#ModelWin rate in head-to-head votesWin rate95% CIElo
0%25%50% · even75%100%
Placement in each domain
How the ranking works
Real tasks
Each task is a brief a professional designer completed. The human work is the reference every model is judged against.
Expert judgement
Designers compare model outputs against the reference and against each other, without seeing which model made what.
Win rate
Each vote is one head-to-head: two outputs for the same brief, one winner. A model's win rate is the share of head-to-heads it won.