Meta’s new AI is likely to be monetized

Meta CEO Mark Zuckerberg, once one of the biggest advocates of open source artificial intelligence and famously declaring in a letter that “open source is the way of the future,” appears to be making a 180-degree turn. New reports suggest that Meta is secretly working on a new model called “Avocado” which, unlike the popular Llama series, may be completely closed and proprietary. This change of approach shows that the competition to reach superintelligence has become so serious that meta is willing to violate its previous principles.

According to Bloomberg, Avocado is the code name of a new project that is being developed at Meta Advanced Labs and is expected to be released in 2026. Unlike previous meta models where the code was publicly available, Avocado will likely be a closed model. The project is led by a small group called TBD under the supervision of Meta’s new AI chief, Alexander Wang. A fan of closed models, Wang apparently managed to change Zuckerberg’s mind.

Artificial Intelligence Avocado Meta

The development of the Llama 4 model (codenamed Behemoth) seems to have run into serious problems. The New York Times previously reported that Meta executives even considered stopping the project altogether, as early versions failed to meet expectations. Long delays and lack of developer response to Llama 4 betas have been one of the main reasons for the meta’s move to the alternative Avocado project.

In general, Zuckerberg is very worried that he will lose the competition with OpenAI and Google. Meta plans to spend $600 billion on AI infrastructure in the coming years. When such huge capital is at stake, and security concerns about superintelligence become serious, free sharing of results with competitors (open source) is no longer economically and strategically justified. Also, the departure of prominent figures such as “Yan Lacan” (who was a staunch defender of open source) from Meta, indicates the victory of the faction favoring closed models in this company.