Apple is vigorously enhancing its device-based artificial intelligence (AI) capabilities, constructing a localized AI framework that spans a range of performance levels, from the iPhone to the Mac. According to Apple's latest data on on-device AI inference capabilities, devices such as the iPhone 18 Pro can now support AI models with up to 14 billion parameters. On the other hand, Mac Studio clusters are designed to manage extremely large models, boasting up to 1.6 trillion parameters, for local training and inference tasks. Apple achieves this by optimizing memory capacity and bandwidth through a unified memory architecture, allowing higher-tier devices to exhibit more robust local AI functionalities.

In parallel, Apple is working closely with Google to develop a new generation of Apple Foundation models. These models leverage a variant of the Mixture of Experts architecture, enabling efficient operation on local devices. Furthermore, Apple is investigating model compression techniques to facilitate the execution of larger-scale AI models on mobile devices like the iPhone. This approach not only reduces reliance on cloud-based services but also bolsters privacy protection capabilities, aligning with Apple's commitment to user data security.