Meta releases Muse Spark Open, an open-weight AI model that runs on two Macs

What is Muse Spark Open?

Muse Spark Open is a Meta AI model whose weights anyone can download. Weights are the files that make up a trained model. Meta calls it "the most capable US open-weight model to date", which is Meta's own claim.

It is natively multimodal, with 816 billion parameters, of which 42 billion are active at a time, and a context window of 1 million tokens. Meta says it is built for coding and agent work, according to Meta's announcement.

Can Muse Spark Open really run on two Macs?

Meta says it is the first model in its class to run on a desktop, using two Apple M5 Ultra Mac Studios and Apple's MLX framework. Each Mac Studio can give its graphics chip up to 512 GB of memory, so two of them behave like one machine with about a terabyte.

Meta says it tested agent work running fully on the Macs, including bug fixes and new features for an open-source iOS app, and editing video in DaVinci Resolve. Meta did not give speeds in its post.

How do Muse Spark Open's scores compare?

Meta published a table of its own benchmark scores that sets Muse Spark Open against eight other models. The tests and the rivals are Meta's choice, and Meta says a full evaluation report will follow, so treat the table as Meta's claim.

Where can you get Muse Spark Open?

The weights can be downloaded from Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. It is also available through the API platforms of Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI.

Meta lists support for inference engines such as SGLang, vLLM, TensorRT-LLM, Apple MLX and llama.cpp.

What does the licence allow?

Meta released the weights under the Muse Model License, with limited commercial terms. Meta says the terms allow many developers to use the model for free. Read the licence before building a product on it.

Is it safe to release?

Meta says it added safety training because open weights carry their own risks, and that the model met the thresholds of its Meta Superintelligence Scaling Framework. Detailed results are in the model card, and Meta says a full evaluation report is coming soon.

Why does it matter?

A model that Meta calls the most capable open one from the US can now run on hardware people can put on a desk, instead of only in a data centre. Whether it holds up outside Meta's own tests is not yet known.

What's next?

Meta says its full safety evaluation report will be available soon. Outside testers can now check the scores themselves.