Apple WWDC Expands On-Device AI Development
Published:
At WWDC 2025, Apple opened more of its on-device foundation-model technology to application developers through the Foundation Models framework.
Developers could call a system-provided language model for tasks such as summarization, extraction, and generation without packaging their own large model or sending every request to a remote server. Local execution can reduce latency, cloud cost, and exposure of private data.
The approach also changes application design because model availability and capability depend on the user’s device. Developers have to treat AI hardware much like graphics capability: check what is available, provide fallbacks, and avoid assuming every machine can run the same workload.