Ollama Enhances Local Model Performance on Macs with MLX Support

Ollama has made significant strides in running large language models locally on Macs by integrating support for Apple's MLX framework. This enhancement, along with improved caching and support for Nvidia's NVFP4 format, optimizes memory usage and boosts performance, especially on Apple Silicon chips like the M1. The timing is ideal as local models are gaining traction beyond just researchers and hobbyists. The success of OpenClaw, which has become a sensation with over 300,000 GitHub stars, highlights the growing interest in local model experimentation. Developers, frustrated by the limitations and costs of cloud-based tools, are increasingly turning to local solutions. Ollama's latest update, available in preview, supports a single model—the 35 billion-parameter Qwen3.5 from Alibaba—requiring robust hardware, including a Mac with at least 32GB of RAM.