WMArena is an open, human-preference World Model Arena. It ranks world-model AI by blind human votes — starting with the most productized category, video generation.
Automated video-quality metrics correlate poorly with what people actually find convincing. WMArena measures the thing that matters directly: human preference. Two anonymous models render the next moment of the same scene, people vote blind, and the votes become a transparent leaderboard. It's the same method that became the reference for ranking language models, applied to world models.
Today: image-to-video world models — "renderers," in the functional taxonomy of world models. The leaderboard reflects live blind votes. As other world-model categories productize, the arena extends to them. Read the methodology for how the rankings are computed.
The ranking method is transparent and the data is meant to be referenced. WMArena aims to be the citable source for "which world model do people actually prefer."