Subscribe to our newsletter

Stay updated with the latest news and trends.

No spam. Unsubscribe anytime.

Browse

Categories, quick links and the newsletter.

Quick Links

Newsletter

Get the latest updates

Subscribe to our newsletter to stay informed about trending topics.

No spam. Unsubscribe anytime.

T
Tech

DeepSeek's New Model Tops Global AI Usage Charts as Chinese Models Sweep the Top Five

DeepSeek's newest model has become the most-used AI model on OpenRouter, the platform developers use to route requests among competing models, with weekly call volumes up 801% in a single week — and Chinese models now fill the entire global top five.

DeepSeek's New Model Tops Global AI Usage Charts as Chinese Models Sweep the Top Five

DeepSeek's newest model is now the most-used AI model in the world, by one telling measure. On OpenRouter — the platform many developers use to route their requests to whichever model they prefer — the Chinese lab's new release has taken the top spot, with weekly call volumes up 801% in a single week. The four places below it are also held by Chinese models.

The ranking matters because of what it measures. OpenRouter traffic is paid usage: developers spending real money on real workloads, not models graded on a benchmark. That distinction is the framing of the question now sitting on Zhihu's hot list — how did DeepSeek's new model top global weekly calls with an 801% surge, and what does it mean? — and of the Hong Kong coverage noting that Chinese AI now holds the entire OpenRouter top five.

A phone screen showing DeepSeek's chat interface: "Hi, I'm DeepSeek." Photo: HK01
A phone screen showing DeepSeek's chat interface: "Hi, I'm DeepSeek." Photo: HK01

The mechanism behind the surge looks less like a new model than like old-fashioned efficiency. Last week DeepSeek gave away its whole playbook, revenue math included, open-sourcing DeepInfer, its inference engine — and settling a long-running industry argument about how it sells tokens so cheaply. The figures circulating on Weibo are striking: a single cluster processes 155 trillion tokens a day, roughly 1.8 billion tokens every second, around the clock. An individual GPU brings in about 38 dollars an hour in off-peak revenue against running costs of five to ten dollars an hour. As one widely read post put it, the industry's two-year narrative was that whoever piles up the most chips wins — and DeepSeek's answer, proven in the numbers, is that clever systems engineering can beat brute-force money.

The surge has a cost, and DeepSeek has already paid it once. Scaling to global traffic leader strained the service badly enough earlier this month that the company refunded users after outages. Holding the top spot means running that 155-trillion-token-a-day machine without another failure — under a load that did not exist two weeks ago.

The timing lands on receptive ground. Chinese official media have spent the week celebrating the country's "new momentum" in industry, with data and artificial intelligence named as its core drivers — and DeepSeek's week, in which a Chinese model processed more paid requests than any other model on earth, is what that momentum looks like when it shows up on a global leaderboard.

OpenRouter's rankings reset weekly, so the table is never settled for long. But for an industry that has spent three years measuring Chinese models almost entirely by benchmark scores, the measure that has flipped this week is the one that carries invoices.