Meta released Muse Spark Open on October 15, offering downloadable model weights for coding and tool-using agents. Meta describes an 816-billion-parameter model with 42 billion active parameters and a one-million-token context window. It also lists hosted API partners and support for major inference frameworks. This is a model release: the announcement does not say the complete Muse personal-agent product can now be deployed locally.
The desktop example still involves substantial hardware: Meta describes running the model across two M5 Ultra Mac Studios using MLX. Its on-device demonstrations include iOS development and DaVinci Resolve workflows. Those are company-run examples, not our tests. For teams considering self-managed agents, the announcement creates another deployment option, but does not establish the cost or reliability of a particular everyday workflow.
The Muse Model License requires attribution. Model-as-a-Service providers whose aggregate gross revenue, including affiliates, exceeds US$20 million in any consecutive twelve months must notify Meta and obtain a separate written agreement before offering the model or derivatives through that business. Internal use and products outside Model-as-a-Service are exempt from that provision. Editorial take: Greater control over where a model runs is useful, but memory capacity, operating costs, licensing and permission controls still belong in the evaluation.