Meta and Apple announce Muse Spark Open, an 816-billion-parameter open-weight AI model optimized to run on Apple’s Mac Studio hardware locally. Meta says the model is aimed at coding and “agentic” workflows, including multi-step reasoning and extended tasks. Multiple outlets report it uses a mixture-of-experts design with 42 billion active parameters per token and supports a large 1-million-token context window.

The partnership centers on running the model across clustered Mac Studio systems using Apple’s MLX framework and unified memory. Most coverage says developers need two M5 Ultra Mac Studios with 512GB unified memory each to fit the model, effectively pooling the memory resources; a configuration with four 256GB units is also described as an alternative. Several outlets note practical timing and cost constraints, including that the 512GB Mac Studio option is expected to arrive later in October and that 256GB configurations are already expensive. Availability details are consistent across reports: Muse Spark Open is free to download, weights are distributed via major model-hosting and cloud platforms, and the model can be accessed through multiple third-party API providers with limited commercial-use licensing mentioned by Meta.