Meta released Muse Spark Open, an 816-billion-parameter open-weight multimodal model designed for coding and long-running agent tasks.
On October 15, Meta released Muse Spark Open, the downloadable-weight version of its Muse Spark model, aimed at coding and long-running AI-agent tasks. The natively multimodal mixture-of-experts model has 816 billion parameters, activates 42 billion per token, accepts text, images, video and audio, and supports a one-million-token context. Meta says it can run across two Apple M5 Ultra Mac Studios using Apple’s MLX framework; weights are available through Hugging Face and major cloud marketplaces, with hosted access from seven inference providers. Meta’s model card reports 61.8 on JobBench and 89.2 on MCP-Atlas, but 76.4 on Terminal-Bench 2.1, below Kimi K3’s 85.0. The release gives developers a customizable alternative to closed models and extends Meta’s open-weight challenge to US and Chinese rivals, though its Muse Model License imposes limited commercial terms. It follows Muse Spark’s April debut as a proprietary model; Meta calls the open release its most capable US open-weight model to date.