Meta's Superintelligence Lab Discusses Abandoning Key Open-Source AI Model, Behemoth, Amid Performance Concerns

Meta Platforms' newly formed Superintelligence Lab is reportedly considering a significant shift in its artificial intelligence strategy, potentially moving away from its long-standing open-source approach. Discussions within the lab, led by new Chief AI Officer Alexandr Wang, involve abandoning the powerful open-source Behemoth model. This potential pivot follows internal performance issues with Behemoth, leading to its delayed release, as reported by The New York Times. The strategic re-evaluation could mark a major philosophical change for the tech giant.

For years, Meta has championed open-source AI, notably with its Llama family of models, positioning itself as a counterpoint to competitors like OpenAI. However, sources indicate that while training on Behemoth was completed, its release was halted due to underwhelming internal testing results. Testing on the model reportedly ceased after the Superintelligence Lab's launch.

Alexandr Wang, who founded Scale AI—a company in which Meta recently invested $14.3 billion—is at the forefront of these discussions within the Superintelligence Lab. This elite team works closely with CEO Mark Zuckerberg, operating separately from Meta's broader AI division. The company is pouring billions into AI development, including massive talent acquisition and infrastructure costs, intensifying pressure to commercialize its AI efforts.

A move towards more proprietary, closed-source AI models could provide Meta greater control over its intellectual property and open new avenues for monetization. While Meta already uses advanced closed-source models internally, its external strategy has largely relied on openness. This potential shift suggests that Meta's commitment to open-source may be more strategic than purely ideological, driven by the need to generate revenue from its substantial AI investments.

Such a pivot by Meta could significantly impact the broader AI landscape, potentially slowing the momentum behind open-source initiatives that Meta has largely spearheaded. It might also shift power dynamics back towards major players with closed ecosystems, posing challenges for smaller companies and startups reliant on open-source foundation models. Globally, a retreat from open-source by Meta could indirectly bolster nations like China, which are increasingly embracing open-source AI projects.

Despite these internal discussions, a Meta spokesperson clarified that the company's official position on open-source AI remains "unchanged." The spokesperson stated, "We plan to continue releasing leading open source models. We haven’t released everything we’ve developed historically and we expect to continue training a mix of open and closed models going forward." This statement indicates a nuanced approach, leaving room for both open and closed models in Meta's future AI development.