The top members of the new Meta Superintelligence Lab Reports New York Times.
Sources told the Times that Meta had completed the training at Behemoth, but delayed its release due to the sluggish internal performance. When the new Superintelligence workshop started, the test on the model reported.
The discussions are exactly that – discussions. Meta Mark Zuckerberg’s chief executive should still sign any changes, and a company spokesman told TechCrunch that Meta’s position at Open Source AI is “unchanged”.
“We are planning to continue to release top open source models,” the spokesman said. “We have not released everything we have developed historically and expect to continue training a mixture of open and closed models moving forward.”
The spokesman did not comment on the possible META shift away from Behemoth. If this happens, so that Meta can prioritize the closed source models, it will signal a major philosophical change for the company.
While Meta develops more advanced closed code models internally, such as those that feed Meta Ai’s assistant, Zuckerberg had made Open Source a central part of the company’s external strategy-a way to maintain AI growth faster. He placed loudly the opening of the Llama family as a differentiated competitors such as Openai, which Zuckerberg publicly criticized that it became more closed after working with Microsoft. But Meta is under pressure to generate revenue beyond ads as it throws billions into AI.
This includes the payment of huge signature bonuses and nine numerical salaries to attract leading researchers, building new data centers and covering the huge cost of developing artificial general intelligence (AGI) or “superconduct”.
Although one of the top AI research laboratories in the world, Meta is still behind his opponents, such as Openai, Anthropic, Google Deepmind and Xai when they are about to commercialize AI’s work.
If Meta prioritizes closed models, it could indicate that the opening was a strategic game, not an ideological one. Past comments by Zuckerberg imply a ambiguity to the opening of Sourcing Meta models. To one podcast Last summer, he said:
We are obviously very pro open source, but I have not committed to release every thing we do. I am basically very willing to believe that the open community will be good for the community and also good for us because we will benefit from innovations. If at some point, however, there is some quality change to what is capable of the thing and we feel that it is not responsible for Open Source, then we will not. It is very difficult to predict.
Closed models would give more control over post-control and more ways to generate revenue-especially revenue if he believes that the talent he has acquired can provide competitive and better performance.
Such a shift could also reshape the AI landscape. The open source momentum, which is largely driven by post-models such as Llama, could slow down, even when Openai leads to release the still late open model. The power could return to the most important players with closed ecosystems, while open source development can remain a product of popular basis efforts. Ripple results will continue throughout the starting ecosystem, especially for smaller companies that focus on alignment of guidance, security and alignment of models based on access to open foundation models.
On the world stage, Meta’s retreat from Open Source could potentially give the ground to China, which has embraced the AI open source – such as Deepseek and Moonshot ai – As a way to build domestic capacity and worldwide influence.
