From the Cornell poker odds calculator, 1,000,000 simulated hands:

AKs (47.84%), 66(52.16%)
AKs (48.24%), 55(51.76%)

Looks pretty accurate to me. The chance that A234 comes in 5 cards is pretty slim, so that's not enough to push 55 better than 66.

Incidentally, in case you weren't already aware, the small differences in the probabilities we got can be explained by the fact that both of these are Monte Carlo simulators -- that is, they run X number of simulated hands rather than being exhaustive searches.