Cupertino — Meta's Open-Weight AI Model Runs Locally on Mac Studio

Meta released an open-weight AI model Thursday that it says is the most capable of its kind available in the United States, and the company built it to run locally on Apple hardware, according to TechCrunch.

The model, called Muse Spark Open, uses 816 billion parameters and a 1 million token context window, TechCrunch reported, citing evaluation results Meta released alongside the launch. Meta says those benchmarks show the model outperforming other popular open-weight systems, though the company has not published independent third-party verification of those claims in the material reviewed.

What makes Muse Spark Open different from other Meta models?

Muse Spark Open is natively multimodal and designed for complex coding tasks and what Meta calls agentic labor, per TechCrunch. The company paired it with Muse Code, its AI coding terminal, and tested it on real-world tasks. In one example cited by TechCrunch, Meta used the model to push bug fixes, new features and a full build for a Mastodon client. In another, the model reviewed and cropped footage inside DaVinci Resolve, the professional video-editing software.

Why does it need so much memory to run?

Running Muse Spark Open locally requires at least 1 terabyte of RAM, TechCrunch reported, citing Engadget. That threshold puts the model out of reach for most consumer machines. TechCrunch noted the model runs best on a Mac Studio equipped with Apple's M5 Ultra chip, a configuration aimed at professional and enterprise users rather than everyday Mac owners.

What is an open-weight model, and why does it matter here?

Open-weight models publish the trained parameters underlying an AI system, which lets developers download, modify and run them on their own hardware instead of calling a company's cloud servers. TechCrunch described this as offering cost efficiency, transparency and flexibility that closed, proprietary models do not provide. Running a model locally also means data stays on the user's own machine rather than passing through a third-party server, a distinction that matters for developers handling sensitive code or footage.

Where can developers access the model now?

Meta is distributing Muse Spark Open now through multiple API platforms, according to TechCrunch. The report did not specify pricing or list every platform carrying the release.

How does this fit into Meta's broader AI push?

The release follows Meta's recent moves to expand its AI product lineup, including subscription offerings detailed in HTT News' coverage of Meta's AI-focused subscription plans. TechCrunch's report also noted Meta has released a separate frontier model, Muse Wonder, though details on that release were not included in the material reviewed for this story.

The hardware requirements mean Muse Spark Open will likely see adoption first among studios, coding shops and enterprises with access to high-end Mac Studio systems, rather than individual hobbyists running AI models on standard laptops.