In Python, len is a function to get the length of a collection by calling an object's __len__ method:

def len(x):
    return x.__len__()

So I would expect direct call of __len__() to be at least as fast as len().

import timeit

setup = '''
'''

print (timeit.Timer('a="12345"; x=a.__len__()', setup=setup).repeat(10))
print (timeit.Timer('a="12345"; x=len(a)',      setup=setup).repeat(10))

Demo link

But results of testing with the above code shows len() to be faster. Why?

The builtin len() function does not look up the .__len__ attribute. It looks up the tp_as_sequence pointer, which in turn has a sq_length attribute.

The .__len__ attribute on built-in objects is indirectly mapped to the same slot, and it is that indirection (plus the attribute lookup) that takes more time.

For Python-defined classes, the type object looks up the .__len__ method when the sq_length is requested.