Artificial Intelligence Index Report 2025 Policy Highlights PDF Free Download
1 / 12/12
100%
Articial Intelligence
Index Report 2025
Policy Highlights
Articial Intelligence
Index Report 2025
Policy Highlights
1
Co-Directors Members
Yolanda Gil
University of Southern
California, Information
Sciences Institute
Raymond Perrault
SRI International
Research Manager and Editor in Chief
Nestor Maslej, Stanford University
Research Associate
Loredana Fattorini, Stanford University
Aliated Researchers
Elif Kiesow Cortez, Stanford Law School Research Fellow
Julia Betts Lotufo, Researcher
Anka Reuel, Stanford University
Alexandra Rome, Researcher
Angelo Salatino, Knowledge Media Institute,
The Open University
Lapo Santarlasci, IMT School for Advanced Studies Lucca
Erik Brynjolfsson
Stanford University
Jack Clark
Anthropic, OECD
John Etchemendy
Stanford University
Katrina Ligett
Hebrew University
Terah Lyons
JPMorgan Chase & Co.
James Manyika
Google, University of
Oxford
Juan Carlos Niebles
Stanford University,
Salesforce
Steering Committee
Sta and Researchers
Vanessa Parli
Stanford University
Yoav Shoham
Stanford University,
AI21 Labs
Russell Wald
Stanford University
Graduate Researchers
Emily Capstick, Stanford University
Njenga Kariuki, Stanford University
Undergraduate Researchers
Malou van Draanen Glismann, Stanford University
Armin Hamrah, Claremont McKenna College
Sukrut Oak, Stanford University
Ngorli Fii Paintsil, Stanford University
Andrew Shi, Stanford University
Articial Intelligence
Index Report 2025
Policy Highlights
2
Analytics and Research Partners
Supporting Partners
Articial Intelligence
Index Report 2025
Policy Highlights
3
Figure 1
The following is a selective summary of key AI Index report highlights that are particularly relevant to
policymakers and other policy audiences. The full AI Index Report 2025 is available at hai.stanford.edu/
ai-index/2025-ai-index-report.
1. Public sector still lags behind industry in frontier AI development
as computation needs continue to soar.
Industry continues to make signicant investments in AI and strengthen its lead in notable AI model development. Nearly
90% of notable models originated from industry in 2024 compared to 60% in 2023 (see Figure 1).
This dominance persists despite substantial global public investment in AI—led, in 2023, by the United States, with $831 million
in public spending on AI-related contracts (see Figure 2)—and academia remaining the leading institutional producer of
highly cited (top 100) AI publications over the past three years.
Large-scale industry investment is continuing to drive model scaling and performance gains as AI models are continuing to
become more computationally demanding and energy intensive (see Figure 3): The training compute for notable AI models
is doubling approximately every ve months, dataset sizes for training LLMs every eight months, and the power required for
training annually.
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
0
10
20
30
40
50
60
Number of notable AI models
0, Academia
0, Academia–government collaboration
0, Academia–research collective collaboration
0, Research collective
0, Industry–research collective collaboration
0, Government
1, Industry–government collaboration
5, Industry–academia collaboration
55, Industry
Number of notable AI models by sector, 2003–24
Source: Epoch AI, 2025 | Chart: 2025 AI Index report
Articial Intelligence
Index Report 2025
Policy Highlights
4
830.98
262.59
49.59
49.55
36.89
31.13
26.08
22.98
18.44
16.84
10.48
10.14
8.35
5.78
4.77
0 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850
Sweden
Czech Republic
Belgium
Austria
Italy
Hungary
France
Poland
Ireland
Romania
Greece
Germany
Spain
United Kingdom
United States
Public spending on AI-related contracts (in millions of U.S. dollars)
Public spending on AI-related contracts in select countries, 2023
Source: AI Index, 2025 | Chart: 2025 AI Index report
Figure 2
DeepSeek-V3
Qwen2.5-72B
Llama 2-70B
Claude 2
PaLM (540B)
Megatron-Turing NLG 530B
GPT-3 175B (davinci)
RoBERTa Large
BERT-Large
Transformer
Segment Anything Model
AlexNet
GPT-4
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
1000
10
K
100
K
1M
10M
100M
1B
10B
100B
Language Vision Multimodal
Publication date
Training compute (petaFLOP - log scale)
Mistral Large 2
Claude 3.5 Sonnet
Gemini 1.5 Pro GPT-4o
ERNIE 3.0 Titan
Training compute of notable AI models by domain, 2012–24
Source: Epoch AI, 2025 | Chart: 2025 AI Index report
Figure 3
Articial Intelligence
Index Report 2025
Policy Highlights
5
2. Remarkable technical performance jumps are accompanied by
gaps in standardized evaluation methods.
