Mark Zuckerberg has for months publicly hinted that he is backing away from open-source AI models. Now, Meta’s latest AI pivot is starting to come into focus. The company is reportedly working on a new model, known inside of Meta as “Avocado,” which could mark a major shift away from its previous open-source approach to AI development.
Both CNBC and Bloomberg have reported on Meta’s plans surrounding “Avocado,” with both outlets saying the model “could” be proprietary rather than open-source. Avocado, which is due out sometime in 2026, is being worked on inside of “TBD,” a smaller group within Meta’s AI Superintelligence Labs that’s headed up by Chief AI Officer Alexandr Wang, who apparently favors closed models.
It’s not clear what Avocado could mean for Llama. Earlier this year, Zuckerberg said he expected Meta would “continue to be a leader” in open source but that it wouldn’t “open source everything that we do.” He’s also cited safety concerns as they relate to superintelligence. As both CNBC and Bloomberg note, Meta’s shift has also been driven by issues surrounding the release of Llama 4. The Llama 4 “Behemoth” model has been delayed for months; The New York Times reported earlier this year that Wang and other execs had “discussed abandoning” it altogether. And developers have reportedly been unimpressed with the Llama 4 models that are available.
There have been other shakeups within the ranks of Meta’s AI groups as Zuckerberg has spent billions of dollars building a team dedicated to superintelligence. The company laid off several hundred workers from its Fundamental Artificial Intelligence Research (FAIR) unit. And Meta veteran and Chief AI Scientist Yann LeCun, who has been a proponent for open-source and skeptical of LLMs, recently announced he was leaving the company.
That Meta may now be pursuing a closed AI model is a significant shift for Zuckerberg, who just last year said “fuck that” about closed platforms and penned a lengthy memo titled “Open Source AI is the Path Forward.” But the notoriously competitive CEO is also apparently intensely worried about falling behind OpenAI, Google and other rivals. Meta has said it expects to spend $600 billion over the next few years to fund its AI ambitions.
This article originally appeared on Engadget at https://www.engadget.com/ai/meta-is-reportedly-working-on-a-new-ai-model-called-avocado-and-it-might-not-be-open-source-215426778.html?src=rss
