The arena platform has reached $100 million in annual revenue.
The Arena project, one of the world's most influential platforms for evaluating artificial intelligence models, has reached an annual revenue of $100 million just eight months after launching its commercial operations, according to Zamin.uz.
Founded in 2023 by students at the University of California, Berkeley, the startup quickly evolved from an open-source research initiative into a major technology business, as reported by TechCrunch.
The Arena platform is widely recognized for its public leaderboard, where users submit prompts and compare responses from two different models to determine which performs better.
To date, over 10 million such evaluations have been conducted on the platform, making it a key benchmark for objectively assessing AI capabilities.
While the public leaderboard remains free, the company began generating revenue in September of last year with the launch of its AI Evaluations service.
According to TechCrunch, this service provides deep analytical insights to major labs and corporations on how their models perform in practice.
As Arena’s founder and CEO Anastasios Angelopoulos noted, many still view the project as a non-profit, open-source platform—but real-world metrics tell a different story.
The company’s $100 million annual revenue run rate marks a significant milestone: just a few months ago, in January, Arena raised funding at a $1.7 billion valuation with an annual revenue of only around $30 million.
This represents more than a threefold increase in revenue over a short period.
Today, Arena competes in the market with large data-labeling firms such as Scale AI, Mercor, and Surge, which rely on human annotators to prepare training data.
AI developers increasingly depend on such services to refine their models after initial training. For example, companies like Mercor and Handshake, active in this space, are already valued at around $1 billion in annual revenue.
The Arena platform ranks not only text-based models but also those in programming, vision, and image generation.
Recently introduced, the Agent Mode feature enables evaluation of complex, long-running agent workflows.
For users and developers in Uzbekistan, the platform remains one of the most trusted sources for identifying which AI model is currently the strongest.
