Stanford HAI AI Index Report
Last reviewed
Jun 8, 2026
Sources
8 citations
Review status
Source-backed
Revision
v1 · 1,270 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 8, 2026
Sources
8 citations
Review status
Source-backed
Revision
v1 · 1,270 words
Add missing citations, update stale details, or suggest a clearer explanation.
The AI Index Report is an annual, independent, data-driven publication that tracks, distills, and visualizes trends in artificial intelligence across research and development, technical performance, the economy, education, policy and governance, responsible AI, and public opinion. It is produced by the Stanford Institute for Human-Centered AI (Stanford HAI) and is widely regarded as one of the most comprehensive and most-cited cross-industry sources of AI metrics. The eighth edition, the 2025 AI Index Report, was released in April 2025 and runs to 456 pages. [1][2][6]
The AI Index aims to provide unbiased, rigorously vetted, and globally sourced data so that policymakers, researchers, journalists, executives, and the general public can develop a grounded understanding of the state of AI. Rather than advancing a thesis, each edition compiles original measurements and third-party data into hundreds of charts, organized into thematic chapters and accompanied by a set of "top takeaways." [1][6]
The report is openly published as a free PDF, with accompanying chapter pages, an interactive set of charts, and an underlying public data set. It is also deposited on arXiv, where the 2025 edition appears as report 2504.07139. [2][6]
The AI Index is frequently compared with two other prominent AI-tracking efforts: the State of AI Report, authored by investor Nathan Benaich and Air Street Capital (also in its eighth year as of 2025), and the research and data published by Epoch AI, which specializes in compute, model-scale, and trend estimates. The AI Index draws on data from many such organizations, and the three efforts are often read together as complementary references. [1][5][8]
The AI Index was launched in 2017 as a project of the One Hundred Year Study on Artificial Intelligence (AI100) at Stanford University, an open, not-for-profit effort to track activity and progress in the field. The inaugural 2017 report was framed as a starting point for rigorously measuring AI activity, with a steering committee that included Erik Brynjolfsson, Yoav Shoham, Raymond (Ray) Perrault, and Jack Clark. [3][4]
After Stanford HAI was founded in 2019, the AI Index became one of its flagship initiatives. Successive editions have grown substantially in scope: early reports were tens of pages, while the 2024 and 2025 editions each exceed 400 pages. The number of thematic chapters has also expanded over time to accommodate new areas such as responsible AI and, in 2025, a dedicated chapter on science and medicine. [1][6][7]
| Edition | Year | Approx. length | Notable additions |
|---|---|---|---|
| 1st | 2017 | Tens of pages | Inaugural report under AI100 |
| 7th | 2024 | 400+ pages | Expanded responsible AI; foundation-model tracking |
| 8th | 2025 | 456 pages | New chapter on science and medicine; AI hardware and inference-cost analyses |
(Lengths are approximate; the 2024 and 2025 figures are the editions most frequently cited.) [1][6][7]
The 2025 edition is organized into eight chapters: Research and Development; Technical Performance; Responsible AI; Economy; Science and Medicine; Policy and Governance; Education; and Public Opinion. Across these chapters the report measures items such as: [6]
The AI Index is best known for a handful of headline findings each year that are widely quoted by the press. Selected examples from the two most recent editions:
2024 edition (7th). Industry continued to dominate frontier AI research, producing 51 notable machine-learning models in 2023 versus 15 from academia, with a record 21 industry-academia collaborations. A total of 149 foundation models were released in 2023, more than double the 2022 count, of which roughly 66 percent were open-weight. The report flagged that AI had saturated several established benchmarks (such as ImageNet, SQuAD, and SuperGLUE), prompting the field to adopt harder evaluations including SWE-bench and MMMU. Overall private AI investment declined year over year, but funding for generative AI nearly octupled to about $25.2 billion. [7]
2025 edition (8th). U.S.-based institutions produced 40 notable AI models in 2024, ahead of China's 15 and Europe's 3, but the report emphasized that Chinese models had largely closed the quality gap on major benchmarks such as MMLU and HumanEval, narrowing from double-digit differences in 2023 to near parity in 2024. U.S. private AI investment reached $109.1 billion in 2024, roughly 12 times China's figure, and generative AI attracted $33.9 billion globally. The cost of querying a model at GPT-3.5-equivalent performance on MMLU fell from about $20 per million tokens in late 2022 to roughly $0.07 by October 2024, a reduction of more than 280-fold. Business adoption rose to 78 percent of surveyed organizations using AI, up from 55 percent the prior year. [1][6]
The AI Index is overseen by an AI Index Steering Committee and produced by a research team at Stanford HAI, supported by external contributors and partner organizations that supply data. In recent editions the report has been led by Research Manager Nestor Maslej as editor-in-chief; the 2025 report lists Maslej alongside 22 co-authors. Long-standing committee members and contributors associated with the project include Ray Perrault, Jack Clark, and Erik Brynjolfsson. [1][3][6]
Methodologically, the report combines original analyses (for example, benchmark tracking and cost estimates) with data licensed or sourced from third parties. The 2025 edition adds in-depth analyses of AI hardware, novel estimates of inference cost, and new analyses of publication and patenting trends, along with expanded coverage of AI in science and medicine. The report stresses that its data is "rigorously vetted" and globally sourced, and it publishes its underlying data and chart code for transparency. [1][6]
The AI Index is routinely cited by policymakers, journalists, executives, and researchers worldwide as a neutral reference point on the state of AI, and its annual release is covered by major technology and business outlets. Its charts and takeaways are frequently reproduced in policy briefings and corporate strategy materials, and organizations such as IBM and IEEE Spectrum publish summaries of each edition's key findings. [1][7][8]
The report's combination of scale, independence, and free availability has made it a default citation for high-level AI statistics, a role it shares with the State of AI Report and Epoch AI's data. Because the AI Index aggregates and standardizes data across many domains in a single document, it is often treated as a starting reference even by audiences who then turn to more specialized sources for deeper technical detail. [1][5][8]