Meta releases Muse Spark Open — ~816B multimodal MoE, one million tokens
Meta published Muse Spark Open, an open-weight multimodal Mixture-of-Experts model for text, images, audio and video. It has ~816B parameters, ~42B active per token and a one million tokens context window.
What changed
Meta released Muse Spark Open, an open-weight multimodal MoE model with ~816B total parameters (~42B active per token) and a one million tokens context window.
Why it matters
Developers can run agentic coding workflows that read a codebase, write and debug code, and call tools on multimodal inputs across long contexts.
Who should care
DevelopersResearchers
Worth trying
Maybe — Open checkpoints but needs large memory and GPUs (BF16 about 1.5 TB; H200/B200)