CUPERTINO, Calif., Oct. 15, 2026 | Apple and Meta announced Thursday that Meta’s new open-weight AI model, Muse Spark Open, can run locally on Apple’s M5 Ultra Mac Studio computers, according to CNBC. Running the 816-billion-parameter model takes at least two Mac Studios, and the performance claims come from Meta and Apple.
What Apple and Meta Announced
Meta calls Muse Spark Open the most capable open-weight model in the United States. TechCrunch reported that it has a 1 million token context window, is natively multimodal and is built for complex coding and agentic tasks. AppleInsider reported that 42 billion of its parameters are active at any moment. Meta CEO Mark Zuckerberg said on X that the model “runs great on new Mac Studios” and thanked Apple CEO John Ternus for the work. Alexandr Wang, Meta’s AI chief, said the companies had worked together for months and that the model is not Meta’s most capable, per MacDailyNews, which cites CNBC.
Meta showed the model working with its Muse Code tool to fix bugs, add features and complete a build in Ice Cubes, an open-source Mastodon client, and controlling DaVinci Resolve to find and crop video segments. TechCrunch notes that Meta’s evaluation results, which say the model beats other popular open-weight models, have not been independently verified. MacDailyNews reports the companies had not publicly partnered since around 2017 and declined to say whether this signals a broader relationship.
What It Takes to Run It
The compressed model weights are about 493GB, so one machine is not enough, according to AppleInsider. Developers need either two M5 Ultra Mac Studios with 512GB of memory each or four with 256GB each. The four-machine setup is available now, while the 512GB option is not expected to ship until late October. Apple’s open-source MLX framework splits the work across the machines, which connect over Thunderbolt 5, and Apple updated it for this model.
In Apple’s own benchmarks, four M5 Ultra Mac Studios generated more than 36 tokens per second with an 8,192-token context, and results vary by workload. AppleInsider estimates a multi-machine setup costs tens of thousands of dollars, so it is better suited to professional development and research than casual use. The weights are on Hugging Face, and developers without the hardware can use hosted API providers.
Frequently Asked Questions
Can one Mac run Muse Spark Open?
No. The compressed weights are about 493GB, so developers need two M5 Ultra Mac Studios with 512GB each or four with 256GB each, according to AppleInsider.
Is Muse Spark Open free to use?
The weights are available to download, and Meta’s license allows many uses for free but includes commercial limits, per AppleInsider.