Physical implementations of General Quantum Computing machines have so far been a bit underwhelming. They may remain so for quite a while yet, though it's always possible we'll see significant improvements.
To the best of my knowledge, 21 is the largest integer yet factored1 using an implementation of Shor's algorithm on a true GQC machine with a program to factor arbitrary integers.
There have been larger numbers factored using Shor's and GQC, at least as great as 4088459, but those are integers of special form, where the factors differ by only a few bits.
There have been larger numbers factored using adiabatic quantum computing (AQC), as implemented by e.g. the D-Wave machine; but AQC has limited application and it's not clear that it offers any real advantage over classical computing, at least for most applications. I mean, if you want to predict how your spin glass will anneal, it could be pretty handy, but you're not using it to break someone's ECCDH key.
In any case, none of these demonstrations is about doing a better job of factoring a number than your six-year-old does. It's about showing that these very preliminary GQC and AQC machines can in fact be used to implement certain algorithms, even if only for trivial inputs.