Language models available to apps
The Foundation Models framework is Apple’s developer framework for using large language models from apps on Apple platforms. Through Swift APIs it brings summarization, classification, extraction, text generation, and tool calling into an application.
Because the on-device model behind Apple Intelligence ships with the OS, supported devices run it without bundling model data in the app and without network calls or metered API costs.
When it appeared, and what changed
The framework was introduced at WWDC25 as the way for developers to reach the on-device language model at the center of Apple Intelligence. It is not new in WWDC26.
WWDC26 extended it to image input, models running in Private Cloud Compute, external models including Claude and Gemini, and open source models — moving toward one API that addresses several models.
What you can build
- Summarizing and classifying text or reviews
- Turning natural language into structured data
- Understanding image content for description or extraction
- Calling search or in-app features as tools
- Selecting between models according to the task
How it differs from Apple Foundation Models
Apple Foundation Models are the models themselves. The framework is the development layer between an app and a model. Since WWDC26 it is no longer limited to Apple’s own models — third-party and local models can be addressed the same way.
Model output varies and can be wrong. Shipping it in an app means designing for device support, a fallback when the feature is unavailable, verification of output, and safety.