Core AI
Load and run AI models on device, tuned for Apple silicon.
Introduced
2026
WWDC26
Kind
Framework
Runs on
iPhone, iPad, Mac and Apple Vision Pro
Get it from
iOS
27.0+
iPadOS
27.0+
Mac Catalyst
27.0+
macOS
27.0+
tvOS
27.0+
visionOS
27.0+
watchOS
27.0+
Overview
Core AI runs models entirely on device, with no server and no token costs. Models are specialized for the hardware they run on and compiled ahead of time so they load fast, from compact vision models to large generative ones.
A memory-safe Swift API gives fine control over inference memory and stateful execution. PyTorch extensions, optimization tools and the Core AI Debugger take a model from training to your app.
Swift API
Memory-safe loading and running of models.
Built for Apple silicon
Models specialized for the hardware and compiled ahead of time.
PyTorch extensions
Export inference functions into a single model artifact.
Core AI Debugger
Inspect graphs, profile and trace back to Python source.
History
How it grew up.
2026
Introduced
Announced at WWDC26.