Meta has introduced a compact AI model called Muse Glimmer, designed to run on a single computer with just one GPU. The company announced the release today, describing it as a slimmed-down version of its larger Spark 1.2 model, built for tasks like scheduling and file management. "We designed Muse Glimmer to balance capability against the memory and compute constraints of local hardware," Meta stated.

The model's weights are now available for free on Hugging Face, along with developer documentation, so users can download and run it on their own PCs. Meta noted that optimized integrations for llama.cpp and other platforms are on the way, allowing users to move from download to a working agent in minutes.

According to Meta, the model performs well for its size, showing strong success rates on benchmarks including DeepSearch QA, MCP-Atlas, and SWE-Bench, which tests code writing and debugging. It also supports reliable tool use, multi-step reasoning, failure recovery, multimodal input, and works with agent orchestrators like OpenClaw. The model was trained on data from over 100 languages.

Meta's move echoes competitors like China's DeepSeek, which also runs on local machines with more open licensing. In an essay accompanying the release, CEO Mark Zuckerberg said, "Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it."

Meta's Spark AI line has been seen as weaker than rivals from OpenAI and Anthropic, and this new model appears to push toward a more open-source approach, appealing to users who prefer local execution. This comes amid broader industry discussions, including NVIDIA's recent launch of the Open Secure AI Alliance for cyber defense, following a rogue attack on Hugging Face by an unreleased, closed OpenAI model.

Source: Engadget