A strong free option for developers running agents on their own computer, with Apache 2.0 weights and support for over 100 languages.
Kenning's take
What it does
Muse Glimmer is a 30 billion parameter open model from Meta Superintelligence Labs, distilled from Muse Spark and released under Apache 2.0. It is built for agents that run on a person's own computer, with tool calling, coding, image input and adjustable reasoning effort, and is trained on more than 100 languages. The weights are on Hugging Face, with support in tools such as llama.cpp, MLX and ExecuTorch.
- 30B parameters, runs on one consumer GPU
- Apache 2.0 open weights on Hugging Face
- Tool calling and multi-step agent tasks
- Text and image input
- Controllable reasoning effort
- Trained on more than 100 languages
Who it suits. Developers building private, local AI agents that work without a cloud connection.
Key facts
- Category
- AI models
- Part of
- Meta AI
- Made by
- Meta (Meta Superintelligence Labs)
- Released
- 10 August 2026
- How to get it
- free download of the weights from Hugging Face, with guides at dev.meta.ai for llama.cpp, vLLM, SGLang and ExecuTorch
- Works on
- self-hosted, macOS, Windows, Linux
- In the UK
- Yes
- Free plan
- free open weights
FAQ
Is Muse Glimmer open source?
Yes. Meta released the weights under the Apache 2.0 licence on Hugging Face.
Can Muse Glimmer run on a laptop?
Meta says it is small enough to run on a Mac or PC with a single consumer GPU.
When was Muse Glimmer released?
Meta announced it on 10 August 2026.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
Meta Llama models
Meta's family of downloadable large language models, from Llama 2 to the multimodal Llama 4
Muse Image
Meta's in-house image model that plans a layout before drawing, used in Meta AI and the Meta Model API
Muse Spark
The first model from Meta Superintelligence Labs, a multimodal reasoning model behind the Meta AI app
Muse Spark 1.1
Meta's July 2026 update to Muse Spark with a 1M-token context, computer use and the first public API