__sub__ is a special command that just replaces:
object = arg1 - arg2
,where arg 1 and arg 2 are Vector3 types, with:
object = arg1.__sub__(arg2)
which makes sense since the first object is self. Basically it mean if you have 2 vectors then you use normal mathematical notation to do it.
The docs are somehow auto generated form a strongly typed language such as c++, and just create some text that happened to be there.