Here’s How Many Gemini Elements Apple Intelligence Actually Contains.
Apple spent a lot of time at WWDC 2026 talking about its updated Apple Intelligence platform and new Siri AI app, and in the past few days, new details have emerged about how Apple and Google’s partnership on AI models will impact the new software—but the question of how much of Siri AI is Apple’s and how much is Google’s remains a complex one.
Back in January, we received official news that Apple would use Google’s Gemini AI models to power Apple Intelligence, that the deal would last for several years, and that Apple’s “industry-leading privacy standards” would be maintained.
Neither Apple nor Google explained details at the time about how exactly this partnership would play out, but it was clear that it was more significant than Apple’s previous deal with ChatGPT, which simply ignored requests Siri couldn’t answer.
I suspect there was a heated debate within Apple about whether to strike a technology deal with a sworn competitor, even if it meant catching up faster in AI. Ultimately, CEO Tim Cook and his colleagues decided it was worth it—and we’ll have more details after WWDC 2026.
Siri AI is not a Gemini app…
Moving on to the WWDC 2026 keynote , Apple’s senior vice president of software engineering, Craig Federighi, described the significantly improved Apple Foundation Models (AFM), developed through “close collaboration” with Google. According to Federighi, Apple “leveraged” the technology behind the Gemini models to create the artificial intelligence that now powers Siri AI and other new Apple Intelligence features.
And Gemini’s impact is certainly noticeable: Apple’s AI is now truly multimodal, capable of processing audio, voice, and text, and is much better at creating custom text and images. Image editing is significantly improved—indeed, very similar to the Nano Banana 2 —and Apple’s AI now has a much better understanding of the world around it, another area where Gemini models outperform the competition.
Ultimately, however, these are still Apple’s own models. Regarding the native models, we have the built-in AFM 3 Core and AFM 3 Core Advanced models, which are installed on iPhone, iPad, and Mac—though the Advanced version (the one that allows you to adjust Siri’s tempo and expressiveness) requires the latest hardware. According to Apple , AFM Core Advanced requires an iPhone Air, iPhone 17 Pro or Pro Max, an iPad M4 or later with at least 12GB of RAM, or a Mac M3 or later with 12GB of RAM.
In subsequent comments (via 9to5Mac ), Federighi said, “We don’t have a Gemini app as our app.” This means Apple Intelligence doesn’t use Gemini’s AI models, the client code that users of the Google Gemini app receive, or the knowledge base built from Google search. All of that work was done by Apple.
There’s no doubt that Apple needed Gemini’s AI models to take its own models to the next level so quickly, but Apple executives are understandably reluctant to make too much of the partnership, for the same reasons they don’t talk about the billions of dollars Google pays annually to remain the default search engine in Safari.
Federighi did have time for a few caustic remarks, however: “It seems some are racing ahead, developing AI for the sake of AI, without paying enough attention to the people—all of us—it’s ultimately meant to serve,” he said, alluding to competitors who have left his company far behind in AI (while Apple paid $250 million in compensation for promised AI features that never arrived).
…but Apple uses some Google servers.
This includes the Apple Foundation’s local models, which ensure the privacy and security of all your requests and data on any device you use. Cloud models are a bit more complex—they are used to solve larger, more complex problems that can’t be solved solely on-device.
As Apple’s team describes (via Ars Technica ), AFM 3 Cloud is designed for general use. There’s also ADM 3 Cloud for image generation and AFM 3 Cloud Pro for “more complex” queries (and, apparently, for getting started with agents). Like the on-device models, they utilize some of Gemini’s capabilities at the most basic levels, but with Apple’s own developments and refinements.
The first two models run on Apple servers, but requests sent to AFM 3 Cloud Pro are routed elsewhere: to Google data centers, where they are processed by Nvidia GPUs. However, according to Apple , these data centers will be subject to the same security measures as Apple-managed data centers.
This means no data is stored (it will be deleted after the request is processed), no one else can see it (not even Apple or Google), and your identity is hidden. Apple allows third-party security auditors to verify its PCC code, and this applies to both AFM 3 Cloud Pro models and Google servers.
There’s one small caveat: Apple states that “PCC on Google Cloud will gradually roll out a full suite of protections over the summer preview period.” So, if you’re using one of the developer betas , some of your most complex AI queries may not yet be as fully protected as you’d like.
Apple promises to provide more technical details as it gets closer to the full launch of new software updates and Siri AI, but from what we’ve heard, the company seems to have found a balance between improving its AI with Gemini and maintaining all the signature Apple features users have come to expect.