Ollama has made waves in the tech world by introducing support for Apple’s open-source MLX framework for machine learning. This development promises to enhance the performance of local language models on computers equipped with Apple Silicon chips, like the M1 or later.
The recent improvements include better caching and support for Nvidia’s NVFP4 format. These updates make memory usage much more efficient—especially beneficial for certain models. It’s an exciting time for local AI models, which are gaining popularity beyond just researchers and hobbyists.
A standout example of this trend is OpenClaw. This project has become a sensation, garnering over 300,000 stars on GitHub. Its unique experiments, such as Moltbook, have captured attention globally, especially in China. As developers seek alternatives due to the high costs and limitations of platforms like Claude Code and ChatGPT Codex, interest in running models locally is on the rise. Ollama has also recently bolstered integration with Visual Studio Code to make experimentation easier.
Currently, in its preview phase (Ollama version 0.19), the new support is compatible with one model: Alibaba’s 35 billion-parameter Qwen3.5. While this offers a chance for hands-on exploration, users must meet some demanding hardware requirements. Specifically, an Apple Silicon Mac and at least 32GB of RAM are necessary.
As we look at the growth of local machine learning models, it’s clear that they are no longer just a niche interest. According to recent surveys, 65% of developers are considering local models for their versatility and cost-effectiveness. This shift signifies a fundamental change in how we think about and use AI technology.
In summary, Ollama’s latest updates mark a significant step forward in the push for accessible machine learning. As more people delve into local model experimentation, the landscape of AI development is likely to change forever. For more in-depth information, you can explore Ollama’s official announcement.
