Arena hits $100 million run rate as AI evaluations business grows

Arena said it had reached a $100 million annualized revenue run rate, eight months after introducing its commercial AI Evaluations product and while keeping its free public leaderboard online. The milestone marks a sharp turn for a system that began in 2023 as a UC Berkeley research project and built its reputation by letting users compare model answers side by side.

The company launched AI Evaluations on September 16, 2025, pitching the service to enterprises, model labs and developers that want deeper performance analytics and human-feedback-based testing. Arena said the business is consumption-based rather than a traditional recurring subscription model, a structure that gives the company a direct financial stake in the volume of evaluation traffic flowing through its platform.

That traffic is substantial. Arena said its site now draws more than 10 million monthly visitors and has hosted more than 700 million total conversations and 82 million total votes. Its public benchmark is built from more than 10 million user evaluations, in which a prompt is sent to two models and users choose the stronger response. Arena also said Agent Mode, its feature for complex multi-step tasks, is already seeing more than 5 million turns per month.

The commercial push comes after a rapid corporate transformation. Arena became an independent company in April 2025, then raised a $100 million seed round in May 2025 at a $600 million valuation, led by Andreessen Horowitz and UC Investments. In January 2026, it followed with a $150 million Series A at a $1.7 billion valuation, with participation from Lightspeed Venture Partners, Felicis Ventures, Kleiner Perkins and The House Fund.

Arena’s rise reflects how central rankings have become to the AI market. OpenAI, Google and Anthropic have partnered with the company to make flagship models available for community evaluation, giving the leaderboard unusual visibility across model releases and product decisions. That visibility has also brought scrutiny: researchers have accused Arena of helping some top AI labs game the system, an allegation the company has denied.

For Anastasios Angelopoulos, the company’s chief executive, the bigger challenge is perception. Many people still treat Arena like an open-source or research project, even as it sells evaluation infrastructure to major AI buyers and scales like a venture-backed software company. The result is a business built on trust, rankings and public legitimacy, now pulling in the economics of a fast-growing AI platform.