It’s almost like they’re interchangeable. We need to start asking these models to solve extremely difficult, contrived DSA coding questions before deciding which ones we employ