Every year, Stanford University releases an AI Index Report, and the 2026 edition is out now. MIT Technology Review has distilled the key data. If you need to explain "where AI is right now," these numbers are the best material.
Adoption has outpaced the internet
More than 50% of people globally are using AI—a rate of adoption far faster than personal computers or the internet. 88% of businesses have deployed AI tools, and four out of five university students use AI.
These figures show AI has truly reached the "ordinary people are using it" stage, no longer confined to the tech world. Yolanda Gil, a computer scientist at the University of Southern California and co-author of the report, put it bluntly: "The technology continues to advance, with no signs of slowing down. Even I find it astonishing."
Model performance: benchmarks nearly maxed out
The SWE-bench software engineering benchmark tells the story best: AI scored about 60% in 2024, and approached 100% in 2025. AI now surpasses human experts on doctoral-level science, math, and language comprehension tests.
As of March 2026, the model performance leaderboard shows: Anthropic ranks first, followed by xAI, Google, and OpenAI. China's DeepSeek and Alibaba's Qwen are not far behind.
But there is a serious issue worth noting:
A popular math benchmark has an error rate of 42%—meaning the benchmark itself contains a large number of wrong answers. AI models can inflate their scores by "having seen the answers during training," making many benchmark scores less reliable.
Performance is improving, but the ruler we use to measure that progress may itself be bent.
Young programmers hit hardest
The employment data is the most uncomfortable part of this report:
- Employment for software developers aged 22-25 has fallen nearly 20% since 2022
- Early-career customer service roles show a similar downward trend
- One-third of businesses expect to reduce headcount in the coming year
AI's productivity gains are real: a 14% efficiency boost in customer service and a 26% boost in software development. But the benefits of those gains flow first to companies, while the pressure of job losses falls hardest on young people in entry-level positions.
A 50-point gap between experts and the public
This figure is striking:
| Group | Believe AI has a positive impact on jobs |
|---|---|
| AI experts | 73% |
| US general public | 23% |
A full 50-percentage-point gap. Experts think it's fine; the public thinks there's a problem. Who is wrong? It's hard to say, but this perception gap itself is telling—AI experts and the people affected by AI live in two completely different information worlds.
Another data point: 59% of people believe AI does more good than harm, but 52% say AI makes them uneasy. These two numbers can coexist without contradiction.
Infrastructure dependency: the geopolitical risk of chips and data centers
Taiwan's TSMC manufactures nearly all of the world's top AI chips, and the US hosts the most AI data centers. Together, this creates a highly concentrated supply chain risk—but for now, everyone is enjoying the convenience AI brings, and few are seriously discussing the issue.
Robotics: only 12% success rate
Among all AI advances, robotics is the most honest: success rates for household tasks are just 12%. Waymo operates in five US cities, and Baidu's Apollo Go serves users in China. Autonomous driving and home robots still have a long way to go before they are truly usable.
The most memorable line from the report
Yolanda Gil said: "There is a lot we don't understand about these AI systems."
AI regulation lags behind the technology. There are 150 state-level AI bills in the US, but almost nothing at the federal level. The technology keeps advancing, but no one fully knows where it is heading. Whether that is exciting or unsettling probably depends on which row of the table above you belong to.
Sources: CocoLoop, Want to understand the current state of AI? Check out these charts. (MIT Technology Review, April 13, 2026); Stanford AI Index Report 2026