What Muse Spark Open is
According to the model page:
- It has 816 billion parameters in total and 42 billion active.
- The context is 1 million tokens.
- It was trained on data from more than 100 languages.
Who it is for
- Organizations that need a strong open model without sending the request to a third-party API.
- Software teams that want a coding assistant hosted for them.
How we deploy
It is a very large model: the page asks for a GPU with at least 700 GB of memory for the NVFP4 build, or two Mac Studio computers with M5 Ultra. So we set up the capacity when your organization asks. The model joins the same API, with the privacy barrier.
How it works
You describe the case
The tasks, the volume and the rules for your data.
We test with your samples
Meta does not list Portuguese among the languages. So we measure the quality in Brazilian Portuguese with your own tasks.
We connect it to your access
We set up the capacity, connect the model to your key and agree on a usage price.
Good to know
- Muse Model License, checked on the model page on Hugging Face on October 17, 2026.
- The license requires the label “Built with Muse Spark Open” to be visible in the product or its documentation.
- To offer the model as a service, an organization with more than 20 million dollars of annual revenue, counting the companies of its group, first needs an agreement with Meta.
- On-demand deployment: the model goes live when an organization asks for it.