Arena, the AI leaderboard everyone uses, is now a $100M business

In an impressive display of growth, Arena, the AI leaderboard platform that began as a research initiative at UC Berkeley, has achieved an annualized revenue run rate of $100 million just eight months after its commercial launch. Known for its widely used crowdsourced AI model performance leaderboard, Arena has garnered over 10 million user evaluations, showcasing its popularity among tech enthusiasts. The platform allows users to input prompts and compare responses from two AI models, enabling them to select the better performer. While Arena offers its leaderboard freely to the public, it began monetizing its offerings in September by introducing AI Evaluations. This new service provides model labs and enterprises with comprehensive performance analytics sourced from its dedicated community of evaluators, who are often eager to try the latest AI models. “Many people are surprised to learn that we are generating revenue; they still perceive us as an open-source project,” remarked Anastasios Angelopoulos, co-founder and CEO of Arena, in an interview with TechCrunch. He clarified that while the company refers to its revenue milestone as ARR (annualized recurring revenue), it actually charges customers based on usage, indicating that the revenue is not consistently recurring. Arena currently faces no direct competitors, especially after the shutdown of Yupp, another crowdsourced AI model selection startup, in March. However, Angelopoulos noted that they compete for funding with human labeling companies like Mercor, Surge, and Scale AI, which help refine AI models post-training. As demand for improved model performance grows, the need for such post-training services has surged. Earlier this year, when Arena announced a $150 million Series A funding round at a post-money valuation of $1.7 billion, its annualized revenue stood at $30 million. In contrast, Handshake’s gross annualized revenue from AI training has nearly doubled since January, reaching almost $1 billion, while Mercor's annualized revenue exceeded $1 billion, up from $500 million last September. Arena evaluates models across various tasks, including text, coding, vision, and image generation, and has recently introduced Agent Mode for complex workflows. Co-founded by Angelopoulos, UC Berkeley postdoctoral peer Wei-Lin Chiang, and Ion Stoica, a prominent UC Berkeley professor and Databricks co-founder, Arena has successfully raised $250 million from notable investors such as Felicis, Andreessen Horowitz, and Lightspeed Venture Partners.