Meta’s approach to AI may be entering a new phase. Once a loud advocate of open-source AI through its Llama models, Meta is now reportedly considering moving away from this philosophy. Internal discussions suggest the company might scrap its most powerful upcoming model, Behemoth, in favor of keeping it closed-source. The move follows disappointing internal evaluations and the arrival of a new elite team.
The Superintelligence Lab, Meta’s latest AI division, is at the center of the shift. Helmed by Alexandr Wang and now joined by Shengjia Zhao—one of the key figures behind ChatGPT and GPT-4—the lab has sparked speculation about Meta’s deeper intentions. Some insiders claim that entire teams stopped testing Behemoth after the lab launched, hinting at a lack of confidence in the model’s potential.
Despite this internal turbulence, Meta has not publicly changed course. A company spokesperson insisted that Meta remains committed to open-source development, and plans to continue releasing leading models. Yet, the signals are increasingly mixed. With high-stakes hires and massive investments being funneled into AI, the company appears to be recalibrating priorities behind the scenes.
If Meta chooses to abandon its open-source stance, it could reshape the AI field. Many researchers and startups have relied on Meta’s accessible models as an alternative to proprietary ones from OpenAI and Google. A turn toward closed systems would signal not only a strategy shift but also a new competitive posture, one that prioritizes control over community.

#Meta #AI #OpenSource #Behemoth #SuperintelligenceLab #Llama #ArtificialIntelligence