This page may contain stale information. Last updated: 2026-06-12

Definition

Core AI is Apple’s new framework for on-device AI model deployment, introduced at WWDC 2026. It provides an architecture optimized for unified memory and the Neural Engine on Apple silicon, enabling developers to run full-scale LLMs locally without cloud inference.

Key Capabilities

  • On-device LLM deployment: Deploy full-scale language models on Apple silicon via CoreAILanguageModel
  • Foundation Models integration: Core AI models plug into LanguageModelSession through the language-model-protocol
  • Model compilation: Ahead-of-time compilation via coreai-build for device-specific optimization
  • Curated model repository: Open-source models (Qwen, Mistral, SAM3) optimized for Apple silicon
  • Xcode tooling: Core AI Instruments, visual debugger tracing tensors to Python source

Architecture

Core AI spans Python tooling (model conversion, authoring, optimization) and a Swift API for inference. Models ship as .aimodel bundles inspectable in Xcode for size, platform targets, and tensor shapes.

let model = CoreAILanguageModel(modelBundle: myBundle)
let session = LanguageModelSession(model: model)
let response = try await session.respond(to: prompt)

Sources