Meta Superintelligence Labs has shaken up the artificial intelligence landscape with the release of Muse Spark Open, an 816-billion-parameter multimodal open-weight model designed explicitly for advanced coding and long-horizon agentic workflows. Billed by Meta as the most capable United States open-weight model to date, the system features 42 billion active parameters and a sprawling 1-million-token context window. According to reports from AI Breakfast and TLDR AI, the release arrives amid fierce competition in the foundational model space, contrasting with rivals like Microsoft that have chosen to keep their massive base models strictly closed.
What sets Muse Spark Open apart is not just its raw scale or benchmark performance, but its deployment versatility. While traditional enterprise deployments rely heavily on cloud infrastructure, Meta has engineered Muse Spark Open to run locally. By leveraging Apple's unified memory architecture and the latest MLX framework, developers can run the massive model across two Apple M5 Ultra Mac Studios. Apple's integrated GPU allows each Mac Studio to use up to 512 GB of unified memory, combining two machines to act as a single supercomputer with a terabyte of memory and 1.2 TB/s of bandwidth, according to Meta's research blog.
For founders, builders, and engineering leaders, this release lowers the barrier to deploying enterprise-grade agentic infrastructure. As detailed in coverage by TechCrunch, Meta paired the model with Muse Code to successfully execute complex software engineering tasks, such as shipping real iOS changes to an open-source Mastodon client, including bug fixes, new features, and full builds. Furthermore, the model demonstrated the ability to drive end-to-end media workflows in DaVinci Resolve, reviewing footage frame by frame to find strong moments and executing vertical crops.
From an infrastructure perspective, accessibility is broad. Beyond local execution on Apple silicon, Muse Spark Open is optimized for NVIDIA, AMD, and Intel hardware, supported by inference engines such as SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, llama.cpp, and Unsloth. Developers can access the model via major cloud providers and API platforms including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, Together AI, Amazon SageMaker AI, Google Cloud Model Garden, and Microsoft Foundry Models.
For business leaders, open-weight models like Muse Spark Open offer critical strategic advantages over proprietary alternatives, including cost efficiency, architectural transparency, and data privacy. Meta is releasing the weights under the Muse Model License with limited commercial terms, allowing developers to utilize the model for free while maintaining compliance with the Meta Superintelligence Scaling Framework for safety and risk management.