Muse Spark Open

In non-nerd speak

Who makes it: Meta. The files come from the maker's own Hugging Face account.

License: Muse Model License. Meta's own license. Free to use, change, and share, even for money. You must show "Built with Muse Spark Open" and follow Meta's usage policy.

Business use: Yes for most businesses. Show "Built with Muse Spark Open" and follow Meta's usage policy. If you sell AI model access to others and make over 20 million US dollars a year, you need a separate deal with Meta.

Training data: We have not found a full public data release. Not yet checked in depth.

Will it say no?

Plays it safe. Meta's card says this open version got extra safety training. It says the model refused over 98% of harmful chemistry and biology requests and 99.2% of cyber jailbreaks in Meta's own tests.

This comes from what the maker wrote, checked Oct 16, 2026. We have not run our own tests yet. It sits next to the trust score and does not change it. How we pick these labels.

Will it run on my computer?

Not for home computers. This one needs server-class hardware, or Meta's own suggested setup, with hundreds of GB of memory. Most people should use it through a hosted service instead. This model uses a mix of experts. Only part of it works on each word, so it runs faster than its size suggests.

Not for home computers: Meta says it needs about 700 GB of graphics memory (source). Our own estimate is about 590 GB. Either way, your laptop is not invited.

Not sure? Use our pick-by-computer guide.

How to run it in LM Studio

1. Get LM Studio from lmstudio.ai and open it.
2. Make a free Hugging Face account. Open the official page and accept the terms.
3. Open the Discover tab. Search for ggml-org/Muse-Spark-Open-816B-A42B-GGUF.
4. Pick a file that ends in Q4_K_M. It is the best mix of size and quality for most people.
5. Click Download. When it is done, open a new chat and load the model.

The files are GGUF builds that LM Studio can load. We checked that this repo exists on Oct 16, 2026. Full LM Studio guide.

How to use it with Hermes

Does it fit? Not yet checked. We could not read its memory window without a sign-up.

1. Load the model in LM Studio. Set the context length to at least 64000.
2. In LM Studio, open the Developer tab and start the server.
3. In a terminal, run hermes model. Choose LM Studio and press Enter to keep http://localhost:1234/v1.
4. Pick this model from the list. Then run hermes to start.

Steps based on the Hermes docs. Full Hermes guide.

Common questions

Is Muse Spark Open truly open source?

No, open weights with custom rules. Meta's own license. Free to use, change, and share, even for money. You must show "Built with Muse Spark Open" and follow Meta's usage policy.

Can I use Muse Spark Open for my business?

Yes for most businesses. Show "Built with Muse Spark Open" and follow Meta's usage policy. If you sell AI model access to others and make over 20 million US dollars a year, you need a separate deal with Meta.

Do I need to sign up to download Muse Spark Open?

Yes. Free Hugging Face account and accept terms.

How much memory does Muse Spark Open need?

Meta says it runs within 700 GB of memory. Our estimate is about 590 GB of RAM for a 4-bit file of about 489.6 GB. This is a rough guide, not a test. So a 700 GB machine will be a tight fit.

Similar models to compare

Same maker first, then models that need about the same memory.

Where Muse Spark Open shows up

• What its license means

Change history

• Oct 16, 2026 New model Muse Spark Open: Added Meta's Muse Spark Open. 816 billion parameters with 42 billion active. Meta says the smaller version needs about 700 GB of graphics memory, so it is not for home computers. Muse Model License: business use with limits, and the download needs a sign-up.

All changes on the site

Sources we checked

If the maker's page and this page disagree, trust the maker. Tell us what we got wrong.