Developers gain local access to an 816-billion-parameter open-weight model with a 1M-token context window and native multimodality, runnable on a pair of Apple Mac Studio desktops with M5 Ultra chips. Meta’s Muse Spark Open ships today for download and is available through hosted inference partners on the same day.
What Muse Spark Open includes
Muse Spark Open is a natively multimodal model with 816B total parameters, 42B of which are active per token, and a 1M-token context window. The weights are released under the Muse Model License with limited commercial terms, which lets many developers use the model at no cost. Meta describes it as the most capable US open-weight model to date.
The model is built for coding and agentic work, with strengths in software engineering tasks spanning extended contexts and multiple reasoning steps.
Running the model on Apple silicon
Muse Spark Open is the first model in its class to run on a desktop using two Apple M5 Ultra Mac Studios with the latest MLX framework. Apple’s unified memory architecture lets the M5 Ultra’s integrated GPU use up to 512 GB per Mac Studio, so the two machines together can hold the full model. The chip’s 1.2 TB/s memory bandwidth is fast enough to read the small slice of the model each token needs.
Meta states that the two Mac Studios together behave like a single supercomputer with a terabyte of unified memory, and that this is the most efficient way to run a powerful model on your own desktop.
Tested agentic workflows
Meta reports testing a range of agentic workflows running entirely on-device using Muse Code powered by Muse Spark, including:
- Shipping real iOS changes to Ice Cubes, an open-source Mastodon client, including bug fixes, new features, and full builds.
- Driving DaVinci Resolve end to end, with the model reviewing footage frame by frame to find the strongest moments and crop them vertically.
Where to download and run the model
Model weights are available for download on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models. Hosted inference is available today through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI.
The model runs on leading inference engines including SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, and Unsloth, with optimizations for AMD, Apple, Intel, and NVIDIA silicon. Developers can customize the model further using the PyTorch framework.
Kernel-level speedups
The open-weight release ships with custom, auto-tunable Helion GPU kernels in SGLang, delivering 1.03-3.95× kernel-level speedups on NVIDIA and AMD GPUs and a 3% end-to-end speedup on AMD MI350.
How to get started
To customize Muse Spark Open, Meta points developers to PyTorch’s TorchTitan for further tuning. To run it on Apple silicon, use MLX, which pulls the weights directly from Hugging Face. The model collection is hosted at huggingface.co/collections/meta-models/muse-spark-open.
Safety and release framework
Meta evaluated Muse Spark Open under the Meta Superintelligence Scaling Framework, covering the most severe risks, and reports that the model met the thresholds the framework requires for release. Detailed results are in the Muse Spark Open model card, with a full evaluation report to follow. The open-weight release also includes additional safety post-training specific to the considerations that come with openly licensed weights.
FAQ
What is Muse Spark Open?
Muse Spark Open is an 816B-parameter open-weight multimodal model from Meta with a 1M-token context window, built for coding and agentic work and released under the Muse Model License with limited commercial terms.
Can Muse Spark Open run on a desktop?
Yes. It is the first model in its class designed to run on a desktop using two Apple Mac Studio machines with M5 Ultra chips, which use the MLX framework and the chip’s unified memory architecture.
Where can developers access Muse Spark Open?
Weights are available on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models. Hosted inference is available through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI.