Meta released an AI model that works locally on a computer and does not need cloud services

Meta introduced Muse Glimmer - an open-weight artificial intelligence model that can be run directly on a personal computer without a constant connection to cloud services.

The model has 30 billion parameters and was developed by the Meta Superintelligence Labs division. It is distributed under the Apache 2.0 license, which allows downloading, modifying, and using it in your own projects.

What computer is needed

Muse Glimmer can work on a Mac or PC with a single consumer graphics card. At full precision, the model would require over 55 GB of memory, but after 4-bit quantization, its size was reduced to less than 20 GB.

To run the model together with image processing components, a system with 24-32 GB of available memory is needed. Meta tested it specifically on Nvidia RTX 5090, Apple M5 Max, and M4 Max.

What Muse Glimmer can do

The model is primarily intended for agentic tasks: it can plan actions, perform multi-step operations, use external tools, and recover from failed calls.

Muse Glimmer supports working with text and images, can analyze documents, charts, and screenshots, write code, and perform administrative tasks. The training data covers over 100 languages.

The model was created by distillation based on the more powerful Muse Spark system. Muse Glimmer weights have been published on Hugging Face, and support for llama.cpp, MLX, ExecuTorch, Ollama, and LM Studio is planned to be added in the near future.

Based on materials from: Business Insider, Fortune India