The Site Where People Decide Which AI Model Is Best Is Worth 3.1 Billion Dollars: Arena Almost Doubled Its Valuation in Ten Months

In 2023 it was a research project at Berkeley: people compare the answers of AI models and vote, and the votes are turned into a leaderboard. Today Arena is worth 3.1 billion dollars (around 2.7 billion euros), after raising 200 million dollars (around 170 million euros) in a Series B round.

The jump is fast even by the standards of the AI fever. In January the company raised 150 million dollars at a valuation of 1.7 billion, when its annual revenue run rate was 30 million dollars. In June it announced it had reached 100 million. Ten months after the January round, the valuation has almost doubled. The new round is led by Lightspeed Venture Partners and Khosla Ventures, with Salesforce Ventures, a16z, Felicis, Dell Technologies Capital, 01 Advisors and Endeavor Catalyst also in.

The business model is worth a closer look. The leaderboard is free for ordinary users - the company says it has tens of millions of visitors a month. The money comes from the other side: since last September Arena has been selling AI labs and companies analytics on how their models fare with the public. So the labs whose models are ranked pay for the ranking data. Who, then, grades the grader?

The company is also opening a new category, for model "alignment". It measures whether a model takes actions without permission, falsely attributes sources and claims to have finished a task it hasn't actually finished. In the preliminary ranking a group of OpenAI models leads, Anthropic's Claude Opus 5.5 is sixth and Claude Fable ninth.

"Artificial intelligence is advancing faster than our ability to evaluate it, and static tests fail as soon as models realise they're being tested," the company says. An accurate diagnosis. Except that the whole industry's problem is now also the product of a company worth 3.1 billion dollars.