Recently, Meta released the weights for Muse Spark Open, the most capable US open-weight model to date. This model is built for coding and agentic work, and excels at long-horizon tasks, complex software engineering, and multi-step processing.
Today we’re releasing Muse Spark Open, the most capable US open-weight model to date.
Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning.
Key Model Specs:
→ 816B total / 42B active parameters
→ 1M token… pic.twitter.com/yAtoxNeYJm— AI at Meta (@AIatMeta) October 15, 2026
Muse Spark Open
Muse Spark Open is a natively multimodal model with 816B parameters (42B active) and a 1M-token context window. This is the first model in its class to run on your desktop using two Apple M5 Ultra Mac Studios with the latest MLX framework.
It runs on leading inference engines like SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, Ilama.cpp, and Unsloth with optimisations for AMD, Apple, Intel, and NVIDIA silicon. Developers can easily further customise the model using the PyTorch framework.
The model showed impressive performance on key benchmarks for agentic, knowledge & reasoning, coding and multimodal (with tools).
Leveraging the latest version of MLX, Muse Spark Open was optimised to run on Mac Studio with M5 Ultra. Apple’s unified memory architecture lets the M5 Ultra’s integrated GPU use up to 512GB per Mac Studio, so two machines can hold an entire model. And the chip’s 1.2TB/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.
The open-weight version of Muse Spark ships with custom, auto-tunable Helion GPU kernels in SGLang, delivering 1.03-3.95x kernel-level speedups on NVIDIA and AMD GPUs, and a 3% end-to-end speedup on AMD MI350.
Open weights also carry their own considerations, and Muse Spark Open has additional post-training safety for that reason. The results demonstrate strong safeguards across risk categories measured. Meta evaluated the model in line with the Meta Superintelligent Scaling Framework, covering the most severe risks, and it met the thresholds its framework requires for release.
The company is releasing the weights under the Muse Model License with limited commercial terms, which allow many developers to use the model for free.
Availability
Muse Spark Open is available through API platforms at Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. Model weights are available for download on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models.
You can customise Muse Spark Open for your case by leveraging PyTorch’s TorchTitan to further tune the model. To run it on Apple Silicon, use MLX, which pulls the weights directly from Hugging Face.
