Comment by jlkuester7

Exactly. Even 32b parameter models you can run locally on consumer hardware are "good enough" at this point for some workflows!