I'm not usually prone to hyperbole (not on POTN at least), so... disclaimers and context before I come over all hyperbole fanboy...
As previously mentioned, my left side bank of cross type points are significantly front focusing. AF is consistent - but obviously I can't MFA correct just one set of points so I assume some part of the optical path to the AF sensor isn't parallel to the image sensor (it's going for a fix next week).
I got some time for what I'd consider the camera's first "proper" use today - namely getting some shots of my toddler running around, on a swing, and a slide. I've done exactly the same "shoot" using my 5D3 + 24-70II so this was a pretty good comparative test.
Checking my 5D3 shots, the swing shoot (positioned right in front - so a pretty hard test of subject moving back and forth, fairly close, at around 35mm focal length) was about 2 stops brighter (i.e. today's conditions had 2 stops more ambient light).
For the slide (much later in the day) we were actually nearer 3 stops lower ambient vs the 5D3 shoot - though still reasonable light (f/4, 1/1000, ISO 800).
I mention the above because obviously the AF performance will be highly dependant on available light (not the aperture, shutter speed and ISO chosen for the actual exposure).
I'd also been shooting some general "running around" shots at f/3.5, rather than my usual f/4-f/8. The previous slide shots were at f/5.6, this time at f/4. I mention that as shallower DOF will be less forgiving of misfocused shots.
Finally, I know my technique needs work when shooting things that move (long story).
So...
<HYPERBOLEFANBOY>
Holey shizzle. This thing is incredible!
On paper there doesn't really seem to be much difference vs the 5D3, but in actual use the difference in keeper rate is incredible. I have never - under any circumstances or lighting conditions - got anything near the hit rate that I've seen from this camera today.
It's the one thing that bugged me before - I'd get a burst of shots, with quite a few usable, but you could guarantee that the ones with the "perfect" expression would be the percentage where focus was missed. The rate of unusable frames here is so low I've come away with multiple good images per "scene".
I'd note that I'm also pretty picky about what I consider a hit - DPReview's 5D4 review has a number of shots that (as far as I read the writeup anyway) are claimed to be good - but are way below what I'd call sharp. I probably am being a pixel peeper TBH.
Perhaps my sucky technique means that a more skilled shooter would see a smaller improvement (because they'd be getting a higher hit rate with the 5D3), but, regardless, this body has way surpassed my expectations when it comes to AF.
I also got chance to snap a couple of landscape/architecture shots in the area, and there are a couple that are clearly high DR. I haven't edited them yet, but knowing that I won't see the shadows falling into a mess is a huge bonus. Ironically it (low ISO DR) was my main "want" for the 5D4; but the subtle (but clear) high ISO improvement and stunning AF really just make it complete.
It's almost enough to make me forgive the abomination they've made of the BG-E20 grip 
</HYPERBOLEFANBOY>
Right... back to levelheaded analysis and data...