Apple's AI partnership with Google is deeper than previously reported. According to a new account from The Information, Apple has "complete access" to Google's Gemini model inside its own data centers — and can use that access to build smaller, specialized models tuned for Apple devices.

The process is called model distillation: a large "teacher" model transfers knowledge to smaller "student" models that are more efficient and fast enough to run directly on-device. Rather than relying on Gemini as a cloud service, Apple can create offshoot models that roughly approximate Gemini's performance while running locally on iPhones and Macs — without round-trips to the cloud.

The arrangement gives Apple more control than a typical API partnership. Apple's student models can learn to imitate Gemini's internal reasoning steps, not just its outputs, which reportedly leads to better results than standard fine-tuning. The deal was first announced in January 2026 and extends to Apple's cloud infrastructure and Gemini 3, which topped AI leaderboards when it launched last November.

There's a catch: Apple's priorities for Siri don't always align with Gemini's strengths. The company's Foundation Models team is continuing in-house development in parallel, with goals that remain unclear.

Apple is expected to unveil a major Siri overhaul at WWDC in June, including features like persistent conversation memory and proactive suggestions. The Gemini-distilled models will likely underpin many of these capabilities, shipping in iOS 27 later this year.

The distillation approach signals a broader industry trend: rather than building frontier models from scratch, companies are increasingly licensing access to top-tier models and refining them for specific use cases. Apple's arrangement with Google may be one of the most expansive of its kind yet.