Cryptopolitan
--
Meta may ditch open-source Behemoth for a private model
Meta’s new superintelligence lab is reportedly considering replacing its open‑source Behemoth model with a private one.
According to the NY Times sources, senior team members, including Meta’s freshly appointed chief AI officer, 28‑year‑old Alexandr Wang met last week to debate pausing the release of their most advanced publicly available model, Behemoth. The focus was instead on developing a proprietary version.
Meta used to make its AI code public so other developers could build on it. A closed‑source model keeps the code private, but Meta argues that sharing openly speeds up progress and gives more people access to powerful tools.
Switching to a private model would be a big change for Meta, which developers have praised for its open‑source approach.
As recently as last year, Yann LeCun, one of Meta’s top AI researchers, asserted that “the platform that will win will be the open one.”
Earlier this year, the Chinese venture DeepSeek leveraged Meta’s public code to create its own sophisticated AI chatbot, underscoring the tangible value of openness.
Although Meta finished training Behemoth on extensive datasets aimed at maximizing its capabilities, sources report that the rollout was postponed after initial in‑house evaluations fell short of projected benchmarks.
Following the inauguration of its superintelligence division last month, the Behemoth project considered a “frontier” effort at the leading edge of AI development, was put on hold pending further review.
These talks are still early with no decisions made, and any change would need CEO Mark Zuckerberg’s approval.
Meta could keep its public models while also building a private one. Either way, moving toward a private model would be a big strategic shift as it tries to keep upwith rivals like Google, OpenAI, and Anthropic.
In a podcast interview last year, Zuckerberg said, “We’re obviously very pro open source, but I haven’t committed to releasing every single thing that we do.”
Meta’s new lab is under scrutiny after several AI missteps
Meta has dealt with management problems, seen key staff depart, and launched products that didn’t catch on. Zuckerberg still wants to build an AI that achieves superintelligence, something some describe as beyond human thinking.
To staff the initiative, Zuckerberg initiated an aggressive recruitment campaign, offering nine‑figure compensation proposals to lure leading researchers from OpenAI, Google, Apple, Anthropic, and others. Moreover, he reassigned the executive formerly in charge of Meta’s generative AI division.
In June, Meta poured $14.3 billion into Scale AI, the company founded and helmed by Alexandr Wang. The deal granted Meta a 49 % equity share, and Wang, together with some of his key associates, transitioned into senior posts within Meta.
Following the investment, Meta rebranded its AI arm as “Meta Superintelligence Labs,” with Wang stepping into the role of chief AI officer. He now leads an elite cohort of about a dozen recent hires, several Scale AI deputies, and Nat Friedman, ex‑CEO of GitHub.
Sources indicate that numerous members of Wang’s group gathered at Meta’s Menlo Park headquarters for their inaugural in‑person meeting last week. They occupy a secluded office suite adjacent to Zuckerberg’s workspace.
On Tuesday, Wang held a Q&A with about 2,000 AI team members. He said his core team’s work would stay private and that everyone in the AI division would now focus on building superintelligence.
Looking forward, some AI employees anticipate that the upcoming vesting window in August, which will allow certain staff to liquidate company shares. It could trigger a wave of departures among those not selected for the superintelligence group.
On Monday, Zuckerberg unveiled intentions to commit hundreds of billions of dollars toward AI compute infrastructure, aiming to have Meta’s inaugural supercluster operational by next year.
In a Facebook update, Zuckerberg stated, “Meta Superintelligence Labs will have industry‑leading levels of compute and by far the greatest compute per researcher. I’m looking forward to working with the top researchers to advance the frontier!”
He called the first supercluster “Prometheus” and disclosed plans for additional multi‑gigawatt installations, such as “Hyperion,” projected to reach five gigawatts over the next few years.
Zuckerberg’s moves show he’s frustrated with how slowly Meta’s AI is progressing. After the underweheling response to Llama 4 in April, he is taking a different road to better compete with OpenAI and Google. “For our superintelligence effort, I’m focused on building the most elite and talent‑dense team in the industry,” he added, marking the start of a new phase in Meta’s AI endeavors.
