Meta has released Muse Glimmer, a 30 billion parameter model available under an Apache 2.0 licence. This shift from their previous restrictive terms marks a return to the open weights market. The release targets local deployment on machines with at least 32 GB of RAM. Simon Willison, who tested the model on his 128 GB system, notes that this size leaves sufficient memory for other applications to run simultaneously.
The model focuses on end-to-end task completion rather than simple text generation. Benchmarks show it can write and debug code while managing multi-turn requests. It also handles function calls with precise schemas during extended workflows. Willison demonstrated this by asking the model to explain authentication in a fresh checkout of Datasette. The agent successfully explored the codebase and produced a response after numerous tool calls.
* The model weighs 18.16 GB in its LM Studio version
* It runs on local hardware without cloud dependency
* Apache 2.0 licensing removes previous distribution restrictions
