PyXLL embeds python, and so runs python in-process (that is, in the same process as Excel, without spawning a child process).

PyXLL fully supports UDFs (user defined functions) in a fast, efficient way. Calling a pyxll function (UDF written in python) calls into the in-process python interpreter and so there is no cost of starting up a new python process or calling into an out of process COM server.

Excel and Python types are converted to and from each other in optimized C code, depending on the argument and return types specified for the function (UDF). The type conversion is user extensible and so can handle automatically converting Excel ranges to and from pandas DataFrames (just one example).

The full Excel API is accessible from pyxll using the well known and established win32com package.

PyXLL works with Python 2 and 3, 32 bit and 64 bit (all versions from python 2.3 and 3.4).

PyXLL supports newer features of recent Excel releases like multi-threading and asynchronous functions, while maintaining compatibility with older versions.

PyXLL can expose python functions as Excel macros and menus as well as UDFs.

If you want to script Excel from a process outside of excel (eg IPython notebook) you can use win32com very easily without the need for any other package. But, if you want to have a python interpreter running inside Excel to be able to write performant UDFs as an alternative to writing VBA or C addins as far as I'm aware PyXLL is still the only package capable of doing that[1].

PyXLL has commercial support and is partnered with a major python training and consultancy company, Enthought[2].

[1]DataNitro claims to be able to do with, but according to their docs you are limited to only being able to run one Excel instance at a time, which for most real world users is too limiting in my experience.

[2]https://www.enthought.com/products/pyxll/