Stanford University’s latest AI Index report says the performance gap between open and closed AI models is close to disappearing, a sign of how quickly the field has shifted in just the past year. The annual report also finds that AI capability is advancing rapidly across coding, reasoning and scientific tasks, while concerns over safety, governance and trust continue to mount.
The 2026 edition of the report describes a landscape in which frontier AI progress is being driven overwhelmingly by industry. It says private companies produced more than 90% of notable frontier models in 2025, and several leading systems now match or surpass human baselines on PhD-level science questions, multimodal reasoning and competition math.
One of the clearest indicators of that pace is coding performance. On the SWE-bench Verified benchmark, which measures software engineering ability, reported scores climbed from 60% to nearly 100% in a single year. Stanford also notes that adoption is expanding well beyond the lab. It says 88% of organizations now use AI, while four in five university students use generative AI tools.
The report says competition between the United States and China has intensified to the point that the performance gap between their top models has effectively closed. The two countries have exchanged leadership several times since early 2025. In one example, DeepSeek-R1 briefly matched the leading U.S. model in February 2025. By March 2026, Anthropic’s top model was ahead by just 2.7%, according to the report.
Even so, the report says the broader picture still favors the U.S. in some areas and China in others. The U.S. leads in the number of top-tier models and in high-impact patents. China, meanwhile, leads in publication volume, citation counts, total patent output and industrial robot installations. Stanford also highlights South Korea as a standout for innovation density, saying it leads the world in AI patents per capita.
The report places these technical trends in the context of a larger infrastructure race. The U.S. has 5,427 data centers, more than 10 times the number in any other country, and uses more energy than any other nation. Stanford also points out that Taiwan-based TSMC manufactures nearly all leading AI chips, making the global hardware supply chain heavily dependent on a single foundry, even as a TSMC-U.S. expansion began operating in 2025.
While model capability is rising, the report says AI remains inconsistent in ways that researchers describe as a jagged frontier. It cites systems that can earn gold-medal-level results in mathematical competition settings but still struggle with simple tasks, such as reading analog clocks accurately. It also says AI agents improved from 12% to about 66% success on OSWorld, a benchmark that tests real computer tasks across operating systems, but they still fail roughly one-third of the time.
Responsible AI remains a lagging area. Stanford says capability benchmarks are widely reported by frontier developers, but safety and other responsible AI measurements are still far less consistently published. The report counts 362 documented AI incidents, up from 233 in 2024. It also warns that gains in one safety dimension can come at a cost in another, such as accuracy.
On the economic side, the U.S. remains the dominant investor, with private AI investment reaching $285.9 billion in 2025, far ahead of China’s $12.4 billion. The report says the U.S. also led in startup formation, with 1,953 newly funded AI companies. But it notes a steep drop in international AI talent moving to the U.S., down 89% since 2017.
Stanford’s survey data show a widening divide between experts and the public. Experts are far more likely to expect AI to improve work, the economy and health care, while public confidence is more cautious. Trust in governments to regulate AI also varies widely. Among countries surveyed, the U.S. recorded the lowest trust in its own government on AI regulation, at 31%.
The report concludes that AI is spreading faster than earlier general-purpose technologies, but that benefits, risks and power remain unevenly distributed.