Apple announces Siri AI overhaul with Google Gemini-derived model at WWDC 2026

Apple announced a redesigned Siri AI at WWDC 2026, according to the company, powered by a custom Gemini-derived model licensed from Google. The model runs on Apple's Private Cloud Compute infrastructure rather than on-device.

Vision LLM integration

According to Apple, the new Siri will use vision language models to extract information directly from users' screens. This approach eliminates the need for individual applications to ship custom integration code for Apple Intelligence, a significant architectural shift from the 2024 Apple Intelligence announcement.

Vision LLMs have matured substantially since June 2024, when Apple first announced Apple Intelligence features that failed to deliver as promised.

Core AI developer library

Apple introduced Core AI, a new library that bridges PyTorch models to Apple hardware. The library includes Core AI PyTorch Extensions (coreai-torch), a Python package that converts PyTorch models exported as torch.export.ExportedProgram into Core AI AIProgram objects.

According to Apple's documentation, the library "traverses the FX graph node-by-node and maps ATen operators to Core AI operations," enabling developers to run existing PyTorch models on Apple silicon without complete rewrites.

Availability and waitlist

iOS 27 Developer Beta is available as of June 8, 2026, but access to the new Siri AI features requires joining a waitlist. Aaron Perris from MacRumors reports gaining access after the waitlist period, suggesting initial rollout has begun.

No pricing information for Private Cloud Compute access has been disclosed. Apple has not specified whether the Gemini-derived model is a custom-trained version or a fine-tuned variant of an existing Gemini model.

What this means

Apple's partnership with Google for a Gemini-derived model represents a significant strategic shift, moving away from purely in-house AI development after the 2024 Apple Intelligence shortcomings. The vision LLM approach for screen context is technically sound and avoids the fragmentation that would come from requiring per-app integrations. However, given Apple's track record with the 2024 announcements, actual performance will need independent verification once the waitlist clears and the features reach broader testing.