. Corporate-sponsorship partners include senior Silicon Valley AI companies and other industry partners; specific sponsor lists vary by research program and research center.

The CRFM specifically operates with research-grant support from federal agencies and from corporate-sponsorship partners, with the academic-research-institute model providing substantially more long-horizon research capability than commercial-revenue-driven research organizations.

Industry position

Stanford HAI / CRFM occupies a structurally distinctive position in the global AI research landscape. The combination of the Stanford academic affiliation, the interdisciplinary research integration, the senior faculty cohort, the Silicon Valley engagement, and the research outputs (Alpaca, HELM, foundation-models paper, AI Index, Marin) produces a profile that no other academic AI research institute matches at the same combination of attributes.

Industry coverage has frequently characterized Stanford HAI as the principal academic AI research institute globally, with the institute's role in shaping the foundation-model research framing and the AI policy discussion as particularly influential. CRFM is regularly cited as the principal academic foundation-model research center.

Strategic risks include intensifying competition for AI research talent from commercial AI labs operating in Silicon Valley, the potential for capability gaps relative to commercial frontier labs as base-model frontiers continue to scale, and the broader academic-funding environment for AI research. Strategic strengths include the Stanford academic prestige, the senior-faculty depth, the Silicon Valley engagement, the research-output legacy, and the AI policy positioning.

Competitive landscape

Stanford HAI / CRFM collaborates with and complements rather than directly competes with most other AI organizations:

• Berkeley BAIR, MIT CSAIL, CMU SCS. Peer US academic AI research institutes. Research-community overlap and collaboration.
• Allen Institute for AI, Hugging Face, EleutherAI, LAION, BigScience, MILA, Nous Research. Open-AI-research peer organizations.
• OpenAI, Anthropic, Google DeepMind, Meta AI / FAIR. Commercial AI labs with Stanford research-community connections; Stanford faculty regularly engage with these organizations through advisory roles, sabbatical appointments, and other collaborations.
• World Labs, Mistral AI, Cursor, and other Stanford-spinout AI companies. Companies founded or co-founded by Stanford HAI faculty and graduates, with continued research connection.
• Tsinghua KEG, KAIST, and other international academic AI organizations. Peer international academic AI research institutes.

Outlook

Several open questions affect Stanford HAI / CRFM's trajectory in 2026 and 2027:

• The continued evolution of the research portfolio across foundation-model research, AI for application domains, AI policy, and AI evaluation.
• HELM successor releases and continued expansion of the foundation-model evaluation framework.
• Continued Marin and Levanter open foundation-model training infrastructure development.
• Senior research-talent recruitment and retention against commercial AI labs operating in Silicon Valley.
• The institute's role in shaping US and international AI policy, particularly around AI evaluation, AI safety, and AI in regulated domains.
• The continued research-and-policy contribution to the AI Index annual report.
• The evolution of Stanford HAI's broader research-community-and-policy engagement as the AI ecosystem continues to mature.

Sources

• Stanford HAI official site. Institute overview and research reference.
• Stanford CRFM. Center for Research on Foundation Models.
• Stanford CRFM: Alpaca. Alpaca release announcement.
• Stanford CRFM: HELM. Holistic Evaluation of Language Models framework.
• Wikipedia: Stanford HAI. Comprehensive institute history reference.
• GitHub: stanford-crfm/levanter. Levanter training infrastructure.
• GitHub: stanford-crfm/marin. Marin foundation-model training pipeline.