The regex [^{}]*public\s+(final)?\s*class\s+(\w+).* does indeed not match the full source code mentioned, it only matches a substring (that's why on regex101 it's accepted). If Codeforces tries to match the full source code, which is what Pattern.matches does in Java (the language the check was likely written in), then it evaluates the regex correctly in this case.

It also seems like Codeforces uses this regex to "parse" your code to determine the correct filename to use, etc., which can cause even more problems (using regex to decide a non-regular language isn't a good idea). I think the check should be removed or done in a different way.

This regex has to match the entire source code, not just some substring of it. In python terms, codeforces uses re.fullmatch whereas the regex site you linked uses re.search. This particular regex simply doesn't handle comments at all.

It's worth noting that you can annul the entire regex check with a comment that looks like a class declaration:

// public class Huzzah! {

/*
 }
*/
public class A {
    public static void main(String[] args) {
    }
}

Who needs Java, lol?