Apple is said to be looking for ways to bring larger-scale artificial intelligence (AI) models to iPhones, in order to increase processing performance on devices and reduce dependence on cloud computing infrastructure.
According to The Information (USA), Apple has had exchanges with startup PrismML to assess the company's ability to apply technology in operating large language models (LLM) directly on iPhone.
Sources said that PrismML company has optimized Alibaba's Qwen 3.6 open source model to be able to run entirely on iPhone 17 Pro.
This model has 27 billion parameters, larger than Apple's current AI model on devices, AFM 3 Core Advanced with 20 billion parameters. This is said to power AI features on iOS 27 such as Siri with a more natural voice and improved spell reading capabilities.
The difference lies in the operation of the two models. AFM 3 Core Advanced uses sparse architecture, meaning it only activates a small fraction of the parameters in each processing, about 1 to 4 billion parameters.
Meanwhile, the Qwen 3.6 model optimized by PrismML can simultaneously activate all 27 billion parameters, helping to maximize the model's capacity during processing.
If Apple can bring larger AI models to operate directly on iPhones, many features of Apple Intelligence will no longer have to depend on the Private Cloud Compute system, the private cloud computing platform that the company is using for complex AI tasks.
According to experts, processing AI right on the device not only helps reduce server operating costs but also enhances privacy protection, because user data will have to be sent to the cloud for processing less.
