WWDC26: Coding Intelligence, Machine Learning & AI Group Lab | Apple
Apple's Foundation Models framework adds a language model protocol letting developers plug in third-party inference backends.
“We've got a Google package that just went live. Anthropic released theirs this morning.”
At WWDC26, Apple introduced a language model protocol in its Foundation Models framework that lets developers swap in different inference backends, including MLX, Core AI, a new Google package, and an Anthropic package released the same morning. Alongside Xcode coding agents, an evaluations framework with a model-judge evaluator, and the Cori framework, this signals Apple opening its on-device AI stack to third-party models. It matters because it positions Apple as a neutral integration layer rather than locking developers to a single in-house model.