Meta launches new AI model Muse Glimmer as Zuckerberg urges U.S. to remove barriers to open-source AI

Meta is putting open-source AI back at the center of its strategy.

The company on Monday unveiled Muse Glimmer, a new open-weight AI model built to run agentic tasks on a Mac or PC using a single graphics card. Mark Zuckerberg paired the release with a warning for Washington: if the U.S. makes open AI harder to build at home, Chinese developers could take the lead.

“We’ve got even bigger models that are coming soon,” Zuckerberg said in a video accompanying a 14-page essay titled “The Future is for Everyone.”

Muse Glimmer is smaller than the flagship models coming from OpenAI, Anthropic and Google. That is part of its appeal. Meta is targeting developers and users who want AI running directly on their own hardware instead of sending every task through a cloud service.

Muse Glimmer benchmarks show how Meta’s 30B model stacks up against rivals

“Muse Glimmer is an open model built for always-on local agents. 30B parameters, licensed under Apache 2.0, and runs on a single GPU — tuned for tool use, long tasks, and failure recovery. Muse Glimmer is trained to deliver competitive agentic and coding performance, with multimodal perception built in,” Meta said on the Muse Glimmer release page.

The release arrives as open-weight models gain new attention across the AI industry. High inference costs have pushed companies to look for cheaper options, and security incidents involving AI agents have raised questions about how much control users have over closed systems.

Open-weight models give developers access to core model parameters, making it possible to modify, fine-tune, and run them on infrastructure they control. Closed models from OpenAI, Anthropic, and Google keep those weights private.

China has become a formidable force in this part of the market. Moonshot’s Kimi K3, Alibaba’s Qwen3.8-Max and DeepSeek’s V4-Flash are competing with some of the strongest U.S.-built systems. Meta’s return to open-weight releases puts one of America’s largest technology companies directly back into that contest.

Zuckerberg says U.S. AI policy could give China an edge

Zuckerberg’s argument goes far beyond Muse Glimmer.

“The notion AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic,” he wrote.

His position comes at an awkward moment for the U.S. AI industry. American companies dominate many of the most capable closed models, yet Chinese labs have built growing momentum around models developers can download, modify and deploy themselves.

Zuckerberg wants U.S. policymakers to rethink rules affecting training data and model distillation, a technique that uses outputs from a larger AI system to train a smaller, less computationally demanding model.

“Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data,” Zuckerberg said.

“U.S. policy must reduce this additional friction if we want American open source models to lead over time,” he added.

Meta plans to put more weight behind that argument. The company said it intends to release the weights for Muse Spark 1.2, its most advanced model, built by the superintelligence team Meta assembled last year after its Llama 4 models received a disappointing reception.

Safety remains part of the equation. Zuckerberg said Meta will create a governance structure that gives independent directors authority to approve safety criteria used in decisions about releasing models.

The debate reaches beyond software. Meta expects to spend as much as $145 billion this year on AI infrastructure, placing the company among the biggest participants in America’s data-center construction boom. Local opposition to data centers has grown amid concerns over electricity demand, water use and costs for surrounding communities, Reuters reported.

Meta is responding with a new $1 billion fund intended to support communities affected by its data-center buildout.

“One significant disadvantage that the U.S. has compared to countries like China is that it is more difficult to build infrastructure here,” Zuckerberg said.

That connects Meta’s two arguments. Zuckerberg sees America’s AI competition with China as a fight over models and the physical infrastructure needed to train and run them.

Muse Glimmer may be a relatively small model, but the message attached to it is much larger. Meta is betting that AI’s next phase will not belong exclusively to a handful of companies operating massive closed systems. Zuckerberg wants downloadable models, local computing, and broader access to remain part of the equation, and he is warning Washington that China may fill that space if American developers cannot.