The main announcement at WWDC this morning was the enhancement of Apple Intelligence capabilities that will power the now smarter Siri AI. Powering these various artificial intelligence (AI) features is the Apple Foundation Model 3rd Generation (AFM 3) with two versions developed to run on devices and three more models developed to run in the cloud browser.
AFM 3
The models that run on devices are the AFM 3 Core and AFM 3 Core Advanced. AFM 3 Core is a model with 3 billion parameters that brings improvements over the previous generation AFM model. Then the AFM 3 Core Advanced is a model with 20 billion parameters that is the most powerful model that runs on Apple devices.
It is multimodal and supports features such as high-precision dictation and more expressive voices for Siri. Apple has already stated that AFM 3 Core Advanced requires more powerful devices such as the iPhone 17 Pro and Macs with at least the M3 chip.
Next are the three cloud browser modes AFM 3 Cloud Pro, AFM 3 Cloud and ADM 3 Cloud (Image). AFM 3 Cloud Pro is the most powerful model to run in the browser and can perform agent tasks and complex reasoning. While AFM 3 Cloud is a workload model that requires high speed, accuracy and performance. Finally, ADM 3 Cloud (Image) is used for image editing such as Spatial Reframing, as well as image generation in Image Playground.
To train AFM, Apple uses a mix of data that includes publicly available information, data licensed or purchased from third parties, open source data, data obtained through specific research and synthetic data. Apple does not use personal user data or user interactions when training AFM.
