Apple adds Claude, Gemini and free Private Cloud Compute to Foundation Models framework

Apple's developer documentation says the Foundation Models framework is a native Swift API for integrating large language models into apps. It provides direct access to Apple's on-device foundation model, Private Cloud Compute, and server models including Claude and Gemini. The framework includes guided generation, tool calling, image understanding, Dynamic Profiles, an Evaluations framework, and a new fm command-line tool in macOS 27. The documentation presents the framework as a single interface for on-device and remote models.

The most concrete commercial detail is the Private Cloud Compute access tier. Developers enrolled in the App Store Small Business Program with fewer than two million first-time App Store downloads can use Apple Foundation Models on Private Cloud Compute with no cloud API cost, according to the page. The framework also defines a Language Model protocol so any model provider can create a Swift package. That protocol is the technical mechanism for swapping Apple models, Claude, Gemini, or other hosted models without changing app code.

This fits Apple's broader pattern of packaging AI features around privacy and local inference. Apple Intelligence already runs some capabilities on-device and routes larger requests to Private Cloud Compute, a design intended to limit data retention. By exposing the same Swift API for Claude and Gemini, Apple turns model choice into a developer configuration rather than a separate vendor integration. Dynamic Profiles, which let apps change model, temperature, reasoning level, and tools within one session, reinforce that abstraction. The developer page does not specify server-model pricing or exact availability.

The change affects iOS and macOS developers, especially smaller teams that cannot absorb per-token API costs. Free Private Cloud Compute for eligible apps under two million first-time downloads lowers the barrier to adding summaries, agents, image understanding, and tool calling without shipping a separate backend. For users, the framework's on-device and Private Cloud Compute options reduce reliance on ordinary cloud inference, while the Claude and Gemini integrations preserve access to larger models when local capacity is insufficient.