Meta open-sources most powerful AI model, challenges OpenAI and Anthropic

On Monday, Meta CEO Mark Zuckerberg announced a major shift in the company's AI strategy through an Instagram video: Meta will open-source its latest AI model Muse Spark 1.2, releasing the weights (the calculations and rules that govern how the AI behaves) so the public can download and use them. The company will also launch Muse Glimmer, a new family of open-source models specifically designed to run on laptops and consumer devices.

The announcement was accompanied by Meta shares rising 2.1% in premarket trading, though the stock remains down around 10% for the year as investors scrutinize Meta's capital spending and competitive position in AI. Zuckerberg framed the move as part of a broader effort by Meta Superintelligence Labs—formed last year—to demonstrate progress and justify the company's forecasted capital expenditure of up to $145 billion for the year.

In a 6,500-word essay published Monday, Zuckerberg positioned Meta's open-source strategy as both a technical and political challenge. He highlighted that Chinese competitors—Alibaba, DeepSeek, and Moonshot—have aggressively released open-weight models that have competed effectively with leading U.S. technology. Zuckerberg directly called on Washington to support American open-source efforts, writing: "Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data. US policy must reduce this additional friction if we want American open source models to lead over time." He argued against restricting foreign open-source models, instead urging that "American open source models should be the best globally. This requires removing the hurdles that make it harder for American open source models to compete."

Muse Glimmer's design offers a technical differentiation. Most AI inference (the step where an AI produces an answer) is processed in cloud data centers, an expensive approach. Muse Glimmer, by running on-device, bypasses those cloud compute costs and can improve speed. Neil Shah, co-founder at Counterpoint Research, told CNBC: "Bringing small, agentic models like Muse Glimmer directly onto PC and mobile hardware bypasses cloud compute costs to outcompete Google, Microsoft and others on the end-user's device."

Zuckerberg also used the essay to make a philosophical and political case against what he saw as excessive concentration of AI power. "Still, it is surprising that the discourse from many developing AI is so filled with doom," he wrote, in what appeared to be a thinly veiled critique of OpenAI and Anthropic, whose leaders have warned of AI's potential impact on employment. "The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic." Instead, Zuckerberg advocated for distributed superintelligence: "Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it." He envisioned a future where "everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care about."

Zuckerberg also called for U.S. policy changes in specific areas, including AI distillation—the use of output from advanced AI models to train other models—and data use in training. Distillation has drawn criticism from some U.S. lawmakers who view it as a form of intellectual property theft, but Zuckerberg signaled Meta believes these restrictions hinder American competitiveness and should be reconsidered.