Why Meta might abandon open-source AI in favour of closed models

Meta’s open-source AI policy has long been seen as a counterweight to the secrecy of rivals like OpenAI and Google. (Courtesy: Reuters file photo)

Meta is weighing a dramatic shift in its AI strategy, as its newly formed Superintelligence Labs considers abandoning its most powerful open-source AI model in favour of developing a closed alternative. The change, if enacted, would represent a major philosophical and strategic pivot for the social media giant, which has long championed open-source development as a key competitive edge, the New York Times reported.

According to two people with direct knowledge of internal discussions, top executives in the lab—including Meta’s newly appointed chief AI officer, 28-year-old Alexandr Wang—met last week to discuss sidelining the company’s open-source model, Behemoth. The model, trained but not yet released due to underwhelming performance, had been positioned as Meta’s latest leap in the race to develop frontier AI.

Why the potential shift matters

Meta’s open-source AI policy has long been seen as a counterweight to the secrecy of rivals like OpenAI and Google. Developers have applauded the company’s approach, with some foreign competitors even building advanced models using Meta’s public code. A move to closed models would not only change Meta’s role in the broader AI community but could also signal the company’s growing urgency to regain lost momentum in the global AI arms race.

Behind the scenes at Superintelligence Labs

The shift is part of a broader internal transformation underway since Meta restructured its AI division as "Meta Superintelligence Labs." Wang, who previously founded Scale AI and joined Meta after a $14.3 billion investment deal, now leads an elite team working directly alongside CEO Mark Zuckerberg at company headquarters. The group operates in a siloed workspace, separate from Meta’s existing 2,000-person AI workforce.

In a recent company-wide Q&A, Wang reportedly told Meta’s AI staff that the division’s new mission was singular: to build superintelligence—AI systems with capabilities that exceed human cognition. However, he did not clarify whether that pursuit would continue to include open-source projects.

Open-source legacy under review

For years, Meta executives, including AI pioneer Yann LeCun, have argued that transparency and collaboration are key to responsible and accelerated AI development. But the performance concerns around Behemoth and rising competition from companies building closed frontier models have led to reconsideration.

Zuckerberg himself has left the door open to selective openness. “We’re obviously very pro open source, but I haven’t committed to releasing every single thing that we do,” he said in a podcast interview last year.

What’s next for Meta’s AI talent?

The impending shift has already created tension within the ranks. Some employees not selected for Wang’s exclusive team reportedly expect a wave of resignations after the company’s next stock vesting period in August. Meta has also sidelined earlier AI leaders, intensifying the internal reorganization.

Whether the company will fully close off its future AI models or maintain a hybrid strategy remains to be seen. But the discussions highlight how the pursuit of superintelligence is reshaping not only Meta’s technology—but also its identity.