HelloPixDig

Directory › Development › Muse Spark Open

Muse Spark Open
Image: official project repository

> GEM DEVELOPMENT · AGENTS & AUTOMATION · LOCAL AI

Muse Spark Open

by Meta

Meta's largest open-weight model: 816 billion parameters, built for coding and agents, with text, image, audio and video input.

Muse Model License FreeOpen weightsLocal
View on Hugging Face →

What it does

  • 816 billion parameters (42 billion active per token), 1-million-token context, text output
  • Near-free license, EU included, with conditions: “Built with Muse Spark Open” notice and a deal with Meta for API resale above $20M
  • Needs ~300 GB of memory at minimum: in practice, an API or a big Mac / server

! “Muse Model License”: commercial use allowed, with no territorial exclusion (EU included). Conditions: show “Built with Muse Spark Open” in your product, follow the usage policy (18+ only), and sign an agreement with Meta if you resell access to the model through an API and your revenue exceeds $20M. Downloading the weights requires accepting the terms on Hugging Face. Figures come from Meta and Hugging Face, not tested by PixDig.

Install it

Requirements: A Mac or server with at least 300 GB of memory. First accept the license terms on the model's Hugging Face page. Without that hardware, use an API (OpenRouter, Together AI…).

Install llama.cpp
curl -LsSf https://llama.app/install.sh | sh
Start the server with the 2-bit compressed version
llama serve -hf ggml-org/Muse-Spark-Open-816B-A42B-GGUF:Q2_K
Open the built-in web interface
http://localhost:9931

The prompt for your agent

Paste it into your coding agent: it checks the requirements and installs it for you.

Claude CodeCodexCursorGemini CLIopencode
Prompt
Install Muse Spark Open (https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B) on my machine.
1. Check the requirements: A Mac or server with at least 300 GB of memory. First accept the license terms on the model's Hugging Face page. Without that hardware, use an API (OpenRouter, Together AI…). If anything is missing, tell me before installing anything.
2. Follow the official method:
- Install llama.cpp: curl -LsSf https://llama.app/install.sh | sh
- Start the server with the 2-bit compressed version: llama serve -hf ggml-org/Muse-Spark-Open-816B-A42B-GGUF:Q2_K
- see http://localhost:9931
3. Check that it works, then explain how to launch it and where to start.

Always review what your agent is about to run, especially install commands.