Meta is reopening its AI playbook and taking direct aim at rivals that keep their most powerful systems locked away.
Meta on Monday released the weights for a new AI model called Muse Glimmer, allowing developers to download, modify, and run it on their own hardware.
The company also said it plans to release the weights for a version of its more advanced Muse Spark model in the coming weeks, marking a return to the open-weight strategy that once set Meta apart from competitors such as OpenAI, Anthropic, and Google.
The announcement came alongside a lengthy essay from CEO Mark Zuckerberg titled “The Future is for Everyone,” in which he argued that advanced AI should be distributed widely rather than controlled by a small group of companies or governments.
A direct shot at OpenAI and Anthropic
Zuckerberg positioned Meta as the company focused on building “personal superintelligence” for individuals, contrasting that approach with rivals that sell controlled access to proprietary AI models.
“Most other labs are focused on building AI for companies, governments, or other institutions, so if those labs lead, then the balance of power will favor larger institutions over individuals,” Zuckerberg wrote.
He also criticized what he described as excessive fear surrounding AI development, writing that “The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic.”
Open models and a $1 billion community fund
Meta said Muse Glimmer is designed to run on a Mac or PC with a single graphics card, making it cheaper and easier to deploy than many frontier AI systems. The company had previously declined to release Muse Spark’s weights earlier this year, citing safety concerns, but has now pledged to open a version of the model.
Zuckerberg also announced a $1 billion fund for communities that host Meta’s US data centers, an effort aimed at easing local opposition as the company accelerates its AI infrastructure build-out.
The bigger fight over AI power
The move comes as Chinese open-weight models from companies such as DeepSeek, Moonshot, and Alibaba have rapidly improved, increasing pressure on US AI companies. Zuckerberg argued that American policy should reduce barriers around data use and distillation so US open-weight models can compete globally.
He also defended distillation, saying it is “important to protect the principle that you can learn from anything you can observe.”
Meta said it will give its independent board of directors authority over safety criteria for future model releases.
What this could change
Zuckerberg’s essay is more than a product announcement; it is a challenge to the business model dominating frontier AI.
OpenAI, Anthropic, and Google increasingly generate revenue by controlling access to their most capable models, while Meta is betting that widespread distribution of AI technology will attract developers, shape standards, and strengthen its ecosystem across Facebook, Instagram, WhatsApp, and future AI products.
The tradeoff is that open-weight models can be customized by anyone, including bad actors, which is why some competitors have chosen tighter controls. Meta is betting that broader access will ultimately make AI systems more useful, more secure, and harder for any single institution to dominate.
Also read: Anthropic says it does not support banning open-weight AI models, but argues that more capable releases should face stronger safety testing.