Meta Releases Muse Spark Open, an 816B-Parameter Open-Weight Model That Runs on Apple Desktops

Meta Releases Muse Spark Open

Meta AI Research has released Muse Spark Open, described as the most capable US open-weight model to date. The natively multimodal model has 816 billion total parameters, 42 billion of them active, and a 1 million token context window. Meta says it is built for coding and agentic work and excels at long-horizon tasks, complex software engineering, and multi-step reasoning.

The company framed the release around the freedom to run advanced models wherever users choose, including at home. 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.

Optimized for Apple Silicon

Apple's unified memory architecture lets the M5 Ultra's integrated GPU use up to 512 GB per Mac Studio, so two machines can hold an entire model. The chip's 1.2 TB/s of memory bandwidth is fast enough to read the small slice each token needs. The result is that two Mac Studios behave like one supercomputer with a terabyte of unified memory.

Meta tested a range of agentic workflows running entirely on-device using Muse Code powered by Muse Spark Open. That included shipping real iOS changes to Ice Cubes, an open-source Mastodon client, such as bug fixes, new features, and a full build, and driving DaVinci Resolve end to end, with the model reviewing footage frame by frame to find the strongest moments and crop them vertically.

Availability and Inference Support

Muse Spark Open is available through API platforms at Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. Model weights can be downloaded from Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models.

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 further customize the model using the PyTorch framework.

The open-weight version ships with custom, auto-tunable Helion GPU kernels in SGLang, delivering 1.03 to 3.95 times kernel-level speedups on NVIDIA and AMD GPUs, and a 3 percent end-to-end speedup on AMD MI350.

Safety and License

Meta said Muse Spark Open includes additional safety post-training, because open weights carry their own considerations. The model was evaluated in line with Meta's Superintelligence Scaling Framework, covering the most severe risks, and it met the thresholds the framework requires for release. Detailed results are in the Muse Spark Open model card, with a full evaluation report available soon.

The weights are released under the Muse Model License with limited commercial terms, which allow many developers to use the model for free. Meta points developers to PyTorch's TorchTitan for further tuning of the model and to MLX on Apple silicon, which pulls the weights directly from Hugging Face.

Sources: research.meta.ai · Techmeme · The New Stack · Engadget