AnyLanguageModel 1.0 unifies Swift model backends
Hugging Face’s AnyLanguageModel 1.0 gives Swift apps one Foundation Models-compatible API for local, hosted, and Apple models, with cross-provider tool calling and guided generation. The release also introduces migration-breaking changes to JSONValue, @Generable enums, and Swift dependencies. [Overview](https://huggingface.co/blog/anylanguagemodel-1-0)
AnyLanguageModel is becoming a serious portability layer for Apple AI apps, but the “drop-in” promise ends at the import line.
- –Supports MLX, llama.cpp, Core ML, Apple Foundation Models, OpenAI, Anthropic, Gemini, Ollama, and Hugging Face Inference Providers
- –Tool calling, structured generation, streaming, and token usage now travel across providers
- –JSONValue is now AnyLanguageModel’s own type, requiring conversions from JSONSchema values
- –@Generable enums changed wire format, so content produced by older versions may not decode
- –Upgrades also require swift-huggingface 1.0 and swift-xet 1.0, adding dependency-level migration work [Release notes](https://github.com/huggingface/AnyLanguageModel/releases/tag/1.0.0)
DISCOVERED
1h ago
2026-10-22
PUBLISHED
2h ago
2026-10-22
RELEVANCE
AUTHOR
DaveTy94