Meta released the weights for Muse Spark Open on October 15, introducing an 816-billion-parameter AI model for coding, long-running tasks and software agents. Developers can access it through hosted services or run it locally on supported Apple hardware.
Muse Spark Open’s core design
Meta describes Muse Spark Open as a sparse mixture-of-experts model with 816 billion total parameters and about 42 billion active for each token. It accepts text, images, audio and video, and supports a context window of 1,048,576 tokens, according to Meta’s announcement and documentation.
Meta designed the model for agentic work. Its developer report says agents built with it can read a large codebase, plan changes, call tools, check results and correct failed steps. Meta also says it tested the model in workflows involving iOS development and video editing.
Local use remains demanding
For local use, Meta says Muse Spark Open can run on two Apple M5 Ultra Mac Studio computers through Apple’s MLX framework. Meta’s hardware documentation specifies two systems with 512 GB of unified memory each for that Apple configuration.
Developers who do not run the model locally can use hosted services from Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. Meta also lists Muse Spark Open through Hugging Face and other model catalogs.
License limits commercial use
Meta released Muse Spark Open under its Muse Model License. The company says the license allows many developers to use the model for free but includes limited commercial terms. The release therefore does not provide unrestricted commercial licensing.
Muse Spark Open gives developers an open-weight option for coding tools, customer-service agents and other software that works across multiple steps. Its hardware requirements make hosted access an alternative to running the model locally.
References
Axios. (2026, October 15). Apple and Meta bury the hatchet—for AI. https://www.axios.com/2026/10/15/apple-and-meta-bury-the-hatchet-for-ai
Meta AI Research. (2026, October 15). Introducing Muse Spark Open, the most capable US open-weight model to date. https://research.meta.ai/blog/muse-spark-open
Loving, A., Shamsoshoara, A., de Paola, B., Kavianihamedani, H., Khan, S., Potu, S. R., Subramanian, S., Choudhury, S., & Vontimitta, V. (2026, October 15). Build with Muse Spark Open. Meta AI. https://dev.meta.ai/resources/blog/build-with-muse-spark-open
Meta AI. (n.d.). Muse Spark Open. Retrieved October 16, 2026, from https://dev.meta.ai/docs/muse-spark-open
Meta AI. (n.d.). Hardware platforms. Retrieved October 16, 2026, from https://dev.meta.ai/docs/muse-spark-open/hardware-platforms