Meta releases Muse Spark Open AI model with 816 billion parameters; runs on two Apple Mac Studios

Meta has released Muse Spark Open, an open-weight artificial intelligence (AI) model designed for coding, complex reasoning and tasks that require multiple steps. The company says the model can run on a desktop setup using two Apple Mac Studios powered by M5 Ultra chips, making it possible to use an advanced AI model locally rather than relying only on cloud services. As stated by the company in its official announcement, Muse Spark Open has 816 billion total parameters, with 42 billion active parameters, and supports a context window of 1 million tokens. Developers can access the model through several AI platforms or download its weights to customize it for their needs.Meta Muse Spark Open: Features and capabilitiesMeta said Muse Spark Open is designed for long-running tasks, complex software development and agentic AI workflows, in which a model can carry out multiple steps towards a goal.The model is natively multimodal, meaning it can work with different types of information. It has 816 billion total parameters, of which 42 billion are active, and a context window of 1 million tokens. A large context window allows a model to process a substantial amount of information in a single interaction.Meta said it tested the model on tasks involving software development and video editing. In one test, the model was used to make bug fixes and add features to Ice Cubes, an open-source Mastodon client, before building the application. In another, it worked with DaVinci Resolve to review video footage frame by frame, identify suitable moments and crop clips into a vertical format.The company also said the model includes optimised GPU kernels that deliver kernel-level speed improvements of 1.03 to 3.95 times on NVIDIA and AMD GPUs. Meta reported a 3% overall speed improvement on AMD MI350.How to access Muse Spark Open AI modelMeta says Muse Spark Open is available through AI inference platforms including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. Developers can download its model weights through Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. The model supports inference engines such as SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp and Unsloth. Developers can also customise it using PyTorch.For local use, Meta has optimised Muse Spark Open for Apple's M5 Ultra Mac Studio through the MLX framework. According to the company, two Mac Studios can work together to provide enough unified memory to hold the model.Meta said it has added safety training to the open-weight version and evaluated it under its Superintelligence Scaling Framework. The company said the model met the safety thresholds required for release.The model is being released under the Muse Model License, which allows many developers to use it for free but includes limited commercial terms. Its weights are available through the Muse Spark Open collection on Hugging Face.You use AI every day. Now get your AI Quotient. Take the AIQ test.