Until a few months ago, companies rewarded employees who massively used Artificial Intelligence, consuming huge amounts of tokens, which is AI's basic unit of usage. This tactic now belongs to the past.
Small and large companies in the US have begun to use cheaper models, including some developed in China.
According to the The Wall Street Journal companies across the country are beginning to adopt this radical idea, which can overturn the sector that feeds the global economy.
But why?
Because according to industry experts, companies do not have to spend huge sums on artificial intelligence, but choose the most appropriate tool for their needs. This does not mean that businesses abandon OpenAI and Anthropic supermodels. Rather, many choose to use them only for the most demanding tasks. But for simple ones they use cheaper models.
As Mike Shakes, a former investment banker and Cursor executive today, put it, was acquired by SpaceX: «It's like driving a Lamborghini to go to the supermarket to buy milk while the car is designed to run on track»He said in an interview.
His role is to advise businesses on how they can measure and improve the performance of their investments in artificial intelligence. The company calls this process «tokenomics», referring to the tokens, the basic unit of measurement of the use of artificial intelligence.
Saving resources - or, as they say in Cursor, «tokenomical» approach - it is a spectacular change of attitude.
Anthropic and OpenAI spent years competing for supremacy in artificial intelligence, trying to create increasingly advanced and expensive models.
But now they find that businesses also want simpler solutions. The increasing popularity of these models has overturned the data in the AI market, threatening the high valuations of the two companies, as they prepare for their admission to the stock market.
Geopolitical dimensions
This change also has a geopolitical dimension, as it confronts the world's two largest economic powers.
Generally, the most famous American models are of a closed type, which means they are strictly controlled by companies that develop them. China, on the contrary, is known for the cheapest open-weight models, which can be downloaded and adapted to the user's needs.
Regardless of how this confrontation will evolve, many investors and executives, invoked by the American newspaper, estimate that the era of cheaper artificial intelligence came to stay. Already several American companies have presented or developed new open models.
Silicon Valley companies were the first to start treating artificial intelligence like any other product, seeking the best price and performance relationship. In recent weeks, this practice has begun to spread to other branches.
«I don't see any loyalty whatsoever to any supplier.», says Marty Kausa, CEO of Pylon, a customer service platform with artificial intelligence. «Right now it looks like a real massacre.». Artificial intelligence companies offer their customers deals «Like crazy.» To keep them, he said.
According to Shakes, cost reduction strategies include limiting access to top models for newly hired, as well as using the most advanced artificial intelligence systems to design a job, before taking on the cheapest models to perform it.
This strategy also works in practice.
Recently, Cursor conducted an experiment to assess the cost of creating a web browser from scratch. Completion of the entire project exclusively with OpenAI's GPT-5.5 cost a little more than $10,000. The combination of Cursor's Composer programming model with Anthropic's Opus 4.8 cost just $1,339.
«The best model for a job used to change every few months»Shakes says. «Now it seems to change many times in the same week.»
