Meta Muse Glimmer 30B: Open-weight agentic model for local deployment

Meta’s technical-report announcement reinforces Muse Glimmer as an Apache-2.0 30B model reportedly quantizable below 20 GB, with day-zero llama.cpp/GGUF and several local-runtime integrations. DistressBench reports 80.6%, but the crisis-support task is narrow; independent usable-memory, throughput, AMD, and reproducibility results remain absent.