If the number of bits is constant, then any polynomial time based on it is constant.

Big-O statements (like an algorithm taking constant or O(1) time) are statements about asymptotic behavior, i.e, how the function behaves in the limit (usually, as input size tends to infinity). If you don't look at them in the limit, then big-O-ish language (like constant time) is meaningless.

How meaningless? Even undecidable languages have a constant time "algorithm" if you consider the input size to be held to a constant. So without viewing things in the limit, all problems become computationally equivalent in the asymptotic language.