Meta has released Muse Glimmer, a 30-billion-parameter open-weight model built specifically to run AI agents locally on a single consumer GPU. Distributed under an Apache 2.0 licence and available on Hugging Face, the model is tuned for tool use, long-running tasks and failure recovery — the skills that matter when an agent is expected to work continuously on a user's own machine.
Meta positions Muse Glimmer for 'always-on local agents': because everything runs on-device, there is no per-token API bill, no cloud round-trip latency and no need to upload private data to a server. The release continues Meta's push to shrink frontier-grade agentic capability down to hardware people already own, and the company says open weights for Muse Spark 1.2 are on the way.
The model's launch comes as the industry debates where agents should live — in big cloud platforms that coordinate fleets of models, or on the devices where the work actually happens. Muse Glimmer is Meta's strongest bet yet on the second camp, giving developers a free, tunable foundation for assistants that stay awake, stay local and keep their data close.
Analysts note the timing: with consumer GPUs now common in gaming rigs and workstations, a 30B model that fits on one card makes credible, private, real-time agents possible for a much wider slice of developers — and challenges cloud-first rivals on price and privacy.
