Claude ships as a backend for Apple's Foundation Models framework on iOS 27

Anthropic on June 26 shipped Claude support for Apple's Foundation Models framework, the system-level LLM API Apple introduced at WWDC 2026. The release is a Swift package that conforms to Apple's LanguageModel protocol, so developers can use the standard LanguageModelSession surface — respond(to:), streaming, guided generation, and tool calling — and pick Claude as the backend without rewriting their app. Claude joins Apple's free on-device model and Google's Gemini as supported options inside the framework. Apple itself is not in the request path: prompts and responses go directly to the Claude API and are billed to the developer's Anthropic account at standard rates.

The technical move that makes this matter is Apple's decision at WWDC to open the Foundation Models framework to third-party providers. Until iOS 27, app developers wanting to use a non-Apple LLM had to write custom networking code per vendor, manage their own keys, and rebuild streaming and tool-use plumbing. The new framework standardizes that surface. A developer writing to LanguageModelSession once can now run the same code path against Apple's on-device model (free, private), Apple's Private Cloud Compute tier (free if eligible), Claude (cloud, paid), or Gemini (cloud, paid) — switching by a dependency change rather than a code rewrite.

The business read is that Apple has effectively turned its platform into a model marketplace where it sets the API and the model vendors compete underneath. That is a familiar Apple play — App Store for binaries, Maps for vendor data, Wallet for issuers — and it concentrates leverage at the framework level rather than at any one model provider. For Anthropic, getting Claude shipped on day one of iOS 27 with parity APIs avoids being the late option in the picker. The package lands the same week the U.S. Commerce Department cleared Mythos 5 for limited release — two distribution wins for Anthropic in 72 hours.

A takeaway for learners: if you build iOS apps, design your LLM layer around LanguageModelSession from day one. The portability across providers — and across the on-device/cloud boundary — is the headline feature; you can ship a v1 against Apple's free on-device model and graduate users to Claude or Gemini only where the task warrants the cost. If you study platform economics, watch which vendor Apple lists first in its picker UI as more models arrive; that ranking will become a real revenue lever.