Meta Releases Muse Glimmer, a Small AI Model for Local Use

Meta has unveiled Muse Glimmer, a novel artificial intelligence model developed by its Superintelligence Labs. This new AI is distinguished by its compact size, enabling it to operate directly on a single consumer graphics card without requiring cloud connectivity.

“

The 30-billion-parameter model is being released with open weights under an Apache 2.0 licence, the company announced today.

The model boasts 30 billion parameters and is released under an Apache 2.0 license with open weights. Meta designed Muse Glimmer for continuous agent workflows, allowing it to handle tasks such as schedule management, message drafting, file organization, and executing multi-step tool commands autonomously. It supports both text and image inputs and can recover from failed tool calls, preventing stalled operations.

“

According to Meta, the model is designed for always-on agent workflows, meaning it can handle tasks such as managing schedules, drafting messages, organising files, and executing multi-step tool commands without needing a cloud connection.

To achieve its small footprint, Meta employed quantization, compressing the model to approximately 4-bit precision. This reduces its memory requirement to under 20 GB, fitting within the typical memory limits of consumer GPUs. Meta claims this compression results in minimal performance loss for agentic tasks, ensuring fast, real-time interaction on devices like MacBooks with M4/M5 chips or Nvidia RTX 5090 graphics cards. The model was trained using a distillation process from a larger teacher model and supports over 100 languages.

“

To fit on consumer hardware, Meta says it used quantisation to compress the model to roughly 4-bit precision, bringing its memory footprint below 20 GB.