Apple Gets Full Gemini Model Access for On-Device Siri Distillation
A new report from The Information reveals that Apple's AI partnership with Google is considerably deeper than the companies have publicly acknowledged. Apple has been granted full access to Google's Gemini model inside its own data centers and is using a technique called "distillation" to create smaller, task-specific models that run natively on device. What makes this particularly significant is that Apple's student models can replicate not just Gemini's outputs but its internal reasoning computations — a far more powerful form of knowledge transfer that yields compact models with capabilities well beyond what their size would normally allow.
The implications for Siri are hard to overstate. Rather than relying on a single large model in the cloud, Apple can now train a fleet of specialized on-device models that carry the intelligence of Gemini without the latency or privacy concerns of a cloud round-trip. It positions Gemini's architecture as the silent backbone of Apple's entire AI strategy — an arrangement that benefits Google in reach and Apple in performance. For the broader industry, it signals that the most consequential AI partnerships aren't always the ones announced at keynotes; sometimes the real action is happening quietly inside private data centers.