Muse Glimmer: Meta open-sources a 30B model for always-on local agents
Meta has open-sourced Muse Glimmer, a 30B-parameter Apache-2.0 model tuned for always-on local agents, tool calling, coding, multimodal input, and failure recovery. Meta says its K-Quant-17GB variant is intended for 24–32 GB devices and that a DFlash drafter raised decode speed up to 3.1× on RTX 5090; the hardware target is a high-end consumer GPU, not a 6 GB card.