Ollama has released v0.40.0-rc0, a new release candidate that significantly enhances performance for local large language models on Apple hardware. Posted recently, this update enables the MLX runtime by default for models running on Apple Silicon devices. This means that supported model architectures will automatically leverage Apple's optimized machine learning framework, aiming to improve efficiency and speed for local inference. During this pre-release phase, Ollama will continue testing and enabling additional models to utilize the MLX runtime.