See this answer.
I'm not sure how to deal with this answer. That's the shorter code at this time, so it may be the winner.
On one hand, the tokenized version is sort of compressed code, but it seems to be totally unreadable by an human. So I'm not sure if counting the characters of the tokenized version is fair. On the other hand, it's a feature from the language so it doesn't break any rule.
What do you think? Are there other languages which can be compressed on similar ways?
I haven't seen this discussion on meta yet so I'm posting it, it may be useful if somebody asks the same thing.