Meta Releases Muse Glimmer Open‑Source 30B Model To Boost AI Agent Development

Meta announces release of Muse Glimmer, a 30B‑parameter AI model licensed under Apache 2.0, now publicly downloadable. The model follows Meta’s Llama series and is positioned as the first open‑source offering after the proprietary Muse Spark launch, aiming to support a broad range of applications.

Developers can integrate Muse Glimmer into conversational agents, benefiting from its optimization for agent‑centric tasks such as tool use and memory management. Open‑source access lets researchers modify the 30‑billion‑parameter architecture, reducing reliance on closed‑source alternatives and accelerating experimentation.

Industry analysts say Muse Glimmer may raise the share of open‑source models powering downstream applications, noting that Meta’s Llama family already underpins roughly 38% of adapted AI systems. The move could pressure competitors to release comparable licenses and broaden the ecosystem for AI agents.

Researchers worldwide can download the pre‑trained weights from Meta’s GitHub repository, which includes documentation for fine‑tuning on custom datasets. Early benchmarks show Muse Glimmer matching Llama‑2 performance on standard language tasks while using fewer compute resources.