Unpacking Apple's Update to Foundation Models Framework

Apple’s latest announcement at WWDC introduces a significant enhancement to its Foundation Models Framework (FMF), expanding the ecosystem of on‑device and cloud‑based large language models available to developers. The FMF, first revealed at WWDC 2025, originally offered a lightweight on‑device LLM for macOS, iOS, iPadOS, and visionOS, alongside a larger model accessible via Private Cloud Compute (PCC). The update broadens the framework’s capabilities, enabling developers to integrate more sophisticated AI functionalities directly into Apple devices while maintaining privacy through on‑device processing and secure cloud support.

Key findings highlight that the new FMF version improves model efficiency, reduces latency, and enhances customization options for developers. Apple emphasizes that the framework now supports a wider range of use cases, from natural language understanding to multimodal content generation, and introduces tighter integration with the Apple ecosystem’s privacy safeguards. The update also includes tooling for easier model deployment, monitoring, and performance tuning across Apple’s hardware platforms.

The scope of the FMF update covers all major Apple operating systems—macOS, iOS, iPadOS, and visionOS—and targets developers worldwide who rely on Apple’s hardware for end‑user applications. The time frame spans the current fiscal year, with rollout expected throughout 2026 as developers adopt the new APIs. Apple’s methodology for this update involves internal benchmarking against existing models, user‑centric privacy testing, and collaboration with select developer partners to refine the framework’s usability. The overarching thesis is that Apple’s FMF will empower developers to create more intelligent, privacy‑preserving applications without sacrificing performance or user experience.