Arena reaches $100M annual revenue run rate eight months after launch
The UC Berkeley spinout built a billion-dollar business by making AI model rankings something companies will actually pay for
Eight months is not a long time. It is, however, apparently enough time for Arena to go from zero commercial revenue to a $100 million annual run rate.
Arena, which runs the AI model evaluation platform LMArena, announced the milestone after launching its first commercial product, AI Evaluations, in September 2025. The product sells something that sounds deceptively simple: trustworthy, community-driven performance data on AI models, ranked in real time by real users.
From Berkeley research project to billion-dollar business
The origin story here matters. Arena did not emerge from a stealth startup or a founder’s garage. It grew out of LMSYS, a research project at UC Berkeley that built Chatbot Arena in 2023 as an academic exercise in comparing large language models.
The concept was straightforward: show users two anonymous AI responses, let them pick the better one, and aggregate millions of those votes into Elo-style rankings.
The research project spun out into a company in spring 2025. Within weeks, it had closed a $100 million seed round at a $600 million valuation. Within months, it had a commercial product. Within four months of that product launching, it had hit a $30 million annualized consumption run rate.
The $150 million Series A followed in January 2026, led by Felicis and UC Investments, with participation from Andreessen Horowitz and others. The post-money valuation landed at $1.7 billion.
Why AI labs are paying for what used to be free
The free public leaderboard still exists. Anyone can go look at how GPT-4o compares to Claude or Gemini based on millions of head-to-head user votes.
What changed is the commercial layer on top. AI labs and enterprises are now paying Arena for structured evaluations, deeper benchmarking data, and the credibility that comes with community-verified results rather than self-reported benchmarks.
Arena’s platform pulls votes from users across more than 150 countries, across tens of millions of conversations. The revenue model reflects this positioning. Paid evaluations and enterprise services generate revenue, while the public leaderboard maintains the community trust that makes those paid services worth buying.