Arena Intelligence (formerly LMArena / Chatbot Arena) runs the internet's most widely used platform for evaluating AI models. The product is simple: users type a prompt, two anonymous AI models answer side-by-side, the user picks the better one, and only then are the models revealed. Millions of these "blind taste tests" are aggregated into a live leaderboard ranking 400+ frontier models by real-world human preference — not curated test sets. AI labs (OpenAI, Google, xAI) pay for a commercial tier called AI Evaluations, where they submit specific use cases — software engineering, law, medicine — and get crowd-driven feedback that tells them how their model performs against competitors in practical tasks. The free leaderboard builds trust and drives traffic; the paid product converts that trust into revenue. Founded in 2025 as a spinout from UC Berkeley's Sky Computing Lab by CEO Anastasios Angelopoulos, CTO Wei-Lin Chiang, and professor Ion Stoica.