Arena AI

Arena is a free, open benchmark platform where you compare frontier AI models in blind side-by-side battles. You submit a prompt, two anonymous models answer, and you vote on the better one. Millions of those votes produce Elo-style leaderboards spanning text, coding, math, creative writing, vision and image generation across 360 or more models. The project began as Chatbot Arena at UC Berkeley, became LMArena, and rebranded to Arena on 28 January 2026. Read the leaderboard as a measure of human preference, not correctness. Votes reward answers that feel better, which rewards formatting, length and tone as much as accuracy, and the widely discussed Leaderboard Illusion critique argues that providers with private testing access can inflate their standing. Arena also tells you nothing about latency, cost or rate limits. Use it to narrow a shortlist, then test those models on your own prompts before committing to a subscription.