Not the parent poster, but basically, they dropped RGB subpixel rendering, and now use plain anti-aliasing for all font sizes. They may have changed hinting and anti-aliasing parameters, too. No difference on a HiDPI screen, since you don't need subpixels there, but on standard definition screens many find it looks a bit blurrier (me included).

Supposed benefits are that it is easier to render text on colored backgrounds (you can just render with alpha to a buffer and combine it with the background, with RGB subpixel rendering you have to consider the background while drawing), and you can scale text without color artifacts.

To be honest, we have plenty of computing power today, and our GPUs are underutilized most of the time anyway. I wonder why we can't just have RGB subpixel aware rendering of everything (not just text, but also lines and polygons) implemented as a kind of shader pass. Render everything at 3x horizontal resolution, and merge using an appropriate shader. And if you have a HiDPI screen, just drop the shader.

It's actually not possible to accurately composite subpixel AA using the GPU since you can't configure blends per-channel. To do it accurately, you need three different alpha channels.

It is actually possible to do using a feature called dual source blending, which unfortunately is so rarely used that drivers tend to fail to implement it properly. But it does give you the possibility to use 3+ alpha channels with full hardware acceleration, at some memory cost.

Though since DPIs are going up and 3 alpha channels is such an unusual thing, I guess they just figured it wasn't worth the trouble.