Meta launches Muse Spark Open for Macs

Meta says its new open-weight AI model is built for coding and agentic tasks, but running it locally on a Mac requires unusually high-end hardware.

Meta has released Muse Spark Open, an open-weight AI model it describes as its most capable in the United States. The launch stands out not just for the model’s scale and capabilities, but also because Meta designed it to run on Apple computers.

The company is positioning the model for complex coding work and other agentic tasks — meaning jobs that involve carrying out multiple steps, not just replying to a prompt. Meta says its internal evaluations show stronger results than other popular open-weight models.

That makes the release notable for developers and businesses looking at AI tools they can deploy more flexibly. But there is a catch: the hardware needed to run it locally is far beyond what most Mac owners have.

Meta built Muse Spark Open for Macs, but the hardware bar is high

Muse Spark Open is natively multimodal and uses an 816 billion parameter architecture with a 1 million token context window. In practice, that means it can work with different types of input and handle very large amounts of information at once.

Meta also built the model to run locally on Macs. That does not mean it will run on an ordinary laptop. According to Engadget, local use still requires at least 1 terabyte of RAM, and the model is expected to work best on the Mac Studio with the M5 Ultra chip.

That requirement sharply limits who can realistically use it on a personal machine. As IT-PUB News notes, the launch may still matter to Mac users and developers, but it is clearly aimed at high-end systems rather than everyday consumer hardware.

Meta says it used the model for coding and video tasks

Meta says it paired Muse Spark Open with Muse Code, its AI coding terminal, and used it in several agentic workflows. One example the company gave was making coding fixes for a Mastodon client, including bug fixes, new features, and a full build.

The company also says it used the model with DaVinci Resolve, the professional video editing tool. In that setup, the model was able to review and crop footage.

Those examples show how Meta wants the model to be seen. It is not presenting Muse Spark Open simply as a general chatbot, but as a desktop tool that can help with programming and media work. That gives the launch practical relevance not only for AI developers, but also for software teams and creative professionals watching how AI might fit into day-to-day workflows.

Open-weight access offers flexibility, but not without limits

Meta’s release also points to the broader appeal of open-weight AI models. Unlike closed proprietary systems, open-weight models can offer more cost efficiency, transparency, and flexibility.

Still, those advantages do not erase the trade-off here. The local hardware demands are so high that access remains limited, even if the model itself is more open. Businesses and developers may get more room to work with it across different environments and platforms, but that does not automatically make it easy to run everywhere.

For companies, that can mean more control over deployment. For users, it can mean more options for integrating AI tools into existing software and workflows.

Muse Spark Open is available now through API platforms

Meta says Muse Spark Open is available now through several API platforms. That widens access beyond local Mac hardware, even if the on-device experience is reserved for especially powerful machines.

The release adds another entry to the growing open-weight AI market, where companies are trying to balance performance, access, and flexibility. In Meta’s case, the mix of Mac-focused design, large-scale architecture, and coding-oriented use cases is what gives Muse Spark Open its distinct angle.

It also underlines a familiar gap in AI rollouts: strong capabilities do not always translate into broad usability. Muse Spark Open may be powerful, but for many users, actual access will depend less on owning a Mac than on the API platforms Meta is using to distribute it.