And the worst part is, these people don't even use the flagship thinking models, they use the default fast ones.