At WWDC in June Apple unveiled the Foundation Models Framework, enabling developers to run AI models locally on devices. The company is promoting the feature in OS 26 releases for iPhone, iPad and Mac.

Framework and device support

The Foundation Models Framework provides tools for integrating compact models directly into apps, with execution occurring on end-user hardware. Apple emphasizes local inference to reduce reliance on external compute and preserve responsiveness for simple tasks.

Model size and typical use cases

Apple’s supplied models are compact, roughly 3B parameters, and not intended to match the capacity of leading cloud models from other providers. These models are positioned for quick tasks such as summarizing text and generating content-based tags.

Integration, tooling and costs

Models support tool invocation, allowing developers to connect them to arbitrary APIs and device capabilities as needed. Because inference runs on the device, developers and users avoid pay‑per‑call cloud compute fees for those local operations.

Private Cloud Compute for larger models

Apple plans to offer Private Cloud Compute to host larger models that cannot run efficiently on-device. The company says this option will let developers combine local models with managed cloud inference for heavier workloads.

Developer showcase and examples

Apple released a showcase highlighting how developers apply local AI across categories, illustrating practical integrations and user scenarios. The showcase includes several third‑party apps that demonstrate targeted uses of on‑device models.

• Lil Artist creates educational stories and quizzes tailored to children, voiced by familiar characters and interactive prompts.
• SmartGym uses health-data inputs to compose personalized training programs and adapt recommendations to user metrics.
• LookUp generates contextual example sentences to help learners practice newly encountered foreign words in realistic contexts.
• Essayist extracts information from PDF documents and restructures it into organized references and citations for easier review.

The Foundation Models Framework and local models present a pathway for developers to build lightweight, privacy-oriented features without continuous cloud dependence. Apple’s approach aims to balance immediate on‑device responsiveness with optional cloud capacity for larger model workloads.