Meta unveiled a new AI model, Muse Glimmer, designed to run directly on consumer computers, allowing developers to build and use AI agents without relying on cloud infrastructure or an internet connection. Announced on Monday by Meta Superintelligence Labs, the model was released with its weights under the Apache 2.0 licence. Meta said Muse Glimmer is optimised for local agentic workflows, including coding, function calling, tool use and AI evaluation, and is small enough to run on a Mac or PC with a single consumer GPU. The model is currently available through Hugging Face, with integrations for platforms including Ollama, LM Studio, llama.cpp, ExecuTorch and MLX expected in the coming days. Muse Glimmer supports agentic tasks such as managing tool calls, sustaining multi-step reasoning, recovering from failed tool calls, and working with text and images. It can interpret screenshots, charts and documents through a dedicated perception encoder, and was trained on data covering more than 100 languages. Meta trained Muse Glimmer using distillation, supervised fine-tuning and reinforcement learning, using outputs from its larger Muse Spark model. The company stated that compared with Gemma4-31B and Qwen3.6-27B, Muse Glimmer performs strongly for its size class on several widely used LLM benchmarks. To address memory requirements, Meta used quantisation to reduce the model to approximately 4-bit precision, bringing its size below 20GB, allowing it to run within a