Apple platforms

Foundation Models gains multimodal input, Python SDK, and third-party provider swap

2026-06-12 digest confirmed

  • Category: Apple
  • Status: confirmed
  • Sources: Apple developer docs, TechTimes WWDC coverage
  • Summary: At WWDC 2026 Apple expanded the Foundation Models framework with: multimodal image input to on-device models; a Python SDK for building with Foundation Models outside Swift; the LanguageModel protocol allowing third-party providers (Anthropic, Google have published Swift packages) to be swapped with no downstream code changes; and PrivateCloudComputeLanguageModel with a 32K context window now available on watchOS 27.
  • Why it matters: The LanguageModel protocol makes provider portability first-class in iOS 27 and macOS 27; apps can run on-device Apple models by default and fall back to cloud providers without changing the inference call site.

Send feedback on this story