Meta has put an 816-billion-parameter model on the desktop. On October 15, 2026, the company released Muse Spark Open, an open-weight multimodal model built for coding and agentic work, with Apple hardware support at its center.
Meta worked with Apple so two Macs can run Spark Open like an AI supercomputer. The setup uses two M5 Ultra Mac Studios with 1TB of combined memory, connected through Apple’s updated MLX toolkit, an array framework for machine-learning tasks on Apple Silicon.
Each Mac Studio offers up to 512GB of unified memory, giving the two-machine configuration enough capacity to run the model locally. Apple’s 512GB-memory configuration does not arrive until late October, and Apple lists one 512GB Mac Studio at $10,799.
That puts a working desktop configuration in the tens of thousands of dollars. Meta describes the setup as requiring a hefty investment, which is corporate language for “the receipt will hurt.”
A large model with a long memory
Muse Spark Open supports a context of up to 1 million tokens and accepts text, images, video, and audio. Its size and input range position it for tasks that combine code, visual information, media, and extended instructions in one session.
Meta’s local demonstrations show Muse Code, powered by Spark Open, modifying the Ice Cubes Mastodon application and operating DaVinci Resolve. Those examples focus on software development and agentic work rather than a chatbot merely producing another paragraph about productivity.
Meta calls Muse Spark Open “the most capable US open-weight model to date.” Its benchmark table places Spark Open ahead of the listed competitors on JobBench and MCP-Atlas, while the model card reports an Artificial Analysis Intelligence Index score of 39.
Meta’s published results also suggest that Spark Open outperforms Chinese models on agentic tasks and visual reasoning. Those results come from Meta’s published comparisons, so the benchmark table and the model card will carry more weight than the launch-day superlatives.
Apple’s hardware bet has a large price tag
The two-Mac arrangement is not the only configuration Meta describes. Muse Spark Open can also run on four 256GB M5 Ultra Mac Studios, a setup that will cost roughly $38,000.
The numbers make the point plainly: local access to a model at this scale depends on memory capacity, and memory capacity on these systems is expensive. The hardware is available as a desktop configuration, but that does not make it a casual purchase.
Meta and Apple’s chief executives publicly supported each other’s work on the model. Mark Zuckerberg wrote, “Muse Spark Open — weights releasing today — runs great on new Mac Studios thanks to work with @johnternus and Apple.”
John Ternus replied, “Congrats to the team! Amazing to run a model of Muse Spark Open’s caliber and scale this efficiently, right from your desk. Mac Studio keeps pushing the boundaries of on-device AI!” Alexandr Wang, Meta’s chief AI officer, added, “We’re excited for mommy and daddy to be playing nice again.”
Apple also published a new technical post covering MLX, Muse Spark Open, and two other open models. The partnership gives Spark Open a route to local execution on Apple Silicon, while Meta gets a prominent demonstration that open-weight models can move beyond server hardware—provided the desktop has enough memory and the buyer has roughly $38,000 available.
Based on
- Meta turns Mac into an AI supercomputer — therundownai.beehiiv.com
- Meta unveils Muse Spark Open, the new top U.S. open weights model — and it runs on two paired Mac desktops | VentureBeat — venturebeat.com
- Meta Built A Huge Open-Weight AI Model For Macs, But You’ll Need 1TB Of RAM To Run It — engadget.com
- Meta and Apple team up to make Muse AI models run better on Macs | The Verge — theverge.com
