They’re testing the same models on the same task but different ways of organising the swarms and the new approach works better.