The problem: This doesn't give a match:
>>> re.match(r'ä\b', 'ä ', re.UNICODE)

This works ok and gives a match:
>>> re.match(r'.\b', 'ä ', re.UNICODE)

Both of these work as well:
>>> re.match(r'a\b', 'a ', re.UNICODE)
>>> re.match(r'.\b', 'a ', re.UNICODE)

Docs say \b is defined as an empty string between \w and \W. These do match accordingly:
>>> re.match(r'\w', 'ä', re.UNICODE)
>>> re.match(r'\w', 'a', re.UNICODE)
>>> re.match(r'\W', ' ', re.UNICODE)

So something strange happens in my first example, and I can't help but assume it's a bug.

As a work-around I currently use a regex like r'ä(?=\W)'. Seems to work ok.

Also, the \b problem doesn't seem to exist in the \W\w case, i.e. at the beginning of words.

Notice that the re.UNICODE flag is only meaningful if you are using Unicode strings; in the examples you give, you are using byte strings.

Please re-test with Unicode strings both as the expression and as the string to match.

FWIW, the first example works fine for me with and without Unicode strings.

Ok so this does work:
>>> re.match(ur'ä\b', u'ä ', re.UNICODE)

If I understand correctly, I was comparing UTF-8 encoded strings in my examples (my Ubuntu is UTF-8 by default) and regex special operators just don't work in that domain.