Meta releases Muse Spark Open, an 816B open-weight multimodal mixture-of-experts model

Meta has released Muse Spark Open, an open-weight mixture-of-experts model with about 816 billion total parameters and about 42 billion active per token.

It accepts text, images, audio, and video, outputs text, and supports a context window of up to one million tokens.

The Hugging Face post provides BF16 and NVFP4 checkpoints, along with inference recipes for llama.cpp, vLLM, and SGLang and a TRL fine-tuning example.