Apple is actively investigating methods to run significantly larger AI models directly on iPhones, aiming to enhance on-device processing capabilities. This initiative involves discussions with PrismML, a startup specializing in AI model optimization.
PrismML has successfully condensed Alibaba’s open-source large language model, Qwen 3.6, to operate entirely on an iPhone 17 Pro. Notably, Qwen 3.6 comprises 27 billion parameters, surpassing Apple’s current on-device model, AFM 3 Core Advanced, which contains 20 billion parameters. A key distinction is that while AFM 3 Core Advanced employs a sparse architecture—activating only 1 to 4 billion parameters at any given time—PrismML’s adaptation of Qwen 3.6 activates all 27 billion parameters simultaneously.
Integrating larger AI models directly onto iPhones could significantly reduce reliance on Apple’s Private Cloud Compute servers. This shift would not only lower operational costs but also bolster user privacy by minimizing data transmission to external servers.
Apple’s pursuit of on-device AI advancements aligns with its broader strategy to enhance device performance and user experience. By enabling more complex AI tasks to be processed locally, Apple aims to deliver faster, more efficient, and privacy-centric features to its users.
As Apple continues to explore these possibilities, the potential for more sophisticated on-device AI functionalities grows. This development could lead to more responsive and capable applications, setting a new standard for mobile AI processing.
