SILICON VALLEY — Meta, which had once shifted to a closed-model strategy, has resumed offering an open-weight artificial intelligence model that anyone can use for free. The move is aimed at countering the closed AI models led by Anthropic and OpenAI, while also blocking the expansion of China's open AI ecosystem. Meta CEO Mark Zuckerberg called on the Trump administration to ease AI controls, arguing that the United States must shed its "ankle weights" to win the U.S.-China race for AI supremacy.
Meta, the operator of Facebook, launched the open-weight model "Muse Glimmer" on the 10th local time. Muse Glimmer is a small model with 30 billion parameters, designed to run an AI agent on a PC or tablet without an internet connection. Meta also said it would soon release the weights of "Muse Spark 1.2," its top-tier closed-model (paid) offering.
Open-weight models make public their "weights," the core values an AI model obtains through training on its parameters. While they do not release the entire code as open-source models do, they are classified as a middle-ground approach closer to openness, because the weights allow developers to build a range of AI applications. For companies or developers without the capacity for capital spending, this means they can use AI without worrying about cost.
Meta's launch of an open-weight model signals a return to an open ecosystem. After entering the AI market late following the arrival of OpenAI's ChatGPT in 2022, Meta focused on developing its own AI model, "Llama," pledging to build an open AI ecosystem distinct from those of OpenAI and Anthropic. But when "Llama 4," released last April, underperformed expectations and the gap with rivals widened, the company switched to a closed-model approach and first unveiled "Muse Spark" this April.
In a letter posted on the company's website titled "The Future Belongs to Everyone," Zuckerberg stressed that an open ecosystem is needed for the artificial superintelligence (ASI) that Meta is pursuing to become a reality. "Instead of concentrating ASI in a single institution, it should be widely distributed so that everyone can use it for themselves," he argued.
Meta turned back to open AI in order to break the closed AI market structure centered on Anthropic and OpenAI. Zuckerberg took direct aim at the two companies, saying, "There is something fundamentally wrong with the idea that AI is so dangerous that the only way to make it safe is to concentrate power to an extreme degree." His remarks were seen as a jab at Anthropic and OpenAI, which have recently disclosed unauthorized AI hacking and called for government control. Some in the industry even interpreted the moves as "noise marketing," in which firms such as Anthropic show off the performance of their AI models ahead of an initial public offering (IPO).
Meanwhile, Meta has positioned itself as a counterweight to Chinese companies making inroads worldwide with open AI. Zuckerberg stressed that the United States must halt its regulatory push if it wants to win the AI war with China. His argument is that blocking the use of Chinese-made AI would shrink the U.S. open ecosystem and could instead work to China's advantage. "Restricting the best open-source models would lower the quality of AI and lead to the centralization of AI," he warned. "Kimi K3," an open-weight model unveiled last month by China's Moonshot AI, has put the United States on edge by showcasing performance on par with the most advanced models from Anthropic and OpenAI.
Palantir, an AI software company, has also criticized the closed models of Anthropic and OpenAI, saying they are encouraging "token maxing," a race to consume tokens. Palantir CEO Alex Karp pointed out that these companies are fueling token maxing and stripping companies and institutions of their AI sovereignty, and late last month Palantir joined a tech companies' open letter opposing regulation of open AI.
