Meta has released Muse Glimmer, a slimmed-down "open source" AI model light enough to run on a single computer. Based on Meta's Spark 1.2 closed model, it requires just a single GPU for agent-oriented tasks like scheduling and file management. "We designed Muse Glimmer to balance capability against the memory and compute constraints of local hardware," the company wrote. The model weights and developer documentation are available for free on Hugging Face, allowing users to run the model on their own PCs. Optimized integrations will soon be available on llama.cpp and other platforms, enabling quick setup from download to working agent.

According to Meta, Muse Glimmer delivers strong performance on benchmarks like Deep Search QA, MCP-Atlas, and SWE-Bench, showcasing its ability to write and debug code. It supports reliable tool use, multi-step reasoning, failure recovery, multimodal input, and compatibility with agent orchestrators like Open Claw. The model was trained on data from over 100 languages.

This release mirrors efforts from Chinese competitors like Deep Seek, which also offer locally runnable models with more open licensing. "Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it," said CEO Mark Zuckerberg. Muse Glimmer appears to be Meta's push toward a more open-source AI strategy, appealing to users who prefer running models locally.

The move comes amid growing interest in local AI deployment, highlighted by NVIDIA's recent launch of the Open Secure AI Alliance focused on cyber defense following a recent attack on Hugging Face by an unreleased model.