• Source:JND

Meta is making another push towards local AI with Muse Glimmer, a new model from its Meta Superintelligence Labs. The 30-billion-parameter model is built specifically for AI agents that can keep running locally, rather than constantly depending on cloud servers. Meta says it can run on a Mac or PC with a single consumer GPU, making it suitable for things such as local coding, function calling and AI agents that can work even without an internet connection.

ALSO READ: Amazon Great Freedom Sale 2026: Galaxy Watch 8 Gets Price Cut, More Watches Follow| Check Best Deals

There is a catch when it comes to hardware. Running Muse Glimmer at full precision requires more than 55GB of memory. Meta has used 4-bit quantisation to bring the model down to less than 20GB, which means it can fit on systems with 24GB or 32GB of memory. The company has also compared it with Google’s Gemma4-31b and Alibaba’s Qwen3.6-27B, claiming Muse Glimmer came out ahead in 12 tests.

The model is designed to handle more complex jobs than a typical chatbot. Meta says it can complete tasks using tools, work through multiple steps, identify errors and retry failed tool calls, while also handling multimodal inputs. It has been trained on data which covers more than 100 languages.

ALSO READ: Amazon Great Freedom Sale 2026: Top 5G Phones From Google, Samsung, Vivo And More Under Rs. 70,000

Meta has released the model weights under the Apache 2.0 licence, and they can be downloaded from Hugging Face. Support through Ollama, LM Studio and Unsloth is expected in the coming days. Meta is also working with AMD, Arm, Dell, Intel and Nvidia on hardware optimisation, with integrations for llama.cpp, MLX and ExecuTorch also in the works.


Also In News