Apple is investigating ways to run significantly larger artificial intelligence models directly on iPhones. The company has met with startup PrismML, which claims to have successfully adapted Alibaba's Qwen 3.6 large language model to operate entirely on an iPhone 17 Pro. This model boasts 27 billion parameters, surpassing Apple's current on-device AFM 3 Core Advanced model which has 20 billion parameters and uses a sparse architecture where only a fraction of parameters are active.

PrismML's approach allows all 27 billion parameters of Qwen 3.6 to be active simultaneously. This capability could enable more advanced Apple Intelligence features, such as enhanced Siri voices and dictation, to process locally on the device rather than relying on Apple's cloud infrastructure. This shift could lead to reduced operational costs for Apple and a further enhancement of user privacy by keeping sensitive data on the iPhone.

The development suggests Apple is prioritizing on-device AI processing for its next-generation features. While Apple's own AFM 3 Core Advanced model powers existing iOS 27 enhancements, the potential integration of larger, fully active models from third parties like PrismML signals a move towards more powerful, localized AI capabilities. This could change how users interact with AI on their devices, offering more complex functionalities without constant cloud connectivity.