Meta Platforms has released Muse Glimmer, a 30 billion-parameter model from Meta Superintelligence Labs, with weights published under a permissive Apache 2.0 licence and available on Hugging Face. Meta positions the model for always-on local agent workflows, stating it runs on a Mac or PC with a single consumer GPU, and lists local agents, function calling, local coding and LLM-as-a-judge evaluation as target use cases. Meta argues that local execution lets users run AI with or without an internet connection, unlike most foundation model deployments that depend on cloud infrastructure. The company describes two optimisations: at full precision, a 30 billion-parameter model would need over 55GB of memory, but quantisation compresses weights to roughly 4-bit precision, cutting the language model to under 20GB, with the remaining headroom holding the KV cache, perception encoder and a speculative decoding drafter within a 24GB or 32GB envelope. Meta validated the compression as causing minimal to no degradation on agentic tasks. The release includes a lightweight drafter based on DFlash that proposes blocks of tokens, which the main model verifies in parallel, running faster than token-by-token decoding with identical output quality. Meta measured throughput on a K-Quant-17GB configuration with the quantised drafter on MacBook M4-Max and M5-Max machines and an RTX-5090. Training ran in three phases: pre-training used logit