Blog
The world of AI: What has changed in 2026
A brief history of artificial intelligence, the collapse of the price of inference and the saturation of benchmarks - based on the numbers of the Stanford AI Index 2026.
2 min read
The hype around artificial intelligence is hard to understand without the numbers. Here's a look at the state of things in 2026, based on key data from Stanford University's annual AI Index report.
A brief history
The term "artificial intelligence" appeared in 1956 at the Dartmouth conference. In 2012, the AlexNet neural network won the image recognition competition by a wide margin, ushering in the era of deep learning. And in 2017, the transformer architecture was proposed - all today's language models are built on this basis.
The biggest change: the price
Most models see a "step up", but the most noticeable change is the price. The cost of using a GPT-3.5 level model (inference) has decreased approximately 280 times since the end of 2022.
The drop in price while the quality remains the same is the result of competition between providers, optimization of infrastructure and cheaper chips.
In practice, this means that a service that could only be afforded by large companies two years ago is now available to a student project.
Benchmarks are saturating
Tests designed to evaluate models are being "unloaded" faster than expected. On the specially designed Humanity's Last Exam, the models improved by 30 percentage points in one year. Important for developers — SWE-bench Verified: the result on solving real GitHub problems approached human level in one year.
The invisible side: infrastructure
AI requires power. The United States has 5,427 data centers—ten times more than any other country. Data center capacity allocated for AI reached 29.6 GW. That is, the conversation about artificial intelligence is actually turning into a conversation about energy.
Summary for the student
Knowing how to call a model is no longer a special skill—it's becoming a common skill. What makes the difference is being able to state the problem correctly, verify the result, and build the system cheaply and sustainably. Therefore, mathematics, algorithms and system design do not lose their value.
Sources: Stanford HAI — AI Index 2026, Department of Research and Development.