LMArena
A public leaderboard that ranks AI models by asking people which of two anonymous answers they prefer.
LMArena, originally Chatbot Arena, ranks models by popular vote. You get the same prompt answered by two models without being told which is which, you pick the better one, and thousands of those votes get turned into an Elo rating, the same system used for chess. There are separate arenas for text, image generation and image editing.
Its strength is that nobody can study for it, unlike a fixed benchmark where questions can leak into training data. Its weakness follows from what it actually measures: which answer people prefer at a glance. That rewards confident, well-formatted, agreeable output, and says little about whether the content is correct. Small gaps near the top are mostly noise, so a model in second place by 20 Elo points is not meaningfully worse than the one in first.