Apple rebuilds Siri as Siri AI and ships five-model Apple Foundation Models 3 at WWDC
The cloud models run on NVIDIA GPUs inside Google Cloud, and Siri AI will skip the EU and China at launch over regulatory hurdles.
Apple announced a rebuilt assistant called Siri AI at its Worldwide Developers Conference on June 8, 2026, alongside a third generation of Apple Foundation Models (AFM 3). The rebuilt Siri AI runs as a standalone app with access to a user's Mail, Calendar, Messages and Photos.
The assistant also gains Dynamic Island integration on iPhone and a camera "Siri mode" that can act on what the camera sees, Apple said. Siri AI will not be available at launch in the European Union or China, which Apple attributed to regulatory requirements.
Underneath sit five models. Two run on-device: AFM 3 Core, a 3-billion-parameter dense model, and AFM 3 Core Advanced, a 20-billion-parameter sparse model that activates 1 to 4 billion parameters per prompt. Three run server-side, including AFM 3 Cloud Pro for agentic tool use and complex reasoning. Apple said the cloud models are custom-built in collaboration with Google, and that for AFM 3 Cloud Pro it worked with Google and NVIDIA to extend its Private Cloud Compute to NVIDIA GPUs in Google Cloud.
Apple reported gains in human preference tests it ran itself. AFM 3 Cloud was preferred on 64.7% of text prompts against 8.7% for the 2025 baseline, while the on-device Core model was preferred on 45.6% versus 23.3%; Apple separately reported a roughly 36% relative improvement in overall response satisfaction for AFM 3 Cloud. Those are Apple's internal evaluations and have not been independently verified; a fuller technical report with benchmarks is planned for later in summer 2026.
The Core Advanced model uses a technique Apple calls Instruction-Following Pruning, storing the full 20-billion-parameter weights in NAND flash and loading only needed expert parameters into memory per prompt. Apple said it does not train AFM 3 on users' private personal data or interactions.
Tim Cook said it was his last WWDC as chief executive before John Ternus takes over on September 1, 2026.
Founder and Chief Editor of Data Phoenix — a San Francisco Bay Area media and education platform focused on AI and Data.
More news

AWS releases six open-source Hugging Face deployment skills for SageMaker

Google Research releases MilleMiglia logistics benchmark generator

AWS launches AgentCore Runtime V2 with elastic memory and snapshot starts
