Muse Spark Open
Meta's open-weight 816B multimodal mixture-of-experts: 42B active, 1M context, text, image, video and audio input.
In short. Muse Spark Open is Meta's most capable downloadable model. It understands text, images, video and audio, and it is built for coding and for acting as an agent. Meta says it can run on two high-end Apple Mac Studios, though most people will use it through a hosted service. Downloads need approval and the license has some conditions.
Summary
Meta Superintelligence Labs released Muse Spark Open on 15 October 2026. It is the open-weights version of the Muse Spark model that Meta shipped in August, shares the same base model as Muse Spark 1.3, and carries extra safety post-training so that it is more robust against misuse without system-level guardrails. Meta calls it the most capable US open-weight model to date.
It is a mixture-of-experts transformer with 816 billion parameters in total and 42 billion active per token: 62 layers, 256 experts with 8 active per token, and a 202,048-token vocabulary. A 1.8-billion-parameter Perception Encoder reads images and video. The model takes text, images, video and audio, writes text, and has a 1-million-token context window. It was trained on data in more than 100 languages.
Meta publishes BF16 weights (about 1.6 TB) and an NVFP4 build (about 0.5 TB). Its model card recommends at least 700 GB of GPU memory for NVFP4 and 2 TB for BF16, and Meta also describes running it on two Apple M5 Ultra Mac Studios with MLX. SGLang, vLLM, Hugging Face Transformers, TensorRT-LLM, Unsloth and llama.cpp are supported.
The download is gated: you must accept the Muse Model License and be approved on Hugging Face. The license allows commercial use, but a business that offers the model to others through an API with group revenue above US$20 million needs a separate agreement, and everyone must display "Built with Muse Spark Open".
Good for
- Coding and agent workloads that need long context and tool use
- Applications that combine text, images, video and audio understanding
- Organizations that want a Meta model they can host themselves
- Enthusiasts with very large unified memory who want to test a frontier model locally
Watch out for
- The download is gated and the weights need Hugging Face approval
- The license requires the "Built with Muse Spark Open" credit from everyone
- Hosting it as a service above US$20 million of revenue needs a separate agreement with Meta
- It writes text only, and the smaller Muse Glimmer is under different terms (Apache 2.0)
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Estimates for the Spark Open variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.
| Variants | Parameters | BF16 | FP8 / INT8 | 4-bit (Q4) | Fits on |
|---|---|---|---|---|---|
| Spark Open BF16 | 816B 42B active | 1.88 TB | 938 GB | 563 GB | GPU server |
| Spark Open NVFP4 NVFP4 | 816B 42B active | 1.88 TB | 938 GB | 586 GB | GPU server |
Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.
The license in plain words
Meta's license for Muse Spark Open. Most companies can use it commercially, but a business selling hosted access above US$20 million of revenue needs a separate agreement, and everyone must display "Built with Muse Spark Open". The download is gated.
- Separate deal needed to sell hosted access above a threshold
- Attribution or model name must be shown
See also
- Mistral Large 4 vs Muse Spark Open: two giant open-weights models
- Best open-source LLMs in 2026
- Best open-source multimodal models in 2026 (images, video and audio)
- Open models that listen: nine take audio, but only three answer in speech
- Ripple-65B Preview: a dense 65B agent model under Apache 2.0
- Giant open models now ship in 4-bit: what it means for memory planning
- Open licenses: free to build on, but hosting needs a deal
- Meta releases Muse Spark Open, an 816B open-weights model
Frequently asked questions
- What is Muse Spark Open?
- Muse Spark Open is Meta's most capable downloadable model. It understands text, images, video and audio, and it is built for coding and for acting as an agent. Meta says it can run on two high-end Apple Mac Studios, though most people will use it through a hosted service. Downloads need approval and the license has some conditions.
- Can I use Muse Spark Open commercially?
- Yes, with conditions. Muse Spark Open is released under the Muse Model License, which allows commercial use but adds extra rules: Separate deal needed to sell hosted access above a threshold; Attribution or model name must be shown. Read the license text before you ship a product.
- Is Muse Spark Open open source?
- Muse Spark Open is open-weight, not open source in the strict sense. You can download the model files, but the Muse Model License adds conditions that the Open Source Initiative definition does not allow.
- Can I run Muse Spark Open on my own computer?
- Not on a desktop. The smallest variant, Spark Open, needs about 563 GB at 4-bit precision, which means a GPU server with several data-center cards.
- When was Muse Spark Open released?
- Muse Spark Open was released by Meta on October 15, 2026.
- How long a context window does Muse Spark Open have?
- Muse Spark Open supports a context window of up to 1M tokens.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- Introducing Muse Spark Open (Meta AI Research, 15 Oct 2026) research.meta.ai
- Muse-Spark-Open-816B-A42B model card and Muse Model License (Hugging Face) huggingface.co
- Muse Spark Is Now an Open-Weights Model (PyTorch blog) pytorch.org
- Build with Muse Spark Open (Meta developer blog) dev.meta.ai
Release date and facts last verified November 8, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| Inkling Thinking Machines Lab | 276B – 975B 41B active | 1M | Apache 2.0 (Permissive) | 188 GB Big-memory desktop | |
| Mistral Large 4 Mistral AI | 1.05T 52B active | 1M | Mistral Open-Weight (Conditional) | 671 GB GPU server | |
| Muse Glimmer Meta | 29.6B | 128K | Apache 2.0 (Permissive) | 20 GB Gaming GPU | |
| Nex-N2.6 Nex-AGI | 35.1B – 1.6T | 256K | Apache 2.0 (Permissive) | 24 GB Gaming GPU |