Meta Unveils Muse Glimmer Open-Weight Model as Zuckerberg Criticizes Closed AI

Meta has unveiled Muse Glimmer, a compact, open-weight agentic model capable of running locally on Mac or PC. The social media giant said Glimmer features 30 billion parameters and is a distilled version of Muse Spark 1.2. Despite its compactness, Muse Glimmer efficiently runs on a single graphics processing unit.

How to Use Meta’s Muse Glimmer

Meta’s new AI model is designed to do more than respond to questions. Meta described it as an agentic model, which means it is built to handle tasks that require multiple steps and actions. It can use tools, maintain a plan, write and debug code, and adapt its approach as a task progresses.

With its multimodal input and reasoning capabilities, Muse Glimmer can interpret charts, screenshots, and documents during conversations. According to the social media giant, the AI model registered higher performance compared to Gemma4-31 and Qwen3.6-27B on various benchmarks.

Muse Glimmer AI has been trained and evaluated for end-to-end agentic task completion. The AI model is the latest product to be developed by Meta’s Superintelligence Labs, which is headed by former Scale AI CEO Alexandr Wang.

Focus on Open-Weight Models

As Meta unveiled Muse Glimmer AI, CEO Mark Zuckerberg outlined his vision for AI in an article titled “The Future Is for Everyone”. According to the Meta CEO, superintelligent AI should not be controlled solely by large technology companies or governments.

"Some argue that superintelligence itself or a small set of experts who control it should decide what is best for humanity. We disagree," Zuckerberg noted.

This position reflects Meta’s open approach to AI development and contrasts with the closed strategies adopted by companies such as OpenAI, Anthropic and Google. Meta unveiled Muse Glimmer at a time when the debate on open-weight AI continues in the tech industry.

Last month, leading AI companies signed a public letter to endorse open-weight AI models as the U.S. government considered sanctioning Chinese companies that distilled models from American rivals to develop open-source models

"Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data. U.S. policy must reduce this additional friction if we want American open source models to lead over time,” Zuckerberg added.

Mirroring Chinese AI Lab Strategy

Meta’s decision to open-source Muse Glimmer mirrors the strategy adopted by Chinese AI labs. Startups such as DeepSeek and Moonshot AI have launched open-weight models that provide developers with cheaper alternatives to American companies. For a long time, Meta has supported open-source models.

However, not every model in its open-weight Llama series received a positive response from developers. Earlier this year, the social media giant took a different approach when it launched Muse Spark as a closed model.