Meta · 2026-10-15 · seismic
Muse Spark Open — Meta's 816B open-weight model for coding and agents
Muse Spark Open is Meta's new open-weight Muse model: 816B total and 42B active parameters, a 1M-token context, and text, image, video and audio input. Meta says it runs on two Mac Studios with M5 Ultra chips.
Meta puts an 816B Muse model with a 1M-token context into open weights, sized to run on two Mac Studios.
Quick facts
| Maker | Meta Superintelligence Labs |
|---|---|
| Parameters | 816B total, 42B active (MoE, 256 experts, 8 active) |
| Context window | 1M tokens |
| Input / output | Text, image, video, audio in; text out |
| License | Muse Model License (limited commercial terms) |
| Weights | BF16 and NVFP4 on Hugging Face |
Benchmarks
| Muse Spark Open | 76.4% | |
|---|---|---|
| Kimi K3 | 85% | |
| GLM 5.3 | 83.9% | |
| Qwen3.8-2.4T-A95B | 82% | |
| Beam | 80.1% |
What is it?
Muse Spark Open is an open-weight model from Meta Superintelligence Labs, released on 15 October 2026 for coding and long agent tasks. Meta calls it the most capable US open-weight model to date. Until now, the only open-weight Muse model was the 30B Muse Glimmer.
How does it work?
Under the hood, Muse Spark Open is a mixture-of-experts transformer with a perception encoder: 256 experts, 8 active per token, so only 42B of its 816B parameters work on each token. Reasoning effort can be set from low to xhigh. Meta ships BF16 weights and an NVFP4 build, and lists support in SGLang, vLLM, TensorRT-LLM, Transformers, MLX and llama.cpp.
Why does it matter?
A model this size usually needs a GPU server, but Meta says two M5 Ultra Mac Studios with 512 GB each can hold Muse Spark Open, and it used that setup to ship iOS changes with Muse Code. On Meta's own table it leads MCP-Atlas at 89.2 but trails Kimi K3, GLM 5.3 and Qwen3.8 on Terminal-Bench 2.1, so the gain is in tool use more than terminal coding.
Who is it for?
teams that self-host models, local-AI builders with big Macs, agent developers
Frequently asked questions
- What hardware does Muse Spark Open need?
- The Muse Spark Open model card asks for at least 700GB of GPU memory for the NVFP4 build or 2TB for BF16. Meta also says the model is optimized for two Apple Mac Studios with M5 Ultra chips using MLX, where each machine offers up to 512 GB of unified memory, so the pair holds the full model.
- Is Muse Spark Open free for commercial use?
- Muse Spark Open ships under the Muse Model License, not an OSI license such as Apache 2.0. Meta describes the license as having limited commercial terms that still let many developers use the model for free. Meta's smaller open model, Muse Glimmer, uses Apache 2.0, so check the Muse Model License text before a commercial deployment.
- Where can I run Muse Spark Open without my own hardware?
- Meta lists hosted Muse Spark Open APIs at Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. The weights are also available through Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. Meta's blog post does not publish prices, so per-token rates depend on each provider.
- How does Muse Spark Open compare to Kimi K3 and GLM 5.3?
- In Meta's model-card table, Muse Spark Open scores 89.2 on MCP-Atlas, ahead of Kimi K3 (82.3) and GLM 5.3 (84.2), and 26.1 on HLE-Diamond without tools versus 22.2 and 16.4. It trails both on Terminal-Bench 2.1 (76.4 vs 85.0 and 83.9) and on the Artificial Analysis Intelligence Index (39 vs 44 and 45).
Try it
vllm serve "meta-models/Muse-Spark-Open-816B-A42B"