Meta is reportedly changing its open source AI development path to develop a closed model, and is development of 'Behemoth' being discontinued?

Top members of Meta's newly formed Superintelligence Lab are considering shifting the company's focus on open source AI development and instead developing closed models, The New York Times reported.

Meta's New Superintelligence Lab Is Discussing Major AI Strategy Changes - The New York Times
https://www.nytimes.com/2025/07/14/technology/meta-superintelligence-lab-ai.html

Meta built its AI reputation on openness -- that may be changing | TechCrunch
https://techcrunch.com/2025/07/14/meta-built-its-ai-reputation-on-openness-that-may-be-changing/

According to information obtained exclusively by The New York Times from sources, Meta has completed training of its open-source AI model , Llama 4 Behemoth , but delayed its release after results from internal performance testing were disappointing.

Meta plans to release the ultra-large-scale AI model 'Llama 4 Behemoth' with a total of 2 trillion parameters, but development has been delayed until fall 2025 or later - GIGAZINE

Later, Meta reported that as part of a $14.8 billion investment in data labeling service Scale AI, the company's CEO Alexander Wang would head a new AI research lab . In fact, in July Meta established 'Meta Superintelligence Labs (MSL)' to develop superintelligence.

CEO Mark Zuckerberg announces the establishment of 'Meta Superintelligence Labs' to develop AI that surpasses human knowledge - GIGAZINE

By Cory Doctorow

According to a report from The New York Times, due to the establishment of MSL, Meta has discontinued the development of Llama 4 Behemoth and is instead discussing the development of closed-source AI models. Regarding this report, a Meta spokesperson told technology media TechCrunch, 'We plan to continue releasing our major open source AI models. We have not released everything we have developed so far, and we plan to continue training a combination of open source and closed source models in the future,' indicating their intention to continue developing open source AI models in the future, while also developing closed source AI models.

A Meta spokesperson declined to comment on whether the company has discontinued development of Llama 4 Behemoth. TechCrunch reported that 'if Meta were to prioritize closed-source AI model development, it would be a major change in philosophy for Meta.'

Meta develops open source AI models like Llama, as well as more advanced closed source AI models like those found in its own AI assistant, Meta AI. Meta CEO Mark Zuckerberg has made open source the core of the company's AI strategy and accelerated development. Zuckerberg claims that Llama's openness has differentiated it from competitors such as OpenAI. Meanwhile, rival OpenAI has become more closed since partnering with Microsoft, which Zuckerberg has openly criticized .

How open are 'open source' AI models actually? - GIGAZINE

Despite having some of the world's best AI researchers, Meta still lags behind rivals such as OpenAI, Anthropic, Google DeepMind, and xAI in commercializing AI. TechCrunch points out that if Meta prioritizes the development of closed-source AI models, it could gain additional avenues for monetization.

Furthermore, TechCrunch pointed out that if Meta were to take a step back from open source AI development, it could cede dominance in the open source AI market to China, which has adopted open source AI such as DeepSeek and Moonshot AI.