Meta has released Muse Spark Open, an open-weight model built for long-context coding and autonomous AI tasks that can run locally on two Apple M5 Ultra Mac Studio machines.
Meta has released Muse Spark Open, a new open-weight model the company says is its most capable US-built system yet for coding, software engineering and AI agents. The model is native multimodal, meaning it can work across different forms of input, and Meta says it is intended for tasks that require sustained planning and long-context reasoning. It arrives as major AI developers continue to push beyond chat-style tools towards systems that can perform longer, more autonomous work.
The model’s scale is central to Meta’s pitch. According to the company, Muse Spark Open has 816 billion total parameters, with 42 billion active at a time, and a 1 million-token context window. In practical terms, that gives it room to process very large prompts, documents or codebases in one session. Meta says the model is tuned for multi-step work, including software development and other agentic tasks that require more than a single response.
Meta has also used demonstrations to show what the model can do. In one example, it helped produce and ship code changes for Ice Cubes, the open-source Mastodon client. In another, it was paired with DaVinci Resolve to inspect video clips, identify notable moments and generate vertical crops automatically. The company says these examples reflect how the model is meant to function in real workflows rather than in narrowly controlled benchmarks.
The launch also highlights an unusual hardware collaboration. Meta says Muse Spark Open can run locally on two Apple M5 Ultra Mac Studio machines using Apple’s MLX framework, which is designed to make efficient use of the company’s unified memory architecture. Axios reported that the partnership marks a rare alignment between two firms that have often been rivals, while MacRumors said Meta worked with Apple to bring the model to the Mac. The model is also being distributed through cloud platforms including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI, with weights available on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models. Meta says it has added custom Helion GPU kernels in SGLang to improve performance on AMD and NVIDIA hardware, and that the release went through extra safety testing before launch.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
