Apple is exploring innovative methods to implement significantly larger AI models directly on iPhones. This initiative follows discussions with the startup PrismML, which has successfully condensed Alibaba’s open-source large language model, Qwen 3.6, to operate entirely on an iPhone 17 Pro. The Qwen 3.6 model, which boasts 27 billion parameters, surpasses Apple’s current on-device AFM 3 Core Advanced model that features 20 billion parameters. The latter supports enhancements in iOS 27, such as more expressive Siri voices and improved dictation on the iPhone 17 Pro and iPhone Air models. Unlike Apple’s model, which employs a sparse architecture activating only 1 to 4 billion parameters at a time, PrismML’s model can activate all 27 billion parameters simultaneously. This advancement could enable more Apple Intelligence features to function on the device, potentially reducing reliance on Apple’s Private Cloud Compute servers, lowering costs, and enhancing user privacy.
Is Apple Developing Ways to Run Larger AI Models on iPhones?
