Bairo GonzalezLeandro · Martinez
Start your guided journey The story

In the world · Artificial intelligence · 7 April 2025

AI costs fell more than 280-fold: what the AI Index says

Stanford's AI Index showed that using an AI at GPT-3.5 level became more than 280 times cheaper in two years. What this drop opens up.

bairogonzalez.com team, drawing on Bairo's story · Published

Using artificial intelligence became much cheaper, very fast. According to Stanford University's AI Index 2025, published in April 2025, the cost of running a system performing at GPT-3.5 level fell more than 280-fold between November 2022 and October 2024. Over the same period, hardware costs fell about 30% a year and energy efficiency improved about 40% a year.

What happened

The AI Index is the annual report of Stanford's Institute for Human-Centered AI (HAI). The 2025 edition brought the sentence that became a reference: driven by increasingly capable small models, the inference cost of a GPT-3.5-level system fell more than 280-fold in less than two years, according to the report.

Inference is the cost of using a model that is already built: every question, every answer, every task. It is what users and companies pay day to day, measured in tokens. The figures that accompanied the launch, released by Business Wire and THE Journal, show the other side of the price drop: 78% of organizations said they used AI in 2024, compared with 55% the year before.

The 2026 edition, published in April, showed that the curve continued and that money followed usage. According to the AI Index 2026 highlights, global corporate investment in AI reached US$581.7 billion in 2025. The report also points to the bill that a low price does not erase: energy, water and data centers that grow along with usage.

Why it matters

When the price of a technology falls hundreds of times, who it serves changes. What used to be a big-company project becomes a tool for a small office. A trader, a farmer or a neighborhood clinic gain access to a kind of help that, three years ago, would have cost too much.

But a low price per question is not the same as a low bill. An agent that works on its own for minutes consumes a lot of tokens. That is why the drop in cost does not end the access problem. It changes the problem: instead of "can I afford an AI?", the question becomes "can I afford the right AI, of the right size, for the right task?".

In Bairo's view

Bairo Leandro Gonzalez Martinez says CLAIN was born precisely from this wall. While building the foundation of CEASA Bank without knowing how to code, he understood that AI consumes tokens and that tokens are expensive. Those who most need help are usually those who can least afford it. The name CLAIN comes from claim: the idea that every person can claim access to an intelligence that used to be expensive.

In his reading, the drop in cost measured by Stanford confirms the direction, and niche AIs are the next step. A small, specialist model spends less than an enormous model that tries to know everything. That is why CLAIN is designed as a team of inexpensive, coordinated specialists, with a person deciding at the end. The goal of 142 niche AIs in 142 days, which he announced on October 1, 2026, is a public goal, not a result.

For Bairo, cheap AI only makes sense if it is an extension of the human: more people with more reach, not more machines in place of people.

Where this meets the ecosystem

  • At CEASA Bank, a financial platform for agribusiness that operates on the infrastructure of an authorized institution, the proposal is for agent-made quotes to be cheap enough to serve the trader at a wholesale food market, not only the large exporter.
  • At SOSPROCESSOS, a tool for lawyers under development, the idea came from the same math: analysis of rulings and scenarios at a cost that fits a small office and, through it, people who could not afford a large law firm.
  • At Hospital Triage, under development, the lower cost is what makes it possible to envision an accessible pre-consultation with a clinical assistant, always validated by a professional.

Sources

Read also

More on this topic

In the world: Artificial intelligence