What happened
Meta has released what it calls the "most capable" open-weight AI model in the United States, named Muse Spark Open, according to TechCrunch. The model was developed in a teaming-up with Apple and is designed to run locally on Apple silicon. TechCrunch reports it uses 816 billion parameters and a 1 million token context window, and that Meta claims it offers superior performance to other popular open-weight models based on evaluation results the company released.
The model is described as natively multimodal and built for complex coding tasks and agentic work. Meta is making it available now through a number of API platforms, and the company has shown it running directly on Apple's computers rather than in the cloud.
Why it matters for smaller businesses
Open-weight models carry real advantages over closed, proprietary systems. As TechCrunch notes, they can be more cost-efficient and offer greater transparency and flexibility. For a small or medium business, running a model locally means two things that matter a great deal: your data stays on your own machine, and you avoid ongoing cloud usage fees that can climb unpredictably.
That said, there is a significant catch. TechCrunch, citing Engadget, reports that running Muse Spark Open locally requires at least 1 terabyte of RAM. The model is said to run best on a Mac Studio fitted with the M5 Ultra chip. In other words, the "no-cloud-cost" promise comes with a heavy upfront hardware cost. For most SMEs, 1TB of RAM is well beyond a typical office machine, so the dream of a private AI workhorse sitting under the desk needs a realistic budget check.
A real-world use case
Meta did not just publish benchmarks — it showed the model working. According to TechCrunch, Meta paired Muse Spark Open with Muse Code, its AI coding terminal, and ran it through agentic workflows. One example involved coding work on a Mastodon client that included, in Meta's words, "bug fixes, new features, [and] full build."
Meta also demonstrated the model operating DaVinci Resolve, the high-end video editing software, where it was reportedly able to review and crop footage. For a business, picture a small software team pointing the model at a repository to clear a backlog of bug fixes overnight, or a content studio using it to pre-sort and trim raw video before an editor steps in. These are concrete tasks, not abstract promises — the question is simply whether the hardware investment pays off for your workload.
What next
Expect more open-weight models aimed at local hardware as companies compete on privacy and running costs. For now, Muse Spark Open sits at the high end: powerful, private, and genuinely local, but demanding serious kit. Smaller teams may wait for lighter versions or lean on API access in the meantime, while well-resourced firms with strict data rules could find the local route worth the outlay.
The sensible first step is to match the tool to a real task and run the numbers — hardware cost versus cloud subscription over two or three years — before committing.
Getting started with AI the right way
Brain.mt can help you using AI for your business. Whether you are weighing a local model against cloud services or working out where AI fits into your daily operations, contact me for more information. I also offer dedicated workshops and training about this subject, so your team can make confident, informed decisions rather than guesses.
