Meta releases Muse Spark Open, an 816B-parameter open-weight model

Meta released Muse Spark Open, an open-weight multimodal model with 816B parameters, on 15 October, describing it as "the most capable US open-weight model to date."

The model uses a mixture-of-experts design with 42B active parameters and a 1M-token context window. Meta AI Research made the claim in a company blog post.

Meta said the model is natively multimodal and focused on coding and agentic work. It lists long-horizon tasks, complex software engineering and multi-step reasoning as target uses.

The company said the model can run locally on two Apple M5 Ultra Mac Studios using MLX. Each machine has up to 512 GB of unified memory and 1.2 TB/s of bandwidth. Meta said the pair acts like one machine with about 1 TB of memory.

Weights are released under the Muse Model License, which carries limited commercial terms. Meta said many developers can use the model free of charge. The weights are on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models.

Hosted APIs are offered by Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. Meta said the model works with SGLang, vLLM, TensorRT-LLM, Transformers, MLX, llama.cpp and Unsloth. It said Helion kernels in SGLang give speedups of 1.03x to 3.95x on Nvidia and AMD GPUs.

Meta said the model received additional safety post-training and met the release thresholds of its Meta Superintelligence Scaling Framework. The extracted text of the blog's benchmark section gave no figures, so the performance claims await independent testing.

Meta said its full evaluation report will be "available soon." It has not given a date.