Meta launches Muse Spark Open, an open-weight AI model built for Macs

Meta has introduced Muse Spark Open, an open-weight AI model that the company says is its most capable open-weight release in the United States. The notable part for creators, developers and small teams is not just the model’s size, but Meta’s claim that it was designed to run locally on Apple computers.

That could make high-end local AI work more practical for Mac-based workflows, though the hardware bar remains steep. According to TechCrunch, the model is available through multiple API platforms, while local use requires substantial memory.

What Meta says Muse Spark Open can do

Muse Spark Open is a multimodal model with 816 billion parameters and a 1 million token context window, according to Meta. The company says evaluation results released Thursday show stronger performance than other popular open-weight models.

Meta describes the model as built for complex coding tasks and agentic work. In practice, that means the company is positioning it not only as a chatbot-style assistant, but as a model that can help drive multi-step tasks across software and development workflows.

The launch follows Meta’s recent push around its Muse family. For readers tracking that broader strategy, Emonarc covered Meta’s earlier frontier-model announcement here: Meta launches Muse Wonder, its first frontier AI model family.

Local Mac AI, but with a serious hardware requirement

The Mac angle is the most important part of this release for many independent creators and small studios. Meta says Muse Spark Open has been designed to run locally on Apple’s computers, which could appeal to users who want more control over files, latency and experimentation than they may get from purely cloud-hosted systems.

However, local does not mean lightweight. TechCrunch cites Engadget in reporting that running the model locally requires at least 1 terabyte of RAM. The report also says the model runs best on a Mac Studio equipped with Apple’s M5 Ultra chip.

That requirement limits the audience. A typical freelancer’s laptop is not the target machine here. The more realistic users are developers, AI-heavy creative teams, research-minded studios and small companies that already invest in high-end desktop hardware.

For everyone else, API access may be the more practical route. Meta is making the model available through several API platforms, according to the report, though the source does not list pricing or specific providers.

Coding and creative desktop workflows are the first signal

Meta says it paired Muse Spark Open with Muse Code, its AI coding terminal, and used it in agentic workflows. One example involved work on a Mastodon client, including fixes, features and a full build, according to the company.

That matters because many small teams are trying to move beyond prompt-and-response AI into systems that can handle larger chunks of work. In coding, the value is not just generating snippets; it is staying aware of a larger codebase, following instructions over many steps and interacting with tools in a controlled way.

Meta also says it used the model with DaVinci Resolve, the professional video editing application. In that test, the model was able to review and crop footage, according to the company.

For creators, that points to a possible future where a local model can assist inside heavyweight desktop applications rather than only responding in a browser tab. The source does not say whether Meta is releasing a consumer-facing Mac app or plug-ins for these workflows, so teams should treat the examples as demonstrations of capability rather than ready-made production tools.

The move also sits alongside a broader industry interest in local-first AI. Google, for example, has also been pushing offline and local workflows in other contexts; Emonarc recently covered that direction in Google launches local-first AI Edge Foresight for offline meeting notes.

Why open-weight matters for small teams

Open-weight models are not the same as fully open-source software, but they can give developers and organizations more room to inspect, adapt and deploy models than closed proprietary systems. TechCrunch notes that open-weight models can offer cost efficiency, transparency and flexibility that closed models do not provide.

For small businesses, that can translate into a few practical questions:

• Do you need local processing for sensitive project files?
• Do you have hardware powerful enough to make local inference realistic?
• Would API access be simpler than buying or maintaining a high-end Mac setup?
• Are your workflows mostly coding, video review or other desktop tasks where an agentic model could help?

Muse Spark Open is not positioned as a casual assistant for everyday Mac owners. Based on the reported requirements and examples, it looks more like a tool for technical teams that want a capable open-weight model close to their work environment.

The bigger implication is that local AI on Macs is moving from small models and experiments toward much larger systems. But for now, the gap between “runs locally” and “runs locally for most people” remains significant.