AI model performance is converging at the frontier. The AI landscape is becoming increasingly competitive with high-quality
models now available from a growing number of developers. Illustratively, in the last year, the gap between the top and 10th-
ranked model narrowed from 11.9% to just 5.4% on the Chatbot Arena leaderboard (see Figure 4).
In particular, open-weight models are catching up. The performance gap between leading open-weight models and their
closed-weight counterparts has narrowed to 1.70% on the Chatbot Arena leaderboard as of February 2025 (see Figure 5).
AI is mastering new benchmarks faster than ever. Model performance on benchmarks that test the limits of increasingly
capable AI systems (e.g., MMMU, GPQA, SWE-bench) saw remarkable improvements from 2023 to 2024, ranging from 19 to
67 percentage points. This is pushing researchers to continually propose more challenging benchmarks (e.g., Humanity’s Last
Exam, FrontierMath, BigCodeBench).
However, research has shown that many benchmarks are poorly constructed, underscoring the need for standardized
benchmarking to ensure reliable AI evaluation and to prevent misleading conclusions about model performance (see Figure 6).
Evaluating AI systems with responsible AI criteria is still uncommon, and benchmarks aimed at evaluating the factuality and
truthfulness of models have failed to gain widespread adoption.
2024-Jan
2024-Feb
2024-Mar
2024-Apr
2024-May
2024-Jun
2024-Jul
2024-Aug
2024-Sep
2024-Oct
2024-Nov
2024-Dec
2025-Jan
2025-Feb
1,050
1,100
1,150
1,200
1,250
1,300
1,350
1,400
Score
1,252, Mistral AI
1,269, Meta
1,284, Anthropic
1,288, xAI
1,362, DeepSeek
1,366, OpenAI
1,385, Google
Performance of top models on LMSYS Chatbot Arena by select providers
Source: LMSYS, 2025 | Chart: 2025 AI Index report
Figure 4
Articial Intelligence
Index Report 2025
Policy Highlights
6
2024-Jan
2024-Feb
2024-Mar
2024-Apr
2024-May
2024-Jun
2024-Jul
2024-Aug
2024-Sep
2024-Oct
2024-Nov
2024-Dec
2025-Jan
2025-Feb
1,100
1,150
1,200
1,250
1,300
1,350
1,400
Score
1,362, open
1,385, closed
Performance of top closed vs. open models on LMSYS Chatbot Arena
Source: LMSYS, 2025 | Chart: 2025 AI Index report
Figure 5
BBQ
BOLD
MMLU
ARC-Challenge
WinoGrande
GSM8K
HellaSwag
AgentBench
GPQA
BIG-bench
Procgen
Wordcraft
RL Unplugged
FinRL-Meta
SafeBench
ALE
0 5 10 15 20
0
5
10
15
Foundation models
Non-foundation models
Design score
Usability score
MedMNIST v2
TruthfulQA
MLCommons AI Safety v0.5
Machiavelli
PDEBench
DecodingTrust
HumanEval
Design vs. usability scores across select benchmarks
Source: Reuel et al., 2024 | Chart: 2025 AI Index report
Figure 6
Articial Intelligence
Index Report 2025
Policy Highlights
7
3. While the United States continues to lead in many aspects of AI
development, competition from China is intensifying.
The United States continues to surpass China and Europe as the leading source of top AI models (40 notable U.S.-developed
AI models in 2024, see Figure 7), the leading contributor of top-100 cited AI publications (173 from 2021 to 2023), and the
leading source of private AI investment ($109 billion in 2024).
However, the gap between Chinese and U.S. model performance on important benchmarks has narrowed substantially (to
less than 10 percentage points across the board by the end of 2024, see Figure 8), and China leads in AI research publication
totals, with 23.2% of global AI publications and 22.6% of global AI research citations.
While North America is maintaining its lead in organizations’ use of AI, other regions are gaining ground. Greater China
demonstrated one of the most signicant year-over-year growth rates, with a 27 percentage point increase in organizational AI
use, closely followed by Europe, which registered a 23 percentage point increase (see Figure 9).
