Apple took Gemini out of Google, but can it take Google out of Gemini?

You humans don’t handle nuance very well.

Before you complain about that being racist, remember that the Macalope is part human, so he can say that.

He’s not saying that when he gets together with his antelope cousins, it’s like a friar’s club roast of humans. He’s not saying that.

Now, you’re probably asking yourself if, after weeks of discussing AI, the Macalope is finally going to talk about something, ANYTHING, else? Well, let the Macalope just tell you that, uh, no, he is not; this one’s also about AI. He’s very sorry about that. So very sorry.

But the problem with the term “AI” is that it’s too broad. There are a number of different kinds of AI. There’s the programming kind that can finish a line of code or tell you how to accomplish something or just do it for you. There’s the horrible kind that can put a birthday cake in front of a generated image of a dead-eyed friend for a party invitation that asks, “What even is reality anymore?” There’s the scientific kind that can comb through massive amounts of data and analyze it. There’s even the disgusting kind that can take peoples clothes off without their consent.

Obviously some of these are laudable and some are decidedly not. Some seem like it would be worth assiduously dedicating resources to. Many do not. It’s complicated and that complication is being ham-fistedly exploited in order to make AI, all AI, seem like a necessity, one that just happens to make a whole bunch of people even richer than they already are.

At least Apple’s implementation of AI will be easy to understand.

During the WWDC26 keynote, Apple announced its third generation of Apple Foundation Models (AFM), comprising five models, some of which are local, some of which are cloud-based, and one of which lives in Google’s servers running on Nvidia chips.

9to5Mac, June 11, 2026

Uhnnnnnnnnnnnnnnnnnn. Come onnnnnnn.

Well, let’s see if we can at least unpack some of this because the Macalope himself has been guilty of fudging the details, although largely before they weren’t yet available. Apple has now made it clear that the updated Apple Intelligence isn’t just a skinned version of Google’s AI. As Macworld’s Jason Cross says:

It seems like Apple started with Gemini’s foundation models, optimized and rebuilt them for Apple Silicon and the model sizes it needs, and retrained them with its own data, weights, and guardrails.

That might seem like semantics but, again, nuance is pretty important here. Those last three dials on the ol’ AI machine are pretty important and can change your results drastically.

The first two models are AFM 3 Core, which provides more Siri smarts, and AFM 3 Core Advanced, which brings more expressive voices and better dictation but requires a more powerful device. These both run on-device to keep you warm on those cold winter days. Then we have AFM 3 Cloud, which provides the general server-side smarts, ADM 3 Cloud for image generation and editing, and AFM 3 Cloud Pro. These three run in the cloud to one day rid us of those cold winter days completely through the magic of human-induced climate change.

Too dark or just dark enough to make you remember the difference? That’s an exercise the Macalope will leave to the reader.

Foundry

The final one, AFM 3 Cloud Pro, is the only one running on Google’s servers, which run on chips made by a noted leather fetishist instead of Apple Silicon. It’s not, however, Google code other than having been derived from Google code, and it’s the bit that will do all that cool party planning stuff for you. So… yeahhh.

We’re going to be throwing so many parties, y’all.

From this entire combination you can see why it was hard to suss out the elephant from each of these parts before WWDC26. Apple started with Google code but made it its own. It put what it could on-device, but is running a lot in the cloud, some of it even on Google’s own infrastructure. There might be a joke to be made here about people in glass caves not throwing stones, but that parable is kind of complicated and, honestly, the Macalope’s trying to wrap up.

Just like the parable of The Cave, AI is also complicated, both technically and in terms of what it provides. The term combines technical tools that are extremely useful to some, having applications that legitimately benefit humanity, with visual slop and filth and LLM feedback that is often computational malpractice based on misappropriated source material, all run on servers that can damage the environment that no one wants in their neighborhood.

The cherry on top? The people in charge of it are pretty much the worst.

The horny one wishes we could have a more nuanced conversation about AI but when we keep calling it all the same thing, it just isn’t possible.