Cryptopolitan Academy: Want to grow your money in 2025? Learn how to do it with DeFi in our upcoming webclass. Save Your Spot
According to the NY Times sources, senior team members, including Meta’s freshly appointed chief AI officer, 28‑year‑old Alexandr Wang met last week to debate pausing the release of their most advanced publicly available model, Behemoth. The focus was instead on developing a proprietary version.
Meta used to make its AI code public so other developers could build on it. A closed‑source model keeps the code private, but Meta argues that sharing openly speeds up progress and gives more people access to powerful tools.
Switching to a private model would be a big change for Meta, which developers have praised for its open‑source approach.
As recently as last year, Yann LeCun, one of Meta’s top AI researchers, asserted that “the platform that will win will be the open one.”
Earlier this year, the Chinese venture DeepSeek leveraged Meta’s public code to create its own sophisticated AI chatbot, underscoring the tangible value of openness.
Although Meta finished training Behemoth on extensive datasets aimed at maximizing its capabilities, sources report that the rollout was postponed after initial in‑house evaluations fell short of projected benchmarks.
Following the inauguration of its superintelligence division last month, the Behemoth project considered a “frontier” effort at the leading edge of AI development, was put on hold pending further review.
These talks are still early with no decisions made, and any change would need CEO Mark Zuckerberg’s approval.
Meta could keep its public models while also building a private one. Either way, moving toward a private model would be a big strategic shift as it tries to keep upwith rivals like Google, OpenAI, and Anthropic.
In a podcast interview last year, Zuckerberg said, “We’re obviously very pro open source, but I haven’t committed to releasing every single thing that we do.”
Meta’s new lab is under scrutiny after several AI missteps
Meta has dealt with management problems, seen key staff depart, and launched products that didn’t catch on. Zuckerberg still wants to build an AI that achieves superintelligence, something some describe as beyond human thinking.
To staff the initiative, Zuckerberg initiated an aggressive recruitment campaign, offering nine‑figure compensation proposals to lure leading researchers from OpenAI, Google, Apple, Anthropic, and others. Moreover, he reassigned the executive formerly in charge of Meta’s generative AI division.
In June, Meta poured $14.3 billion into Scale AI, the company founded and helmed by Alexandr Wang. The deal granted Meta a 49 % equity share, and Wang, together with some of his key associates, transitioned into senior posts within Meta.
Following the investment, Meta rebranded its AI arm as “Meta Superintelligence Labs,” with Wang stepping into the role of chief AI officer. He now leads an elite cohort of about a dozen recent hires, several Scale AI deputies, and Nat Friedman, ex‑CEO of GitHub.
Sources indicate that numerous members of Wang’s group gathered at Meta’s Menlo Park headquarters for their inaugural in‑person meeting last week. They occupy a secluded office suite adjacent to Zuckerberg’s workspace.
On Tuesday, Wang held a Q&A with about 2,000 AI team members. He said his core team’s work would stay private and that everyone in the AI division would now focus on building superintelligence.
Looking forward, some AI employees anticipate that the upcoming vesting window in August, which will allow certain staff to liquidate company shares. It could trigger a wave of departures among those not selected for the superintelligence group.
On Monday, Zuckerberg unveiled intentions to commit hundreds of billions of dollars toward AI compute infrastructure, aiming to have Meta’s inaugural supercluster operational by next year.
In a Facebook update, Zuckerberg stated, “Meta Superintelligence Labs will have industry‑leading levels of compute and by far the greatest compute per researcher. I’m looking forward to working with the top researchers to advance the frontier!”
He called the first supercluster “Prometheus” and disclosed plans for additional multi‑gigawatt installations, such as “Hyperion,” projected to reach five gigawatts over the next few years.
Zuckerberg’s moves show he’s frustrated with how slowly Meta’s AI is progressing. After the underweheling response to Llama 4 in April, he is taking a different road to better compete with OpenAI and Google. “For our superintelligence effort, I’m focused on building the most elite and talent‑dense team in the industry,” he added, marking the start of a new phase in Meta’s AI endeavors.
Cryptopolitan Academy: Want to grow your money in 2025? Learn how to do it with DeFi in our upcoming webclass. Save Your Spot
Disclaimer: Includes third-party opinions. No financial advice. May include sponsored content. See T&Cs.