Arena is a community-powered platform for evaluating AI model performance in real-world conditions. Created by researchers at UC Berkeley and formerly known as LMArena, the platform hosts public leaderboards that rank frontier models based on actual usage patterns and community feedback. Millions of people use Arena monthly, with tens of millions of builders, researchers, and creative professionals accessing the platform to test models and contribute evaluations.
The platform's core mechanism relies on human-centered feedback. Users interact with frontier models through Arena's interface and provide direct assessments of model responses, which feed into performance rankings and reliability evaluations. This community-driven approach aims to ground model assessment in practical use cases rather than abstract benchmarks alone.
Arena's leaderboards and evaluation data are used by leading AI labs and enterprises to understand model reliability, alignment, and real-world impact. The platform positions itself as a source of transparent, rigorous evaluation for the broader AI community, influencing how the field measures progress and understands model capabilities.
