Meta built its AI reputation on openness — that may be changing
Top members of the new Meta Superintelligence Lab discussed the powerful Open Source AI model of the company, Behemoth, and developed a closed model instead reports the New York Times.
Sources told The Times that Meta had completed the Behemoth training, but the release had delayed because of the overwhelming internal performance. When the new Superintelligence Lab was launched, the testing on the model reportedly stopped.
The discussions are exactly that – discussions. Meta CEO Mark Zuckerberg would still have to be signed in the event of changes, and a spokesperson for the company told WAN that the position of Meta on open source AI is ‘unchanged’.
“We are planning to continue to release leading open source models,” said the spokesperson. “We have not released everything we have developed historically and we expect to continue training a mix of open and closed models in the future.”
The spokesperson did not comment on the potential shift of Meta van Behemoth. If that happens so that Meta can give priority to models with closed source, this would mark an important philosophical change for the company.
While Meta uses more advanced closed-source models internally, just like those of its Meta AI assistant, Zuckerberg Open Source had made a central part of the company’s external AI strategy-a way to make AI development move faster. He positioned the openness of the Lama family loudly as a distinctive factor of competitors such as OpenAi, who publicly criticized Zuckerberg because he was closed more after he collaborated with Microsoft. But Meta is under pressure to make money about advertisements, because it pours billions in AI.
This includes paying massive signing bonuses and nine digits of salaries to poach top researchers, to expand new data centers and to cover the enormous costs for developing artificial general intelligence (AGI), or “super intelligence”.
Although he has one of the best AI research laboratories in the world, Meta is still lagging behind rivals such as OpenAI, Anthropic, Google DeepMind and Xai when it comes to commercializing his AI work.
If Meta gives closed models priority, this could suggest that openness was a strategic game, not an ideological one. Previous comments from Zuckerberg point to an ambivalence to connect to open the models of Sourcing Meta. On one podcast Last summer he said:
We are clearly a very pro -open source, but I have not promised to release all the thing we do. I am actually very inclined to think that open sourcing will be good for the community and also good for us because we will benefit from the innovations. However, if at some point there is a qualitative change in what the thing is capable of, and we feel that it is not responsible for opening it, then we will not do that. It is all very difficult to predict.
Closed models would give Meta more control and more ways to earn money-especially if it believes that the talent it has acquired competing, best-in-class performance can deliver.
Such a shift can also reform the AI landscape. Open Source Momentum, largely powered by meta and models such as Llama, could slow down, even if OpenAI is a matter of bringing out his open model that is still moving. Power could swing back to the most important players with closed ecosystems, while open source development can remain a product of grassroots efforts. The Ripple effects would continue with the startup ecosystem, especially for smaller companies aimed at refined, safety and model lines that depend on access to open foundation models.
On the world stage, the retreat of Meta from Open Source may be able to give up the ground to China, which has embraced open source AI – such as Deepseek and MOONSHOT AI – As a way to build up domestic skills and global influence.
