What Arena does
Arena (formerly LMArena) is a community-powered AI evaluation and benchmarking platform that turns large-scale human feedback into public model leaderboards. Users interact with “Arenas” (e.g., Text, Search, Code, Vision, and Agent-mode experiences depending on the track) where they compare model outputs side-by-side and vote on preference; Arena aggregates those votes into leaderboard rankings using an open, published evaluation methodology.
More
Arena positions its core differentiator as using real-world, human preference signals (including enterprise/professional use cases) rather than relying only on static benchmark scores, and it emphasizes transparency and open-sourced components of its evaluation pipeline. Business model: Arena charges for enterprise-grade evaluation services and related analytics for model labs and companies, while keeping public participation in the core community evaluation experience accessible. In its Series A announcement, Arena describes revenue as coming from paid evaluation services sold to AI labs and enterprises. Arena also periodically releases datasets/research derived from community interactions to support transparency and academic work. Strategic position (as of 2026): Arena has expanded from a public research/community leaderboard into an enterprise evaluation product and supporting infrastructure for “live” model comparison. It reports substantial scale in monthly users, conversations, and votes, and it continues broadening evaluation coverage into new modalities and agentic workflows (e.g., Agent Mode) while adding additional ranking dimensions such as factuality in relevant arenas.