The thing is that Euclidean TSP needs a lot of data to encode hard instances.

N=15 was even considered solved in the 60s, and N=20 has never been considered large instances, especially not of Euclidean TSP.

I cannot see how anyone could say 13 is the max: you need 100k memory slots and 1M comparisons. This has been trivial for quite some time.