Meta's Muse Glimmer brings AI agents to one consumer GPU

Meta's Muse Glimmer is a 30-billion-parameter open-weight model built for local agents, not just chat. Meta released the weights under Apache 2.0 and says the model can run on one consumer GPU or Mac. 1

The hardware claim needs context. Meta says full precision would require more than 55 GB of memory; its four-bit build falls below 20 GB and fits within a 24 GB or 32 GB memory envelope. The model is trained for tool use, coding, long tasks, multimodal input and recovery after failed tool calls. 2

That makes local agents more practical for private files and offline work, but it does not make the hardware cheap. NVIDIA's deployment guide uses a local inference stack built around vLLM and DGX Spark, while Meta's performance numbers remain vendor benchmarks that still need independent replication. 3