2003
2006
2009
2012
2015
2018
2021
2024
0
10
20
30
40
50
60
70
Number of notable AI models
3, Europe
15, China
40, United States
Number of notable AI models by select geographic
areas, 2003–24
Source: Epoch AI, 2025 | Chart: 2025 AI Index report
Figure 7
Articial Intelligence
Index Report 2025
Policy Highlights
8
2022 2023 2024
0%
20%
40%
60%
80%
100%
2022 2023 2024
0%
20%
40%
60%
80%
100%
2022 2023 2024
0%
20%
40%
60%
80%
100%
2022 2023 2024
0%
20%
40%
60%
80%
100%
United States China
Average accuracy
Overall accuracy
Accuracy
Pass@1
General language: MMLU General reasoning: MMMU
Mathematical reasoning: MATH Coding: HumanEval
Performance of top United States vs. Chinese models on select benchmarks
Source: AI Index, 2025 | Chart: 2025 AI Index report
78%
72%
80%
82%
75%
77%
55%
58%
57%
61%
48%
49%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Developing markets
(incl. India,
Central/South America,
MENA)
Greater China
(incl. Hong Kong,
Taiwan, Macau)
North America
Europe
Asia-Pacic
All geographies
2024
2023
% of respondents
AI use by organizations in the world, 2023 vs. 2024
Source: McKinsey & Company Survey, 2024 | Chart: 2025 AI Index report
Figure 8
Figure 9
Articial Intelligence
Index Report 2025
Policy Highlights
9
4. Governments are stepping up on AI—with regulation and
investment —amid growing evidence of AI’s economic opportunities,
increasing AI incidents, and mounting public distrust.
AI is beginning to deliver nancial impact across business functions as the technology continues to boost productivity and
bridge skill gaps. Recent research conrms that AI can have a positive impact on productivity and often helps narrow skill gaps.
Businesses report cost savings, especially across functions like service operations, supply chain management, and software
engineering—even as most are still in the early stages of adoption.
Governments across the world are investing in AI infrastructure at scale. Canada, for example, announced a $2.4 billion AI
infrastructure package, while China launched a $47.5 billion fund to boost semiconductor production.
In the United States, the number of introduced AI-related federal regulations more than doubled in 2024; 59 AI-related
regulations came from 42 unique agencies. U.S. states are leading the way on AI legislation amid slowing progress at the
federal level (see Figure 11): In 2024, the number of state-level AI-related laws passed more than doubled from 2023, while the
number of proposed bills at the federal level grew by just 29.2%.
This policy action is set against the backdrop of continually increasing reports of AI-related incidents (see Figure 12) and a
signicant decrease globally in public condence that AI companies protect personal information and that AI systems
are unbiased.
10%11%26%
13%9%29%
10%14%32%
19%7%17%
7%15%39%
7%11%39%
10%12%21%
13%16%23%
15%35%
19%8%16%
47%
51%
56%
43%
61%
58%
56%
52%
44%
44%
Knowledge management and other
internal functions
Strategy and corporate nance
Software engineering
IT
Service operations
Supply chain and inventory management
Product or service development
Human resources
Risk, legal, and compliance
Marketing and sales
8% 24% 34%
12% 15% 25%
19% 15% 32%
18% 14% 31%
12% 13% 32%
11% 12% 47%
66%
51%
67%
63%
70%
57%
Decrease by ≤10% Decrease by 11–19% Decrease by ≥20% Increase by >10% Increase by 6–10% Increase by ≤5%
Function
Cost decrease and revenue increase from generative AI use by function, 2024
Source: McKinsey & Company Survey, 2024 | Chart: 2025 AI Index report
% of respondents
Figure 10
Articial Intelligence
Index Report 2025
Policy Highlights
10
233
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
0
50
100
150
200
Number of AI incidents
Number of reported AI incidents, 2012–24
Source: AI Incident Database (AIID), 2024 | Chart: 2025 AI Index report
AL
7
AK
0
AZ
5AR
0
CA
42 CO
7
CT
3
DE
1
FL
9
GA
3
HI
4
ID
4IL
11 IN
4
IA
4
KS
0KY
2
LA
4
ME
1
MD
17
MA
11
MI
7
MN
4
MS
6
MO
0
MT
0
NE
1
NV
2
NH
6
NJ
3
NM
3
NY
8
NC
6
ND
3
OH
2
OK
0
OR
2PA
3
RI
0
SC
1
SD
1
TN
4
TX
5
UT
17
VT
7
VA
17
WA
11
WV
4
WI
2
WY
1
Source: AI Index, 2025 | Chart: 2025 AI Index report
Number of state-level AI-related bills passed into law in the
United States by state, 2016 24 (sum)
Figure 12
Figure 11
