Meta Considers Abandoning Open-Source AI Model in Major Strategy Shift

Meta built its AI reputation on openness -- that may be changing | TechCrunch

Top members of Meta's new Superintelligence Lab discussed pivoting away from the company's powerful open-source AI model, Behemoth, and instead developing a closed model, reports The New York Times. Sources told The Times that Meta had completed training on Behemoth, but delayed its release due to underwhelming internal performance. When the new Superintelligence Lab launched, testing on the model reportedly halted. The discussions are just that - discussions. Meta CEO Mark Zuckerberg would still need to sign off on any changes, and a company spokesperson told TechCrunch that Meta's position on open source AI is "unchanged." "We plan to continue releasing leading open source models," the spokesperson said. "We haven't released everything we've developed historically and we expect to continue training a mix of open and closed models going forward." The spokesperson did not comment on Meta's potential shift away from Behemoth. If that happens so that Meta can prioritize closed-source models, it would mark a major philosophical change for the company. While Meta deploys more advanced closed-source models internally, like those powering its Meta AI assistant, Zuckerberg had made open source a central part of the company's external AI strategy -- a way to keep AI development moving faster. He loudly positioned the Llama family's openness as a differentiator from competitors like OpenAI, which Zuckerberg publicly criticized for becoming more closed after partnering with Microsoft. But Meta is under pressure to monetize beyond ads as it pours billions into AI. That includes paying massive signing bonuses and nine-figure salaries to poach top researchers, building out new data centers, and covering the enormous costs of developing artificial general intelligence (AGI), or "superintelligence." Despite having one of the top AI research labs in the world, Meta still lags behind rivals like OpenAI, Anthropic, Google DeepMind, and xAI when it comes to commercializing its AI work. If Meta prioritizes closed models, it could suggest that openness was a strategic play, not an ideological one. Past comments from Zuckerberg hint at an ambivalence toward committing to open sourcing Meta's models. On a podcast last summer, he said: "We're obviously very pro open source, but I haven't committed to releasing every single thing that we do. I'm basically very inclined to think that open sourcing is going to be good for the community and also good for us because we'll benefit from the innovations. If at some point, however, there's some qualitative change in what the thing is capable of, and we feel like it's not responsible to open source it, then we won't. It's all very difficult to predict." Closed models would give Meta more control and more ways to monetize - especially if it believes the talent it has acquired can deliver competitive, best-in-class performance. Such a shift could also reshape the AI landscape. Open-source momentum, largely driven by Meta and models like Llama, could slow, even as OpenAI gears up to release its still-delayed open model. Power could swing back toward the major players with closed ecosystems, while open-source development might remain a product of grassroots efforts. The ripple effects would continue across the startup ecosystem, especially for smaller companies focused on fine-turning, safety, and model alignment that rely on access to open foundation models. On the world stage, Meta's retreat from open source could potentially cede ground to China, which has embraced open-source AI - like DeepSeek and Moonshot AI - as a way to build domestic capability and global influence.

