Muse Glimmer is a 30-billion-parameter open-weight model released by Meta under the permissive Apache 2.0 license, marking Meta's first fully open release since it transitioned away from its earlier open-weight Llama line and introduced the proprietary Muse Spark. By shipping the weights openly and without the commercial restrictions that drew criticism to Llama's bespoke community license, Glimmer is meant to make a capable foundation broadly available to developers who want to fine-tune, redistribute, or self-host rather than depend solely on cloud APIs. The weights are published on Hugging Face, and Meta framed the launch as a return to a more familiar open-source posture for the community.

Glimmer is positioned specifically as an agentic model, optimized for running autonomous AI agents directly on high-end consumer machines such as Macs and PCs instead of relying on cloud infrastructure. That on-device focus, combined with a 30B parameter scale, suggests a practical fit for tool-using assistants and multi-step workflows where local inference, data control, or lower-latency execution matter, while still leaving room for cloud deployment when more capacity is needed. Because it is open-weight, integrators can adapt the model to specialized agent tasks, swap in domain-specific tools, or run it through local runtimes without licensing friction.