Foundation Models provider interoperability and release status

Facts and statements

• Apple's public LanguageModel protocol describes model capabilities and executor configuration. A LanguageModelExecutor translates Foundation Models transcript entries and request options into provider-native requests, then streams results to LanguageModelSession. Apps add provider packages with Swift Package Manager and initialize the same LanguageModelSession(model:) API. (high confidence) [source]
• Apple's developer overview lists Claude and Gemini under supported third-party models and says any provider can publish a Swift package conforming to LanguageModel. This is the framework's stated provider surface, not a closed list of hard-coded endpoints. (high confidence) (verified as of 2026-10-05) [source]
• Anthropic's ClaudeForFoundationModels package conforms Claude to Apple's LanguageModel protocol and uses LanguageModelSession; its README labels the package beta, while the repository's latest tagged release is v0.2.3 dated October 7, 2026. (high confidence) (verified as of 2026-10-23) [source]
• The ClaudeForFoundationModels README labels the integration Beta and says it targets the OS 27 server-side language-model API introduced in OS betas, with APIs subject to change before general availability. (medium confidence) (verified as of 2026-10-23) [source]
• Google's GeminiLanguageModel is distributed through Firebase AI Logic. Firebase release notes say Firebase 12.19.0 added it as a public preview on 2026-09-09; Firebase 13.0.0 notes direct use through ai.geminiLanguageModel(name:). The guide's 2026-10-20 revision still calls the integration Preview and lists iOS and macOS as the initial supported platforms. (high confidence) (verified as of 2026-10-20) [source]
• Apple's Foundation Models Utilities package exposes ChatCompletionsLanguageModel for any server that implements the Chat Completions REST API. Hugging Face separately documents OpenAI Chat Completions and Responses support in AnyLanguageModel. This supports API-level compatibility, but it does not establish an OpenAI-owned LanguageModel provider; Apple's support overview names Claude and Gemini as its supported third-party models. This last distinction is an inference from the cited package descriptions and support list. (medium confidence) (verified as of 2026-10-23) [source]
• Hugging Face announced AnyLanguageModel 1.0 on October 20, 2026; the repository lists v1.1.0, released October 22, as latest on October 23. The package documents OpenAI Chat Completions and Responses APIs, Anthropic Messages, Gemini, and Open Responses-compatible endpoints as integrations in its own cross-provider implementation. (high confidence) (verified as of 2026-10-23) [source]
• Protocol conformance alone does not establish feature parity. Apple exposes provider-declared capabilities, while Firebase's current integration guide lists supported Gemini models, says Gemini Live API is unsupported, and documents only iOS and macOS for this integration. (high confidence) (verified as of 2026-10-23) [source]
• Firebase announced its Gemini adapter as a preview on June 9, 2026. Its launch post instructed developers to use the Firebase Apple SDK branch wwdc26-preview during the preview. (high confidence) (verified as of 2026-06-09) [source]
• Firebase's guide, last updated October 20, 2026, labels the integration Preview, says it has no SLA or deprecation policy and could change in backwards-incompatible ways, and lists iOS and macOS as the initial supported platforms. Its prerequisites require Xcode 27 from the Mac App Store and an Apple platform simulator or physical device running the corresponding OS version, for example iOS 27. (high confidence) (verified as of 2026-10-23) [source]
• Firebase's October 7, 2026 SDK 13.0.0 release notes remove its interim GenerativeModelSession, SystemLanguageModel, and HybridModel bridge types and direct developers to use GeminiLanguageModel through Apple's native framework. (high confidence) (verified as of 2026-10-07) [source]