Meta's New Superintelligence Lab Is Discussing Major A.I. Strategy Changes

Meta's newly formed superintelligence lab has discussed making a series of changes to the company's artificial intelligence strategy, in what would amount to a major shake-up at the social media giant. Last week, a small group of top members of the lab, including Alexandr Wang, 28, Meta's new chief A.I. officer, discussed abandoning the company's most powerful open source A.I. model, called Behemoth, in favor of developing a closed model, two people with knowledge of the matter said. For years, Meta has chosen to open source its A.I. models, which means it makes the computer code public for other developers to build on. Closed models keep their underlying code private. Meta executives have long argued it is better for the technology to be built in public so that A.I. development will move faster and be accessible to more developers. Any move toward a closed A.I. model would be a philosophical change at Meta as much as a technical one. Meta has won plaudits from developers for open sourcing its A.I. models and one of its top A.I. executives, Yann LeCun, had said "the platform that will win will be the open one." This year, the Chinese A.I. company DeepSeek released an advanced A.I. chatbot thanks in part to Meta's open source code. Meta had finished feeding in data to improve its Behemoth model, a process known as "training," but has delayed its release because of poor internal performance, said the people with knowledge of the matter, who were not authorized to discuss private conversations. After the company announced the formation of the superintelligence lab last month, teams working on the Behemoth model -- which is known as a "frontier" model -- stopped running new tests on it, one of the people said. The superintelligence lab's discussions are preliminary and no decisions have been made on potential changes, which would need sign-off from Mark Zuckerberg, Meta's chief executive. Meta could keep its open source A.I. models while prioritizing a closed model. If these scenarios happen, they would be a significant shift for the company as it tries to stay competitive in the A.I. race against rivals like Google, OpenAI and Anthropic. A Meta spokesman declined to comment on the superintelligence lab's discussions. In a podcast interview last year, Mr. Zuckerberg said, "We're obviously very pro open source, but I haven't committed to releasing every single thing that we do." Meta's superintelligence lab is being closely watched after the company recently stumbled with A.I. technology, including internal management struggles, employee churn and product releases that fell flat. Mr. Zuckerberg's ultimate goal is to create A.I. that is "superintelligent," which means it would hypothetically exceed the powers of the human brain. Mr. Zuckerberg has embarked on a spending spree to create the new lab, offering as much as nine-figure pay packages to hire top researchers from companies like OpenAI, Google, Apple and Anthropic. Meta also sidelined its executive who had been leading generative A.I. In June, the company made a $14.3 billion investment in the A.I. start-up Scale AI, which was founded and led by Mr. Wang. Under the deal, Meta took a 49 percent stake in the company, and Mr. Wang and a team of top Scale employees joined Meta in leadership roles. The company has since renamed its entire A.I. division "Meta Superintelligence Labs," with Mr. Wang as chief A.I. officer. Within the larger A.I. division, Mr. Wang has led an exclusive team of around a dozen newly hired researchers, a handful of his deputies from Scale AI, and Nat Friedman, the former chief executive of GitHub, a software start-up. Many members of Mr. Wang's team reported to Meta's headquarters in Menlo Park, Calif., last week for the first time, the two people with knowledge of the matter said. The group is working in an office space siloed from the rest of the company and next to Mr. Zuckerberg, the people said. On Tuesday, Mr. Wang held a question-and-answer session with Meta's A.I. workers, who number about 2,000. In the meeting, he said the work of his small team would be private, but Meta's entire A.I. division would now be working toward creating superintelligence, the people with knowledge of the matter said. He did not address whether A.I. models would be open or closed. In August, at the end of the company's next vesting period, which is when some workers are able to sell portions of their stock, some employees expect an exodus of A.I. talent who were not chosen to join Mr. Wang's superintelligence team, one of the people with knowledge of the matter said.

