Meta’s Muse Spark Open: an 816-billion-parameter AI that runs on Mac Studio
Working with Apple, Meta has released its most powerful open-source AI model yet, and it can run on desktop computers without the cloud.
Meta has released Muse Spark Open, the most capable open-source artificial intelligence model it has built so far. The company says it is the best-performing open-source model made in the United States. Its headline feature is that it can run on a user’s own computer instead of relying on cloud servers.
Teaming up with rival Apple
Using Apple’s MLX machine-learning framework, the model can run on two Mac Studio computers with M5 Ultra chips linked together. According to Axios, the two long-time rivals had been working on this together for several months.
After the launch, Meta CEO Mark Zuckerberg credited Apple CEO John Ternus and his team in a post on X. Ternus replied that running such a large model so efficiently on a desktop shows how fast on-device AI is advancing. Meta’s chief AI officer Alexandr Wang joked to Axios that the team was thrilled to see “mom and dad” being nice to each other again.
What the model can do
Meta describes Muse Spark Open as natively multimodal, meaning it was built from the start to understand more than just text. It has 816 billion parameters in total. The company says it performs strongly on long-running tasks, complex software engineering, and multi-step reasoning.
Apple published new details about MLX at the same time. The framework, built for Apple silicon, will let Muse Spark Open and two other open models run on Apple devices.
Why it matters
Running a model this large on local hardware means using advanced AI without sending data to outside servers. That is a big advantage for researchers, developers, and organisations handling sensitive information. Because it needs two high-end Mac Studios, though, it remains largely within reach of professional users for now.
Comments