Meta has unveiled a new open-source AI model, Muse Glimmer, that runs directly on devices. What are its features?
Meta introduced the Muse Glimmer AI model with 30 billion parameters, optimized for constantly active local agent workflows.
Meta introduced the Muse Glimmer AI model with 30 billion parameters, optimized for constantly active local agent workflows.
Meta introduced the Muse Glimmer AI model with 30 billion parameters, optimized for constantly active local agent workflows.
This model is small enough to run on a Mac or PC with a single GPU and can run directly on the device without using the internet or cloud infrastructure.
«Locally running models allows you to use AI anywhere, anytime, with or without an internet connection. The open source community has shown that smaller models, when trained effectively, can achieve cutting-edge performance for targeted tasks. Muse Glimmer is optimized for these local use cases,» Meta blog notes .
Muse Glimmer is an open source model, and the corresponding developer documentation to help you get started building and running your own agents is published on Hugging Face.
The company also assured that optimized model integrations with llama.cpp, MLX, and ExecuTorch will be available in the coming days.
The model, among other things, is designed for complex execution of agent tasks, and achieves high success rates in the DeepSearch QA, MCP-Atlas, 𝛕-Bench, and SWE-Bench tests.
Muse Glimmer is also reported to work with OpenClaw and other agent orchestration patterns. The model is trained on data from over 100 languages.
In light of the release of this model, CEO Mark Zuckerberg also called for lowering barriers in the US for open-source AI to better compete with Chinese rivals.
Recall that at the beginning of the year, Meta launched Meta Compute, a separate direction that should accelerate the creation of infrastructure for artificial intelligence: new data centers, networks, and computing power.


