Since the new firmware was launched yesterday, I've been putting the updated A9 through its paces with a few torture tests.
To put it briefly, this isn't a firmware update. It's the A9 Mk 1.5.
The basic A9 is no slouch - it keeps up with the 1Dx2 and D5, being better at some things (locking precisely onto eyes, rather than eyelashes or other nearby features) and worse at others (tracking erratic subjects in difficult AF environments vs the D5, e.g. insects or small birds at close distances). But the firmware update takes it to a completely new level, probably surpassing the other two cameras, and just goes to show how much a modern camera is no longer a mechanical or simple electronic device, but a computer attached to an imaging system whose performance is just as dependent on the speed of the processor and the software it's running than on the imaging hardware attached to it.
Not to say that it's perfect. Like any other AF system, it still has the tendency to lock onto the highest-contrast part of an animal's head or body and track that, rather than the eye, although that is largely a function of initial, manual targeting than any issue with tracking - if you manage to target the eye directly and tell it to track it, it will track accurately, but if you just point the cursor at the animal, it will lock onto whichever part of the animal within the square has the highest contrast, which may not be the eye. But this is a feature common to every AF system (AF modules don't know you're targeting a cheetah and want the eyes in focus - they just focus on whichever thing has the highest contrast in the targeting square) and, unlike with other AF systems, should have a solution in a few months' time when animal eye AF is introduced. Eye AF to recognise the eye in the first place, real-time tracking to keep it in focus as it moves. Already, it's not a problem with humans, since the current eye AF system recognises human eyes just fine, and wasn't a problem with the monkeys I tested it with at the zoo today (which are human-shaped enough for eye AF to work). And there are some hardware issues that can't be fixed with firmware - the 5fps limit with mechanical shutter for example, which greatly hinders the usability of EFCS when you want to avoid the electronic shutter's potential for banding when dealing with extremely fast-moving objects. But it's a huge step up from an already-good baseline A9, and a step above the competition, at least AF-wise.
I'm expecting a similar jump in performance when the A7r3 gets its firmware update in April. It won't have the AF tracking update (I doubt the old sensor design allows for it) but will have updates to eye AF, making it 'stickier', and animal eye AF (coming to the A9 in a few months' time), and should make for a good performance update while we wait for the next generation of bodies (probably a 36MP A9II in time for the Tokyo Olympics).