The shift towards running advanced AI tools directly on personal computers is gaining momentum as Meta introduces Muse Spark Open, an open-weight model designed for coding, software engineering and complex, multi-step tasks. The company has partnered with Apple to bring the model to Mac Studio desktops powered by the M5 Ultra chip.
Muse Spark Open brings large-scale AI to desktops
Meta describes Muse Spark Open as the most capable US open-weight model to date. It has 816 billion total parameters, including 42 billion active parameters, and supports a 1-million-token context window. The model is designed to handle long-running tasks, complex software development and multi-step reasoning.
With Apple’s MLX machine learning framework, the model can run locally across 2 Mac Studio systems. Meta says the setup uses Apple’s unified memory architecture, allowing the integrated GPU in each M5 Ultra to access up to 512GB of memory. Together, the 2 systems can provide up to 1TB of unified memory to accommodate the model’s weights, effectively working as a single supercomputer for local AI workflows.
Muse Spark Open also supports several inference engines, including SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, MLX and llama.cpp. It is designed to run on hardware from AMD, Apple, Intel and Nvidia. Developers can customise the model using PyTorch and TorchTitan.
Performance, availability and licensing
The model introduces custom, auto-tunable Helion GPU kernels in SGLang. Meta claims these can deliver up to 3.95 times faster kernel-level performance on Nvidia and AMD GPUs, along with a 3% end-to-end speed improvement on AMD MI350 hardware.
Developers can access the model through API platforms such as Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. Its weights are also available through Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models.
Meta has released Muse Spark Open under its Muse Model License, which allows free developer use subject to certain commercial licensing terms.
For security, the company says it conducted safety evaluations under its Superintelligence Scaling Framework and that the model meets the thresholds required for release.
Also read: Viksit Workforce for a Viksit Bharat
Do Follow: The Mainstream LinkedIn | The Mainstream Facebook | The Mainstream Youtube | The Mainstream Twitter
About us:
The Mainstream is a premier platform delivering the latest updates and informed perspectives across the technology business and cyber landscape. Built on research-driven, thought leadership and original intellectual property, The Mainstream also curates summits & conferences that convene decision makers to explore how technology reshapes industries and leadership. With a growing presence in India and globally across the Middle East, Africa, ASEAN, the USA, the UK and Australia, The Mainstream carries a vision to bring the latest happenings and insights to 8.2 billion people and to place technology at the centre of conversation for leaders navigating the future.