What games like AC unity, second son, TLOU, most unreal games, killzone etc demonstrate is just how much mileage baked lighting gives you. But from a production stand point, you make a huge trade: higher quality and accuracy, often simpler/faster to render (with baked lighting) but dramatically increased asset sizes and an enormous impact on iteration times within the production pipeline. You really need a well sorted pipeline to attempt a project with primarily baked lighting.
If a game design can accept fixed times of day (like those games mentioned) and you are willing to throw massive computing resources at the problem (baking lighting at that quality is an enormous computational challenge) then it's certainly worth it.
I suspect in the case of unity, the lighting isn't stored as lightmaps. More likely very dense probe volumes heavily compressed (denser near surface geometry). Hence sensible lighting on the characters, but no especially obvious lightmap seams, pixilation or texture compression artifacts. Also lightmaps *really* don't scale well with modern geometry density. I've seen evidence in game to suggest individual blocks of buildings are baked separately, which would make a lot of sense.
For the interiors, most of the shiny rooms are very rectangular. I believe they render out a cubemap from the centre of each room, projected onto a box to roughly match the walls. This is a pretty common trick for baking reflections.
Even with that, I believe one of the developers said that lighting data accounts for half of the game install.
The really impressive games, technically, will be those that can achieve this kind of lighting fidelity without baking, with dynamic time of day, etc. Frankly that's probably still a long way off.
Those games at the cutting edge are still dealing with an order of magnitude lower lighting resolution. For example arguably the most advanced and impressive, the tomorrow children's GI system uses three 32×32×32 cascade volumes which it generates in ~ 1/100th of a second - there is no way that could possibly compare to a baked system that might take a day to render over several hundred servers.
Gotta build a rig with decent upgradeability since I have a feeling that building a new one right in the midst of this transitional period will result in the new rig underperforming when games get even more technically impressive after this. I'm pretty forgiving with console games but I play on pc to get the high end experience at 60fps.
