Meta Unveils 30-Billion-Parameter Open-Source Model Muse Glimmer; Zuckerberg Urges US Policy Shift

On August 10, 2026, Meta Platforms released Muse Glimmer, a lightweight open-source agentic model. The model, a distilled version of Muse Spark 1.2, has 30 billion parameters and can run on a Mac or PC with a single graphics card. It is designed for agentic tasks such as scheduling and file organization. Model weights will be available for download on the Hugging Face platform, and Meta also plans to release weights for the more powerful Muse Spark version. The release follows the company's formation of a superintelligence team in 2025, as it seeks to strengthen its position in the AI race.

In a statement, Meta CEO Mark Zuckerberg called on the United States to revisit its open-source model policies. He said foreign labs currently hold several advantages in the open-weights space because US labs face additional restrictions on training data, and US policy must reduce these extra hurdles. He also stressed that restricting access to foreign open-source models is not an effective solution. Zuckerberg further announced that Meta will soon resume releasing some open-source models and will establish a new governance structure giving an independent board authority to approve safety standards for model releases.

Open-weight models are less costly than closed-source models, with core components made public, allowing easier customization. As companies grow more cautious about AI spending and recent cybersecurity incidents involving multiple major AI platforms draw attention, the open-weight model has gained more market traction. After AI platform Hugging Face suffered an intrusion by a rogue OpenAI model last month, it used a Chinese open-weight model to respond, because closed-source models had functional limitations in cybersecurity applications.

Meta, along with Amazon, Alphabet, and Microsoft, has committed cumulative investments exceeding $2 trillion to expand AI capabilities, covering hardware purchases and data center construction. Meta is planning a cloud infrastructure business to sell AI computing power and model access to external customers. Additionally, on Monday Meta announced a $1 billion fund to invest in US communities where it owns and operates data centers, aiming to ease local resistance to data center expansion.