Key Highlights:
- Meta plans to launch a closed AI model, codenamed Avocado, likely next spring.
- The move, if true, will be seen a major shift from Meta’s open-source AI strategy, including Llama models.
- CEO Mark Zuckerberg is reportedly closely involved with the new team, dubbed TBD Lab, overseeing development and strategy.
Meta is among the few tech giants that advocate open-source AI (like its Llama models) for innovation, security through community testing, cost-effectiveness for businesses, and preventing gatekeeping. But it seems Meta has realized the future with open-source AI isn’t worth it, as that goodwill won’t pay the bills for its $600 billion investment pledge in the US infrastructure.
Meta is reportedly working on a “closed” model, codenamed “Avocado”
Well, this comes after disappointment around Meta’s recent Llama 4 model. For the uninitiated, Mark Zuckerberg hit the brakes and swapped out the people in the driver’s seat and is now steering Meta toward something much more… Google/OpenAI-like. According to a report from Bloomberg, the upcoming model — codenamed Avocado — will be Meta’s first “money-making AI model.”
Per the report, the model could launch “sometime next spring” as a “closed” model. Apparently, Meta wants tight control over who it sells access to. From what’s reported, Meta’s new AI head, Alexandar Wang, is an advocate of closed models. It’s important to note that open-source models offer developers and researchers the ability to review and build upon the code.
That’s not all. Mark Zuckerberg is also working very closely with the team. Bloomberg reports that the new AI group, dubbed “TBD Lab,” sits around his desk so he can check in constantly. That’s either leadership or micro-management, depending on who you ask inside Meta.
You may also like: Meta Scores Major Coup, Poaches Apple’s Alan Dye
How the “Avocado” model is being trained
The TBD group is reportedly using multiple “third-party models” as part of the training process for Avocado. This includes distillation of rival models like Google’s Gemma, OpenAI’s gpt-oss, and the Alibaba-owned Chinese model Qwen — the same China Zuckerberg warned might be shaping AI with “state censorship.”
Meanwhile, the open-source side of Meta is basically in shadow mode. Employees have been told not to hype Llama. On top of that, Yann LeCun, considered one of the “godfathers” of modern AI, left Meta due to strategic differences. According to different news outlets, LeCun is working on his own French AI startup that will reportedly launch at the end of this year. According to the report, the AI startup has Mark Zuckerberg’s backing.
At this point, there is no official confirmation on when the “closed” model will launch. But new rumors certainly hint that the idea of being “the open AI company” wasn’t as fruitful as Zuckerberg had hoped.
