Meta released the weights for Muse Spark Open on October 15, 2026, making the model available through Hugging Face downloads and hosted access from Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI, according to the Meta AI Research announcement.
Muse Spark Open is a natively multimodal mixture-of-experts model with 816B total parameters, 42B active parameters, and a 1M-token context window. It accepts text, images, video, and audio as input and produces text output.
Meta says the model runs on SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, and Unsloth, and is optimized to run on two Apple M5 Ultra Mac Studios with MLX. For teams considering self-hosting, the Hugging Face model materials cite roughly 700GB of VRAM for NVFP4 and about 2TB for BF16.
The practical tradeoff is clear: the weights are open, but deployment is still heavy enough that many teams may prefer managed inference. Meta also says Muse Spark Open ships under the Muse Model License with limited commercial terms.