AnyLanguageModel 1.0: One Year of Local Models on Apple Platforms

Hugging Face has released version 1.0 of AnyLanguageModel, a Swift package that mirrors Apple's Foundation Models framework while letting developers pick the model and provider behind it. In a post dated October 20, 2026, Hugging Face's Mattt says the milestone comes a year and dozens of releases after the project's debut, and that swift-huggingface 1.0 and swift-xet 1.0, which speed up model downloads from the Hub, shipped alongside it.

Adopting the library mainly means changing one import statement. The same session code then runs against Apple's system model, open-weight models through llama.cpp, MLX or Core ML, and hosted services including OpenAI, Anthropic, Google's Gemini, Ollama, Private Cloud Compute on OS 27, and Hugging Face Inference Providers. According to the post, tool calling and guided generation work with every provider, image prompts are supported, and the code also runs on Linux. API keys now come from a closure, so apps can fetch short-lived tokens from their own servers.

The post also argues that smaller open models have become more capable. It cites Google's Gemma 4 E2B at about 3.4 GB and Liquid AI's LFM2.5 2.6B at about 1.7 GB, both small enough for an iPhone, while Alibaba's Qwen3.8 27B needs roughly 16 GB and a Mac with at least 32 GB of memory.

For Apple developers, the release offers a way to mix on-device and cloud inference without maintaining separate integrations. Hugging Face says speech and decision models are next.