Arena Intelligence, Inc.

Real-world evaluation platform for AI models

Arena measures how AI models hold up on real work. Each month tens of millions of people put frontier systems side by side on their own tasks, and the preferences they record become the basis for open, human-centred evaluations. AI labs, enterprises and independent researchers draw on those results and Arena's open datasets to understand model behaviour in areas such as agentic coding, creative generation and professional productivity, breaking human feedback down well beyond a single leaderboard ranking. Researchers from UC Berkeley's SkyLab started the company, and the team now includes people from Google, Stanford and DeepMind, with a preference for craftsmanship and curiosity over hierarchy.