Comment by simonw

Because it's one of the few ways of comparing models that lets you instantly evaluate them visually. That makes it more comprehensible than a numeric score on a benchmark.