Meta has launched Muse Glimmer, a 30-billion-parameter AI designed to run directly on high-end personal computers, marking a shift towards decentralised, user-controlled AI technology.
Meta has released Muse Glimmer, an open-weight AI model designed to run on personal computers rather than in the cloud, as the company continues to push a strategy built around local, controllable AI. According to the announcement reported by AP, the model is intended for broad distribution and reflects Mark Zuckerberg’s argument that AI should not be concentrated in the hands of a few large platforms. Meta is positioning it alongside Muse Spark 1.2, the more capable model from its Superintelligence Labs.
The company says Muse Glimmer is a 30-billion-parameter model built for “always-on local agent workflows”, meaning it is meant to perform tasks directly on a user’s machine. That local approach can reduce dependence on remote servers and keep more data on the device, but it also places heavy demands on hardware. Meta says the model was trained through pre-training, mid-training and post-training stages to improve agentic behaviour, multi-step reasoning and recovery when a tool call fails, and it was trained on data covering more than 100 languages.
Meta argues that Muse Glimmer is efficient for its size, but it still will not suit every laptop. The company says a 30-billion-parameter model would normally need more than 55GB of memory, although quantisation reduces that requirement to under 20GB, with working memory pushing practical use closer to 24GB or 32GB. Meta says it tested the model on MacBook M4 Max and M5 Max systems and an RTX 5090 GPU, which suggests the target audience is users with high-end consumer hardware rather than standard notebooks.
On performance, Meta says Muse Glimmer beats Google’s Gemma4-31b and Alibaba’s Qwen3.6-27B across 12 benchmarks, although those claims stop short of comparisons with OpenAI, DeepSeek, Z.ai or Moonshot AI. Independent rankings cited by Lifehacker place Muse Glimmer well behind the strongest open-weight systems currently available, with Qwen3.6-27B slightly ahead and Gemma4-31b further back. The model is already available through Hugging Face, and Meta says support is coming to tools such as Ollama, LM Studio and Unsloth.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
