Apple’s Foundation Models framework now works with other models
Published
The Foundation Models framework is Apple’s Swift API for the on-device model behind Apple Intelligence. With the updates Apple announced at WWDC26, it becomes a single way for an iOS app to work with language models, whether they run on the device or in the cloud.
What changed
According to Apple’s developer documentation:
- Any model, one API. Apps can use Apple’s own models, cloud models such as Claude and Gemini, or any provider that conforms to Apple’s new Language Model protocol.
- Images as well as text. Prompts can include images, so an app can reason about what the user photographed or scanned.
- On-device tools. Vision tools such as text recognition and barcode reading can be called by the model directly, on the device.
- Dynamic Profiles. An app can swap the model, its tools or its instructions within one continuing session.
- Private Cloud Compute for smaller apps. Apps with fewer than two million first-time App Store downloads can use Apple’s latest Foundation Model on Private Cloud Compute.
- An Evaluations framework. Tests for AI features that go beyond unit tests, checking that behavior holds up across changing inputs.
Apple also introduced Core AI, a new framework for running your own models entirely on the device.
What it means for app projects
The biggest change is that the model becomes a decision you can revisit. An app can start with Apple’s on-device model for privacy and cost, and move a feature to a cloud model later, or the other way around, without rewriting it.
On-device processing is also a strong answer to the first question clients ask about AI features: where does our data go? For many features, the answer can now be nowhere.
The Evaluations framework deserves early attention. AI features drift as models change, and a test suite written alongside the feature is what keeps an update from quietly making it worse.
Separately, Apple’s submission rules have moved on: since 28 April 2026, apps uploaded to App Store Connect must be built with Xcode 26 or later, and since 9 September 2026 iOS apps must target iOS 13 or later.
Sources: What’s new in machine learning and Upcoming requirements, Apple Developer.
