Zuckerberg Signal Meta does not open source all of its “Superintelligence” AI models

Meta CEO Mark Zuckerberg shared him vision The idea is that people should be able to use AI to achieve their personal goals on Wednesday for “personal superintelligence.”

Smuggled into the letter is a signal that Meta is shifting its plans to release AI models as it pursues “super intelligence.”

“We believe that the benefits of superintelligence should be shared as widely as possible with the world,” writes Zuckerberg. “That being said, super intelligence will raise new safety concerns. We need to be strict about mitigating these risks and paying attention to what we choose to open source.”

The language about open source is important. Zuckerberg has historically placed open-model Lama family of open-model Lama families as key differentiators from competitors such as Openai, Xai and Google Deepmind. The goal of Meta was to create an open AI model that was either as good as a closed model. in 2024 LetterZuckerberg writes: “From next year, we expect future Llama models to be the most advanced in the industry.”

Zuckerberg previously left room for manipulation of this commitment. “But at some point, if there’s some qualitative change in what can be done and you feel like you’re not responsible for it to open source,” he said. Podcasts last year.

And while many say that llamas don’t fit the strict definition of open source AI, sometimes Meta hasn’t released a large training dataset, Zuckerberg’s words indicate the possibility of changing priorities. Open source may no longer be the default for cutting edge AI in meta

There’s a reason why Meta’s rivals keep the model closed. The closed model gives companies more control over product monetization. Zuckerberg pointed out last year that “Llama’s release does not undermine our ability to invest in revenue, sustainability, or research like a closure provider.” Of course, Meta makes most of its money from selling internet ads.

Still, that’s the way the open model viewpoint was. in front Meta is beginning to feel like he’s falling behind his competitors, and executives are Obsessed with defeating Openai’s GPT-4 model Lama 3 is under development.

Cut in June 2025 when Meta launched the AGI sprint, which was released in earnest through investment. $14.3 billion in scale AI, Get the founder of the scale and CEO, and Restructuring AI initiatives Under a new unit called Meta Superintelligence Labs. Meta spent Billions of dollars To acquire researchers and engineers from top AI companies and build new data centers.

Recent reports have been shown All of that investment has led Meta to pause testing with its latest Llama model, Behemoth, and instead focuses its efforts on developing a closed model.

Zuckerberg’s mission to bring “personal closeness” into the world – the determined shift from his rivals who are working on “automating all valuable work” has come to shape his AI monetization strategy. From Zuckerberg’s words, it is clear that Meta plans to provide “personal tensions” through its own products, such as augmented reality glasses and virtual reality headsets.

“Personal devices like glasses that understand our context can see what we see, hear what we hear and interact with us all day long will become our main computing device,” Zuckerberg wrote in a letter Wednesday.

If we leave the potentially cutting edge models on Meta closed, a Meta spokesperson said the company remains committed to open source AI and is also hoping to train closed source models in the future.

“Our position on open source AI remains the same,” the spokesman said. “We plan to continue to release major open source models. We have not released everything that we have developed historically, so we look forward to continuing to train the mix of open and closed models in the future.”

This article has been updated with more information on Mark Zuckerberg’s stance on the Open AI model.