In cases where you're really okay with knowing the concrete class, there's no benefit. However, in many cases you want to be able to write code which only knows about the base class or interface.
For example, look at Iterables in Guava - that's a lot of methods which (mostly) don't care what implementation of Iterable
is being used. Would you really want all that code separately for every implementation?
Where you can code to an abstract base class or an interface, you allow yourself to later use other implementations which share the same public API, but may have different implementations. Even if you only want a single production implementation, you may well want alternative implementations for testing. (The extent to which this applies very much depends on the class in question.)