Great news from the OGs of open source LLMs! Meta released Muse Glimmer-30B, a new multimodal model designed for local agentic use cases.

Distilled from Muse to 30B parameters and released under the Apache 2.0 license, it’s ideal for local deployment, privacy, cost‑reduction, or hacking. It targets privacy‑aware applications such as coding, document analysis, personal assistants, and Claw‑ or Hermes‑like setups. Meta ships day‑0 support in transformers parehong, llama.cpp, v LLM, Inference Endpoints, and other libraries.

Architecture-wise, the dense 30B model contains a 2B Vi‑T‑style vision encoder (Perception Encoder) and a 28B text decoder. An optional speculative decoding drafter boosts speed for structured content generation. Benchmarks show strong reasoning, coding, and multimodal performance.

Demos and tutorials are available on the Hugging Face Hub, and the community can fine‑tune or modify the model freely.