Naturally if you're dealing with big arrays/tensors, numpy is the best choice for operating on sequences.
However, ndarrays have downsides for certain use cases - as ndarrays are fixed size, adding elements is very slow, also they don't support functional methods (or rather you have to create a new array every time you apply e.g. a map), and ndarrays of any other type than numbers doesn't really make sense.
Many of the methods are wrappers for built-ins, but I find the syntax of Arrays cleaner than the weirdness of the builtins.
For example, while applying an async "starmap" to an Array is just a method call, with built-in lists you would have go through the whole hassle of importing both ThreadPoolExecutor and starmap, creating an executor, scheduling the function, and finally converting the result back to a list.