2026-06-08 — views

Apple rebuilds Siri on new Foundation Models at WWDC 2026, with a Google Gemini assist and a 12GB hardware floor

Read this because Tim Cook's last keynote pairs a genuinely rebuilt Siri with two asterisks practitioners should not skip: a 12GB-unified-memory device floor and EU/China launch exclusions.

WWDC 2026 unveiled Siri AI on next-gen Apple Foundation Models built with Google's Gemini, a dedicated app, an expanded developer framework, and EU/China gaps.

Apple rebuilds Siri on a new on-device + cloud model stack

At WWDC 2026 on June 8, Apple previewed the next generation of Apple Intelligence and an entirely rebuilt assistant it now calls Siri AI. The pitch: a more conversational, knowledgeable assistant that understands personal context, has onscreen awareness, can reach broad world knowledge for up-to-date answers, and can take systemwide actions across apps. Craig Federighi, Apple’s SVP of Software Engineering, framed it as “a profoundly more intelligent, knowledgeable, and capable Siri.”

The headline architecture point matters for practitioners: Siri AI runs on “Apple Foundation Models that run on device and on servers using Private Cloud Compute,” with Apple repeating that under Private Cloud Compute personal data “is not stored nor made accessible to Apple or anyone else.” In its keynote, Apple said it collaborated with Google and the Gemini family of models to develop the next generation of Apple Foundation Models — a notable shift for a company that had emphasized homegrown models.

What Siri AI actually does

A new dedicated Siri app uses iCloud to privately sync conversation history, so you can start on Mac and continue on iPhone, iPad, Apple Watch, or Vision Pro. Personal-context examples Apple gave: surface a hotel confirmation number from an old email, find a restaurant a friend texted you, or pull trip photos. Visual Intelligence expands to iPad, Mac, and Vision Pro — on Mac you can select something onscreen and type to Siri; on Vision Pro you can ask about what you are looking at; on iPhone a Camera “Siri mode” reads the field of view.

Hardware bar and regional gaps

The most capable on-device Siri AI features carry a real hardware floor: iPhone Air, iPhone 17 Pro, iPhone 17 Pro Max, iPad (M4) or later with at least 12GB of unified memory, and Mac (M3) or later with at least 12GB of unified memory. Two large markets are excluded at launch. Apple says the Digital Markets Act blocks Siri AI on iOS 27 and iPadOS 27 in the EU; Federighi said Apple is “deeply disappointed” EU users won’t have it on iPhone or iPad — though it stays available on macOS 27, visionOS 27, and watchOS 27 in the EU. Siri AI is also held back in China pending regulatory work.

Developer surface: Foundation Models framework + Core AI

For developers, Apple expanded the on-device Foundation Models framework with image input (passing images alongside text), custom skills, and server-side model execution, and announced a new Core AI framework. App Intents got a push so apps can let users trigger in-app actions through Siri. Xcode’s coding assistant now handles app localization and can interact with simulated devices, extensible via custom skills; Federighi pitched Xcode as the “best place” to build with agentic coding.

The rest of the OS lineup

The releases span iOS 27, iPadOS 27, macOS 27 (reported codename “Golden Gate”), watchOS 27, visionOS 27, and tvOS 27. The Liquid Glass design gains an opacity slider so users can dial the effect up or down. Apple touted performance gains including AirDrop transfers up to 80% faster, app launches up to 30% faster, and camera-roll loading up to 70% faster, with iOS 27 billed as its broadest release (back to iPhone 11). Photos adds Reframe, Extend, and an upgraded Cleanup, with generated edits carrying SynthID watermarks. This was also Tim Cook’s final WWDC keynote as CEO; John Ternus takes over September 1.

Timeline

Developer testing opened June 8, a public beta is due next month, and the free updates ship this fall.

Item	Detail
Assistant	Siri AI: dedicated app, on-device + Private Cloud Compute models
Models	Next-gen Apple Foundation Models, built with Google’s Gemini family
Min hardware (top features)	iPhone Air / 17 Pro / 17 Pro Max; iPad M4+ and Mac M3+ with 12GB unified memory
Excluded at launch	EU (Siri AI on iOS/iPadOS, per DMA) and China
Dev tools	Foundation Models framework (image input, custom skills, server models), Core AI, App Intents, Xcode agent

Practitioner note: the 12GB-unified-memory floor is the real gate. If you build on the Foundation Models framework, segment users by device tier and keep a cloud or smaller-model fallback for anything below the bar — the keynote’s flashiest Siri demos will not run on most installed iPhones for a while.

Under-considered angle: Apple leaning on Google’s Gemini for its next foundation models quietly reframes the “private, on-device, all-Apple” story. The privacy envelope (Private Cloud Compute) is unchanged in Apple’s telling, but the model supply chain now runs partly through a rival — and the EU’s DMA standoff shows regulation, not silicon, may decide where this assistant is even allowed to ship.