
Muse Glimmer
An open 30B multimodal model for always-on local agents.
Live drop
MetaDropped Aug 10, 20269 builds8 makers3 prompts
Open 30B agentic model running locally on one consumer GPU with strong tool use and failure recovery.



About this release
30-billion-parameter open-weights model (Apache 2.0) distilled from larger Muse Spark models, optimized for always-on local agents. Supports long-horizon execution, precise tool calling, multimodal inputs (text+images), error recovery, and controllable reasoning; runs quantized on consumer hardware via integrations like llama.cpp, MLX, and ExecuTorch.
Read the official Meta launch pageWorth opening first
Paste the URL and Dropday.ai will credit you, track the build as it evolves, and put it on this drop.
Submit a build