Meta Plans to Abandon Llama 4 Behemoth. But Why? | AIM

The company is reportedly focusing on building a closed source model instead. Meta has reportedly considered abandoning the Llama 4 Behemoth, according to a July 14 report by the New York Times. The report indicates that a small group of senior staff at Meta's newly announced superintelligence lab are now believed to be developing a closed-source model instead. The Llama 4 Behemoth is currently the company's biggest and most powerful AI model that has been announced. According to the NYT report, Meta has completed training the Behemoth model but delayed its release due to 'poor internal performance'. And after Meta announced the superintelligence lab last month, teams working on the model stopped running tests on it. Earlier, the Wall Street Journal reported, citing people familiar with the matter, that the tech giant delayed the rollout of the model, 'prompting concerns about the direction of its multi-billion dollar AI investments." These individuals also told the publication that Meta's engineers and researchers were concerned that its performance would not match the public statements made about its capabilities. The model was expected to be released later this year. However, recent updates suggest that a release is now unlikely anytime soon. If Meta develops a closed-source model, it would represent a significant departure from the company's long-standing approach of creating open-source AI models. In April, the company introduced the Llama 4 family of AI models, which includes three variants: Behemoth, Maverick, and Scout. The Behemoth is the largest, with a total of 2 trillion parameters. Meta dubbed the model as one of the most innovative AI models in the world. More recently, Meta intensified its focus on building a team for its superintelligence initiatives. Besides assembling a team led by former Scale AI CEO Alexandr Wang and recruiting several others from OpenAI, CEO Mark Zuckerberg stated that the company would spend hundreds of billions of dollars on AI data centres for these efforts. Despite these efforts to stay ahead of competitors, why is the company planning to abandon the Behemoth model? SemiAnalysis, one of the world's leading AI and semiconductor research companies, outlined some of the reasons in its latest report. Semianalysis suggests that Meta's decision to use the chunked attention technique for memory efficiency may have been a mistake. Standard attention allows every token to access all previous tokens, forming a complete context. Chunked attention splits tokens into fixed blocks, limiting each token's attention to only its current block. "Behemoth's implementation of chunked attention chasing efficiency created blind spots, especially at block boundaries," read the report from SemiAnalysis. Each divided chunk can access tokens within its own block, but not those in the preceding block. Therefore, if a logical argument or chain of thought extends from one chunk to another, the model loses the connection. This weakened the model's ability to follow and reason across long chains of thought. "We believe part of the problem was that Meta didn't even have the proper long context evaluations or testing infrastructure set up to determine that chunked attention would not work for developing a reasoning model," added the report. SemiAnalysis also said that Meta is 'very far behind on RL and internal evaluations' and the new superintelligence team is set to close the gap. Besides, the report added that Meta's Behemoth model switched its Mixture of Experts routing method midway through training, disrupting how its expert networks specialised. This led to instabilities, ultimately limiting the model's overall effectiveness. Among other reasons, SemiAnalysis also states that Llama 4 Behemoth faced bottlenecks with regard to training data. "Prior to Llama 4 Behemoth, Meta had been using public data (like Common Crawl), but switched mid-run to an internal web crawler they built. While this is generally superior, it also backfired," read the report, stating that Meta struggled to clean and deduplicate the new data stream. "The processes hadn't been stress-tested at scale." The report further notes that Meta also struggled to scale research experiments into full-fledged training runs. The company had to deal with competing research directions and a lack of leadership to decide the most productive path for the model. "Certain model architecture choices did not have proper ablations but were thrown into the model. This led to poorly managed scaling ladders," said the report.

Meta's new Superintelligence Lab is discussing major AI strategy changes - The Economic Times

Meta's new superintelligence lab is considering abandoning its open-source AI model, Behemoth, in favour of developing a closed model, marking a major strategic shift. While no final decision has been made, the move could signal a departure from Meta's long-standing open-source philosophy to better compete in the AI race.Meta's newly formed superintelligence lab has discussed making a series of changes to the company's artificial intelligence strategy, in what would amount to a major shake-up at the social media giant. Last week, a small group of top members of the lab, including Alexandr Wang, 28, Meta's new chief AI officer, discussed abandoning the company's most powerful open source AI model, called Behemoth, in favour of developing a closed model, two people with knowledge of the matter said. For years, Meta has chosen to open-source its AI models, which means it makes the computer code public for other developers to build on. Closed models keep their underlying code private. Meta executives have long argued it is better for the technology to be built in public so that AI development will move faster and be accessible to more developers. Any move toward a closed AI model would be a philosophical change at Meta as much as a technical one. Meta has won plaudits from developers for open-sourcing its AI models, and one of its top AI executives, Yann LeCun, had said, "The platform that will win will be the open one." This year, Chinese AI company DeepSeek released an advanced AI chatbot thanks in part to Meta's open source code. Meta had finished feeding in data to improve its Behemoth model, a process known as "training," but has delayed its release because of poor internal performance, said the people with knowledge of the matter, who were not authorized to discuss private conversations. After the company announced the formation of the superintelligence lab last month, teams working on the Behemoth model -- which is known as a "frontier" model -- stopped running new tests on it, one of the people said. The superintelligence lab's discussions are preliminary, and no decisions have been made on potential changes, which would need sign-off from Mark Zuckerberg, Meta's CEO. Meta could keep its open-source AI models while prioritizing a closed model. If these scenarios happen, they will be a significant shift for the company as it tries to stay competitive in the AI race against rivals like Google, OpenAI and Anthropic.

