Seonglae Cho
Open
Muse Spark Open uses a mixture-of-experts architecture with 816B total parameters, 42B active parameters per token, and a 1M-token context window. Meta reports running it locally on two M5 Ultra Mac Studios using MLX. The open weights are distributed under the Muse Model License with limited commercial terms.
wonder
Muse Spark Open: 816B total, 42B active
Muse Spark Open accepts text, images, audio, and video and produces text with a 1M-token context window. Its MoE backbone routes each token to 8 of 256 experts. Meta reports local execution on two M5 Ultra Mac Studios using MLX; sufficient memory for the complete weights is still required despite sparse activation. The weights use the Muse Model License with limited commercial terms.