Arena AI evaluations hit $100M ARR eight months after launch

Eight months after switching on a paid product, an AI leaderboard project that started as a Berkeley research effort is reportedly running at $100 million in annualized revenue. That is the headline number from a TechCrunch report on Arena, the team behind the crowdsourced model rankings that nearly every AI lab cites at launch time.

The growth curve, as reported, is sharp. Arena's commercial AI Evaluations service, which sells deep-dive performance analytics drawn from its community of users, launched in September 2025. When the company raised a $150 million Series A at a $1.7 billion post-money valuation in January 2026, annualized revenue was $30 million. Several months on, that number has more than tripled, with $250 million raised in total. Co-founder and CEO Anastasios Angelopoulos told the publication that "a lot of people don't even understand that our business is making any money at all; people still see us as an open source project."

The honest caveat sits in the same article. Angelopoulos clarified that Arena charges for consumption, so the ARR label is not annualized recurring revenue in the classical software sense. It is a snapshot of usage extrapolated forward, and usage can move in both directions if labs cut evaluation budgets or pull the work in-house. The reporting also doesn't break out customer concentration or margin, both of which matter for a number this big this fast.

Why this matters if you are picking a model rather than training one: the same leaderboard that has shaped public perception of frontier systems is now selling private slices of that signal back to the labs being ranked. That is a useful product, and it is also a structural tension. Arena lines up against Scale AI, Mercor and Surge in evaluation and post-training work, where the question of who gets to grade the test is becoming a real business in its own right. The thing worth watching is whether independent benchmarks stay independent once they are also a paying customer relationship.