Apple Can Now Distill Google Gemini Into Smaller On-Device Models for Siri

Apple Can Now Distill Google Gemini Into Smaller On-Device Models for Siri

Apple's deal with Google is considerably deeper than most observers initially assumed. New reporting from The Information reveals that Apple's multi-year agreement grants the company "complete access" to Gemini models running in Google's own data centers — not just API endpoints for user-facing queries, but the full model weights and infrastructure needed to use Gemini as a training foundation. That access lets Apple distill Google's large models into smaller, highly optimized versions built for on-device inference on iPhone hardware.

In practice, this means the next generation of Apple Foundation Models powering Siri and Apple Intelligence may carry significant Gemini DNA — trained on or shaped by capabilities that Google spent years and billions of dollars developing. It reframes the Apple-Google AI relationship from a simple distribution arrangement into something closer to a technology transfer at the model level, a dynamic that has major implications for both companies' competitive positions.

For developers building on either platform, the takeaway is clear: the boundaries between Google's AI capabilities and Apple's on-device intelligence are blurring in ways that will reshape what Siri can do, what Apple Intelligence means, and ultimately what users expect from AI features across the entire iOS ecosystem.

Read the full article at MacRumors →