You can kinda get a feel for what they're good at, if you get what I mean?

Even the big online models have very specific styles and preferences for similar tasks. You can easily test this by giving them all some generic task without too many limits, each of them will gravitate towards a different solution to the same problem.