Meta released Muse Spark Open on October 15, describing it as "the most capable US open-weight model to date." Open-weight means anyone can download and run the model, rather than only reaching it through a company's API.

Meta says the model is built for coding and agentic work, and that it performs well on long-horizon tasks, complex software engineering, and multi-step reasoning. Its stated specifications:

  • 816 billion total parameters, with 42 billion active at a time
  • a 1 million token context window
  • native multimodal input across text, image, video, and audio

Where you can run it

Meta says Muse Spark Open is optimized to run on two Apple M5 Ultra Mac Studios using the MLX framework, and that it also works with common inference engines including SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, llama.cpp, and Unsloth, with optimizations for AMD, Apple, Intel, and NVIDIA hardware. Developers can fine-tune it with PyTorch, including TorchTitan.

The weights are available on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models. Hosted API access is offered by Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI.

The fine print worth reading

Two details matter if you plan to build on it. First, Meta says the weights are released under the "Muse Model License," which it describes as having limited commercial terms that still let many developers use the model for free. As with most open-weight releases, it is worth reading the license rather than assuming it matches a standard open-source one. Second, "open weight" does not mean light: a model with this many parameters still needs serious hardware, which is why Meta frames two high-end Mac Studios as the local target.

Meta also introduced the Muse Wonder series on October 12, which it positions for personal and professional agent tasks and offers through its own API and Muse Code rather than as open weights. Muse Spark Open is the open follow-up the company had already signaled was coming.

Source: Meta's announcement on X.