Apple Shows On-Device AI Inference Range at JNUC: Up to 1.6 Trillion Parameters on Mac Studio Cluster

At the Jamf Nation User Conference (JNUC), Apple presented a chart during an AI-focused on-device session showing the local AI inference limits across its hardware lineup.

According to a post on X by @aaronp613, the chart indicates that iPhone and iPad models can run models with up to 14 billion active parameters, MacBook Air up to 35 billion, Mac mini up to 70 billion, MacBook Pro up to 120 billion, Mac Studio up to 480 billion, and a Mac Studio cluster up to 1.6 trillion parameters.

Apple's approach relies on unified memory architecture, where memory capacity and bandwidth directly determine the parameter size a device can load. The higher the device tier, the stronger its on-device AI capability. JNUC is Jamf's annual global user conference, bringing together Apple IT administrators, industry leaders and technology enthusiasts to share the latest practices in device management, security and automation.