Meta releases Muse Spark Open, an 816B open-weight model with 1M context

Meta released the weights of Muse Spark Open on October 15, 2026. It is a natively multimodal mixture-of-experts model with 816 billion parameters in total, 42 billion of them active per token, and a 1 million token context window. Meta calls it “the most capable US open-weight model to date.” The weights are free to download, but the license is not a standard open-source one: Meta describes it as having “limited commercial terms.”

What you get

Meta says the weights are on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. The model is also available as a hosted API from Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. Meta’s post gives no prices for those APIs, so check each provider. Muse Spark Open is separate from Muse Spark 1.3, which Meta sells through its own API and which we list on our AI models page.

The Hugging Face model card says BF16 and NVFP4 versions are available. Running it yourself takes “a GPU with at least 700GB of VRAM for NVFP4 and 2TB for BF16,” according to the card. Meta also says it is tuned to run on two Mac Studios with M5 Ultra chips using Apple’s MLX framework.

What the license allows

The Muse Model License, dated October 15, 2026, has no monthly active user cap. Its main limits:

Rule	What the license says
Model as a Service	Notify Meta and sign a separate agreement if your company’s gross revenue tops $20 million in any 12 months
Internal, research and other commercial use	Not covered by the revenue rule
Attribution	Display “Built with Muse Spark Open” in your product, docs, website or settings
Access	Gated: you must sign in and give your legal name, date of birth and full organization name

Meta’s post has a “Key Benchmarks” heading but no figures under it; it points to a separate evaluation report.

Who it affects

Companies that want a frontier-scale model they can host on their own hardware, and smaller teams that want to resell it as an API. Firms over $20 million in revenue that plan to offer the model as a service need a deal with Meta first. Everyone else can use it under the attribution rule. If you plan to rent GPUs to run it, our GPU cloud comparison lists current hourly prices.