Meta AI Research released Muse Spark Open's weights on October 15. The company describes it as a natively multimodal model with 816 billion total parameters, 42 billion active parameters and a one-million-token context window. Meta distributes the weights under its Muse Model License, which it describes as allowing many developers to use the model for free with limited commercial terms.
Hosted access and a large desktop configuration
Meta lists hosted API access through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI, with weight downloads on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry. The company says the model can run locally across two Apple M5 Ultra Mac Studios using MLX, and names SGLang, vLLM, TensorRT-LLM, Transformers and other inference engines.
Meta calls it its most capable U.S. open-weight model and reports kernel speedups on NVIDIA and AMD hardware. Those comparative and performance statements come from Meta's release and evaluation materials; they are not independent measurements in this article.