Meta releases Muse Glimmer: a 30B open agentic model under Apache 2.0

Meta has released Muse Glimmer, a 30-billion-parameter open-weights model, under an Apache 2.0 license, according to a blog post on the company’s research site, “Introducing Muse Glimmer,” flagged by developer and commentator Simon Willison on his blog on August 10.

Willison notes the release marks a return by Meta to the open weights space, calling the Apache 2.0 terms “a step up from the janky Llama licenses of old,” a reference to the licensing restrictions attached to Meta’s earlier Llama model family.

Per Willison’s summary of Meta’s announcement, the company states it optimized Muse Glimmer for agentic task completion, describing the goal as “end-to-end agentic task completion,” among a list of capabilities Meta highlights for the model. Willison writes that this focus aligns with what he looks for in models meant to run locally, on a user’s own device rather than through a cloud service.

The provided material does not include further specifics from Meta’s post beyond the license terms, the parameter count, and the partial capability list quoted above. No release date for general availability, benchmark figures, or additional technical detail appear in the excerpted source.

Willison’s post links directly to Meta’s research blog for the full announcement. The Specola’s log records only what appears in the cited summary.