Just eight months after commencing commercial operations, Arena, an AI leaderboard platform that began as a research initiative at UC Berkeley in 2023, has achieved an impressive annualized revenue of $100 million.
Arena has gained prominence for its widely-used crowdsourced leaderboard, which evaluates AI model performance based on over 10 million user assessments. On its user-friendly website, individuals can input prompts that are sent to two distinct AI models, allowing them to select which one performed better.
While the leaderboard features free access for users, Arena started monetizing its platform in September by launching AI Evaluations. This new service provides model labs and businesses with comprehensive performance analytics sourced from its evaluator community.
The swift growth in Arena's revenue indicates that its commercial services resonate with clients just as strongly as they do with its evaluator audience, who often flock to the platform for early exposure to the latest AI models that may not yet be publicly available.
Co-founder and CEO Anastasios Angelopoulos shared with TechCrunch that many observers are not aware that Arena is generating revenue, with some still viewing the venture primarily as an open-source project. Although Arena refers to its $100 million milestone as annualized recurring revenue (ARR), Angelopoulos noted that the company's revenue structure is based on “consumption,” meaning it does not follow traditional recurring revenue patterns.
While Arena currently faces no direct competitors—having outlasted crowdsourced AI model competitor Yupp, which ceased operations in March—Angelopoulos acknowledged that the startup competes for the same financial resources as human labeling companies like Mercor, Surge, and Scale AI, which support model developers in enhancing their AI post-training.
As AI firms push to improve model performance, demand for post-training optimization services is rapidly increasing. Earlier this year, Arena secured a $150 million Series A funding round at a post-money valuation of $1.7 billion, with its annual revenue reported at $30 million at that time.
In a related development, Handshake’s revenue from AI training has almost doubled since January, rising from $550 million to nearly $1 billion, as reported by The Information in April. Additionally, Mercor surpassed an annualized revenue of $1 billion earlier this year, doubling from the $500 million reported last September.
Arena evaluates AI models across various tasks, including text processing, coding, image generation, and complex workflows, thanks to its newly launched Agent Mode.
Alongside Angelopoulos, Arena was co-founded by fellow UC Berkeley postdoctoral researcher Wei-Lin Chiang, who serves as the startup's CTO, and Ion Stoica, a distinguished UC Berkeley professor and Databricks co-founder who provided guidance during the project's initial stages prior to its establishment as a formal company in April 2025.
To date, Arena has successfully raised $250 million in funding from a range of investors, including Felicis, Andreessen Horowitz, The House Fund, LDVP, Kleiner Perkins, Lightspeed Venture Partners, Laude Ventures, and UC Investments.