Meta's Superintelligence Lab Contemplates AI Strategy Overhaul, Including Abandoning Open Source Model: Report - Alphabet (NASDAQ:GOOG), Apple (NASDAQ:AAPL)

Meta Platforms Inc. META is reportedly considering a major shift in its artificial intelligence (AI) strategy, including the potential abandonment of its open-source AI model, Behemoth. What Happened: The newly formed Superintelligence Lab at Meta, led by 28-year-old Alexandr Wang, the company's new Chief AI Officer, is contemplating a shift from the open-source Behemoth model to a closed model, reported The New York Times. The lab's discussions are in the early stages, and decisions are yet to be made. However, any changes would require approval from Meta's CEO, Mark Zuckerberg, the report said. Check out the current price of META stock here. Meta's open-source AI models have been a cornerstone of its AI strategy, with the company arguing that open-sourcing the technology would accelerate AI development and make it more accessible to developers. Moving to a closed A.I. model would represent a major philosophical shift for the company. The potential shift comes after Meta faced challenges with the Behemoth model's internal performance, leading to a delay in its release. Despite these challenges, Meta may continue supporting its open-source AI models while ultimately prioritizing a closed system. Wang recently hosted a Q&A with Meta's AI staff, where he stated that although his team's work would remain private, the entire AI division would now be dedicated to building superintelligence. However, he did not clarify whether the A.I. models would be open or closed. Meta did not immediately respond to Benzinga's request for comment. SEE ALSO: Pudgy Penguins Coin PENGU Leaps Past Dogecoin, Shiba Inu, Pepe With 93% Weekly Gain - Benzinga Trending Investment OpportunitiesAdvertisementArrivedBuy shares of homes and vacation rentals for as little as $100. Get StartedWiserAdvisorGet matched with a trusted, local financial advisor for free.Get StartedPoint.comTap into your home's equity to consolidate debt or fund a renovation.Get StartedRobinhoodMove your 401k to Robinhood and get a 3% match on deposits.Get Started Why It Matters: This news comes on the heels of Meta's aggressive push towards AI development. In a recent announcement, Zuckerberg confirmed that the company would invest hundreds of billions to build massive AI clusters in pursuit of superintelligence. Earlier, Meta had embarked on a massive recruitment drive, offering multimillion-dollar packages to assemble an elite AI team for its superintelligence labs. The company even poached talent from other tech giants like OpenAI and Apple Inc. AAPL with remuneration packages that outstripped those of Fortune 500 CEOs and professional sports stars. Furthermore, Meta expanded its AI talent pool by acquiring PlayAI, an AI startup focused on natural-sounding voice technology. The entire PlayAI team joined Meta, further bolstering its AI capabilities. With these moves, Meta strives to remain competitive in the AI race against rivals such as Google GOOG GOOGL, OpenAI, and Anthropic. The company's ultimate goal is to create a "superintelligent" AI, which would hypothetically surpass human cognitive abilities. READ MORE: Mark Zuckerberg Expected To Appear As Star Witness In $8 Billion Trial Over Alleged Privacy Violations Image via Shutterstock Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors. AAPLApple Inc$208.01-0.29%Stock Score Locked: Want to See it? Benzinga Rankings give you vital metrics on any stock - anytime. Reveal Full ScoreEdge RankingsMomentum25.11Growth32.39Quality75.55Value9.40Price TrendShortMediumLongOverviewGOOGAlphabet Inc$182.45-0.20%GOOGLAlphabet Inc$181.23-0.18%METAMeta Platforms Inc$720.85-0.01%Market News and Data brought to you by Benzinga APIs