BETA
LMArena, formerly Chatbot Arena, ranks AI models by Elo-style scores from blind, head-to-head human votes on real user prompts.
Users submit a prompt, receive answers from two anonymous models, and vote for the better one; millions of such battles feed an Elo-style rating and a public leaderboard. Started as an academic project at UC Berkeley in 2023, it became one of the most-watched model rankings and later spun out as a company.
Its strength is measuring what benchmarks cannot: real human preference on unconstrained prompts, with model identity hidden. Its known biases follow from the same design — votes reward confident, well-formatted, agreeable answers, so style can outweigh substance, and labs have been accused of tuning submissions for arena appeal.
Read it as a preference signal alongside task benchmarks rather than a capability ranking, and note that arena rank says nothing about price: models near the top can differ enormously in per-token cost.
Last revised 2026-07-05 · All glossary terms → · Live AI model pricing →