Meta releases Muse Glimmer 30B open-weight model optimized for local agentic workflows

Meta Superintelligence Labs released Muse Glimmer, a 30-billion-parameter open-weight model under Apache 2.0 license, optimized for always-on local agent workflows that can run on a Mac or PC with a single consumer GPU. The model was trained via distillation from Muse Spark and supports end-to-end agentic task completion, reliable tool use, multi-step reasoning, failure recovery, multimodal input, and controllable effort. Weights are available on Hugging Face with documentation for local deployment via tools such as llama.cpp, MLX, Ollama and vLLM. Quantization and speculative decoding enable practical on-device performance.