MLX
Apple's array framework for running models on Apple Silicon, tuned to use the GPU and unified memory directly. MLX builds often run faster than the equivalent GGUF on a Mac and ship in their own format. It is the Apple-native alternative to the llama.cpp stack.
MLX is a framework, not a container format, which is the distinction that makes the choice easy. An MLX model is weights plus a config laid out the way the framework wants them. Its kernels are written for Metal, and because unified memory means nothing is copied between CPU and GPU, it skips overhead the llama.cpp Metal backend has to work around.
That is home-field advantage rather than magic, and the size of it varies by model and by task.