At WWDC 2026, Apple expanded the Foundation Models framework into something that actually changes how I’d approach adding AI features to an iOS app — a single Swift API that covers on-device models, server-based models, and now custom skills.

What’s new

The updated on-device models now support image input, not just text. Alongside that, Apple introduced a new Core AI framework for running larger, full-scale LLMs locally, built specifically around the unified memory and Neural Engine on Apple silicon — rather than treating on-device AI as a scaled-down afterthought of server models.

On the model side, developers can use Apple’s next-generation Foundation Models (built in collaboration with Google’s Gemini) or plug in alternatives like Claude or Gemini directly through a new language model protocol. There’s also Dynamic Profiles, which let you adjust how a model behaves within your app on the fly, without shipping an update for every tuning change.

The part that matters for smaller teams

Developers on the App Store Small Business Program (under 2 million downloads) get access to Apple’s Foundation Models via Private Cloud Compute at no extra API cost. That materially lowers the barrier to trying AI features on a client project without the usual per-token cost conversation.

Where I’d actually use this

For the kind of operational apps I tend to build — point-of-sale, staff/rostering tools, supply chain systems — the interesting angle isn’t chatbots. It’s things like natural-language search over historical records, on-device summarization of a shift or a day’s transactions, or flagging anomalies without sending operational data to a third-party API. Keeping that processing on-device is a real privacy and compliance advantage for exactly the kind of enterprise apps this site is about.