Sorry, my mistake, it is 1.6 EV more DR than the 5D3, I corrected my post above. They are going to do some noise analysis too I believe.
Canon hasn't really improved the high ISO ceiling much over the past few releases (look at the 5D3, 6D, 1DX and 1DX2). Perhaps they have been fully concentrating on the shadow detail retention and reducing the low signal noise, and there might not be much more they can do at the high end? It has been several years, and we have seen only very gradual improvements at the high ISO.
That being said, if I can get keepers from ISO 25600 or 32000, I am not sure I need much more high ISO for what I shoot. I would love to start working on my NR methods for the 5D4, as soon as somebody provides a few real-life raw files.
From the comparison tools I've seen so far, it does look as though there's a small high ISO improvement, so that's good.
The low ISO DR improvement seems ballpark with other reported sources.
Linked from the FM thread is this interesting bit of info (https://www.dpreview.com/forums/thread/4047260?page=2); essentially there's some musing that the nature of the Dual Pixel raw files may mean that you have an extra stop of data, because it contains the summed pairs of pixels (like a normal image) but also the data from just one of each pair (i.e. half the sum, if each pair of pixels actually contained exactly the same value). Time will tell if that's valid, but it'll be good to see what people may be able to pull from the files.
I do even wonder if it'll be possible to extrapolate further depth information than is actually present, but that's bordering on wild speculation on my part.