Apple is distilling Google's massive Gemini AI models into smaller, more secure components through a knowledge transfer technique. This process allows the tech giant to create specialized models that can perform specific tasks with the same speed and accuracy as the original Gemini. The partnership gives Apple extensive access to Gemini within its own data centers, enabling engineers to closely examine and adapt the technology for on-device processing.

The distillation method works by having Apple's external models act as students to learn Gemini's internal computations. These smaller models can run on devices like iPhone without requiring server connectivity, reducing both computational costs and hardware requirements. While Gemini will continue powering Siri responses for now, Apple's focus remains on creating efficient on-device AI capabilities rather than developing a direct competitor to Google's full-scale model.

Apple Foundational Models leads this initiative, which represents the company's strategic commitment to on-device AI processing. Despite having the capability to build larger models, Apple appears focused on specialized functions that can run efficiently on consumer devices. This approach aligns with Apple's emphasis on privacy and performance, positioning the company to deliver AI features that work seamlessly without relying on constant cloud connectivity.