Meta has released the weights for Muse Spark Open, describing it as the most capable US open‑weight model to date. The model is natively multimodal with 816 billion total parameters, of which 42 billion are active, and provides a 1‑million‑token context window. It is built for coding and agentic work, excelling at long‑horizon tasks, complex software engineering, and multi‑step reasoning.
According to the announcement, Muse Spark Open is the first model in its class that can run on a desktop, specifically on two Apple M5 Ultra Mac Studios using the latest MLX framework. The unified memory architecture of the M5 Ultra allows each machine’s integrated GPU to address up to 512 GB of memory, so two studios together hold the full model. The chip’s 1.2 TB/s memory bandwidth is said to be sufficient to read the small slice each token needs, making the pair behave like a single supercomputer with a terabyte of unified memory.
The model is accessible through several API platforms: 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. It runs on a variety of 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 who wish to further customize the model can do so using the PyTorch framework.
Performance notes mention that the open‑weight version ships with custom, auto‑tunable Helion GPU kernels in SGLang, delivering kernel‑level speedups of 1.03‑3.95× on NVIDIA and AMD GPUs and a 3% end‑to‑end speedup on an AMD MI350.
Safety work is highlighted: Muse Spark Open received additional safety post‑training for open‑weight release. Evaluations were performed under Meta’s Superintelligence Scaling Framework covering the most severe risks, and the model met the required thresholds for release. Detailed results are available in the model card, with a full evaluation report promised soon.
The weights are released under the Muse Model License with limited commercial terms, which allow many developers to use the model for free. To get started, users can customize Muse Spark Open with PyTorch’s TorchTitan, run it on Apple silicon via MLX pulling weights directly from Hugging Face, or try the hosted inference partners listed. Meta invites anyone who builds something with the model to share their work on the Hugging Face Forums.