Meta Launches Muse Glimmer, New AI Model for Local Agents

Meta launched Muse Glimmer, a 30-billion-parameter language model optimized for AI agents that run locally on a continuous basis. The model is available under the Apache 2.0 license. Using 4-bit quantization, Muse Glimmer can be executed on Mac or PC computers with 24GB to 32GB of graphics memory, maintaining performance comparable to uncompressed models. Tests on MacBook M4 Max, M5 Max, and RTX 5090 demonstrated that the model provides sufficient speed for fluid conversations and real-time agent interactions, with all processing occurring locally. Compared to similarly-sized models like Gemma4-31B and Qwen3.6-27B, Muse Glimmer demonstrates competitive performance across various language model benchmarks.