On October 15 Meta released the weights of Muse Spark Open: 816 billion parameters in total, 42 billion active, a 1 million token context. The license is Meta's own: use is free, but every product carries a badge, and anyone selling model access above 20 million dollars in combined revenue signs a contract with Meta.
- October 15, 2026: Meta Superintelligence Labs released Muse Spark Open, an open weights model: 816 billion parameters in total, 42 billion active, a 1 million token context. It takes text, image, video and audio, and returns text.
- The Muse Model License lets you use, modify, distill and sell, but the product shows “Built with Muse Spark Open”. For model-as-a-service above 20 million dollars in combined revenue over 12 months, a separate written agreement with Meta is required.
- Running it yourself takes at least 700GB of VRAM for NVFP4 and 2TB for BF16. By Meta's own table, GLM 5.3 and Kimi K3 are ahead of the model on the Artificial Analysis Intelligence Index: 45 and 44 against 39.
Downloading is free. The conditions start the moment you put the weights in a product.
Before you download the weights, Meta's form on Hugging Face asks for your full legal name, date of birth and the name of your organization with all corporate identifiers. The license is the next door, and it is the more interesting one.
The badge applies to derivatives too, and the license counts them broadly: a model distilled from Muse Spark Open or trained on its outputs is a derivative as well. The product that uses it still shows the badge. The contract is for a different group: those who sell model access through an API and are already past 20 million dollars in combined revenue over 12 months.
At an event the sponsor always wants its logo at the entrance, where every guest walks past. Here the entrance is your product's interface.
The contract touches few. If you use the model internally, for research, or as part of your own product that doesn't sell model access, you don't need a contract with Meta. The badge stays, though.
The table in the card is Meta's, and in it the model doesn't lead everywhere. GLM 5.3 and Kimi K3 are ahead of Muse Spark Open on the Artificial Analysis Intelligence Index: 45 and 44 against 39. We haven't seen an outside test of the model itself yet.
Whoever sells model access through an API and is close to 20 million dollars in revenue reads the model-as-a-service clause before uploading the weights. Everyone else puts the badge on and gets to work.