Apple has unveiled its third generation of Apple Foundation Models (AFM) during the WWDC26 keynote, introducing a total of five models that include both local and cloud-based options. Among these, AFM 3 Cloud Pro operates on NVIDIA GPUs hosted in Google Cloud, marking a significant shift in Apple’s cloud strategy.
Initially launched in 2024, the AFM lineup featured an on-device language model with approximately 3 billion parameters and a more extensive server-based model utilizing Apple silicon servers. The recent announcement indicates that the AFM 3 models were developed to enhance user privacy while leveraging advanced cloud capabilities.
The models consist of AFM 3 Core and AFM 3 Code Advanced, both designed for on-device use, along with the server-based AFM Cloud, ADM 3 Cloud (Image), and AFM 3 Cloud Pro. The ADM 3 Cloud (Image) incorporates diffusion technology, while the majority of these models are optimized for Apple silicon.
This initiative follows Apple's collaboration with Google, allowing the integration of Gemini into their AI projects. The transition to third-party infrastructure for the first time under the Private Cloud Compute architecture aims to uphold Apple’s commitment to security and privacy while expanding its AI capabilities.