Meta has released what it calls the most capable open-weight A.I. model built in the United States to date — Muse Spark Open — in a notable collaboration with Apple that lets the model run locally on Macs.
The model is natively multimodal and built for coding and agentic work, with 816 billion total parameters (42 billion active) and a 1-million-token context window. Meta says it excels at long-horizon tasks, complex software engineering, and multi-step reasoning, and it outperforms other popular open-weight models in the company's evaluations.
The headline feature is local execution on Apple silicon. Muse Spark Open is optimized to run on the Mac Studio equipped with the M5 Ultra chip using Apple's MLX framework. Because the model is large, it needs at least 1 terabyte of memory: Apple's unified-memory architecture lets each M5 Ultra Mac Studio use up to 512 GB, so two machines effectively behave like one desktop supercomputer with a terabyte of unified memory.
Meta AI chief Alexandr Wang called the partnership a moment when the two longtime rivals are "playing nice again," saying the companies had worked together for several months. Muse Spark Open is available now through API platforms including Baseten, Fireworks AI, Groq, and OpenRouter, with weights downloadable from Hugging Face. Meta paired it with its Muse Code coding terminal for agentic workflows, including shipping real changes to an open-source Mastodon client.