In Python, when are two objects the same?

It seems that 2 is 2 and 3 is 3 will always be true in python, and in general, any reference to an integer is the same as any other reference to the same integer. The same happens to None (i.e., None is None). I know that this does not happen to user-defined types, or mutable types. But it sometimes fails on immutable types too:

>>> () is ()True>>> (2,) is (2,)False

That is: two independent constructions of the empty tuple yield references to the same object in memory, but two independent constructions of identical one-(immutable-)element tuples end up creating two identical objects. I tested, and frozensets work in a manner similar to tuples.

What determines if an object will be duplicated in memory or will have a single instance with lots of references? Does it depend on whether the object is "atomic" in some sense? Does it vary according to implementation?