Meta has postponed the release of its new AI model with the codename “Avocado” once again. Originally planned for March 2026, the market launch is now expected at the earliest in May 2026. The reason for the delay is sobering: the model cannot keep up with the leading AI offerings from competitors.
Performance deficits compared to competitors
Internal tests have shown that “Avocado” lags behind the latest models from Google, OpenAI, and Anthropic in crucial areas. Particularly in the disciplines of logical reasoning, software development, and “agentic” behavior—that is, the ability to autonomously plan and execute complex tasks—Meta’s model performs worse.
While “Avocado” does outperform Meta’s earlier AI models and Google’s Gemini 2.5, it reportedly stands no chance against Gemini 3.0. This performance gap weighs all the more heavily given that Meta is investing between 115 and 135 billion US dollars in artificial intelligence this year, including the development of its own chips.
Consideration of a Google license
The situation is so precarious that Meta’s AI division leadership has even discussed the possibility of temporarily licensing Google’s Gemini to support its own AI products. While a final decision is still pending, the mere fact that this option is being discussed shows the urgency of the situation.
Strategic shift: from open source to proprietary models
The delay of “Avocado” is symptomatic of a fundamental shift in Meta’s AI strategy. Under CEO Mark Zuckerberg’s clear mandate to pursue “superintelligence” (AGI), there are increasing signs that Meta is moving away from its long-standing open-source approach and pivoting toward proprietary large language models.
“Avocado” itself is expected to be developed as a proprietary model, representing a clear departure from Meta’s previous open-source strategy for the Llama models. This reversal is being driven internally above all by Alexandr Wang, founder of Scale AI, whom Meta brought into the company in June 2025 through a massive investment of 14.3 billion dollars and appointed as Chief AI Officer.
Internal turbulence and personnel upheaval
The strategic shift is accompanied by significant internal restructuring. In the course of focusing on “superintelligence,” hundreds of employees from the traditional FAIR unit (Fundamental Artificial Intelligence Research) were laid off. Even more dramatically, longtime Chief AI Scientist and open-source advocate Yann LeCun announced his departure.
LeCun has since founded his own startup, AMI Labs, in Paris, which develops world models and recently raised a billion dollars in capital. Rather than betting on large language models, LeCun is now building world models that don’t reproduce text but can understand physics and other relationships.
Zuckerberg’s change of heart
Particularly noteworthy is Mark Zuckerberg’s own reversal. The Meta CEO, who not long ago rejected closed platforms with a clear “fuck that” and authored a memorandum titled “Open Source AI is the Path Forward,” appears to be revising his stance.
Intense concern about falling behind OpenAI and Google, along with increasingly complex security concerns in the context of AGI development, are driving this pragmatic shift. Added to this is economic pressure: the new “Avocado” model is reportedly meant to generate revenue, a goal that can only be achieved with a paid, proprietary model.
Further models in the pipeline
Meta is working in parallel on additional AI models. “Mango” is intended to enable high-resolution image and video generation and compete with rivals like OpenAI’s Sora. “Watermelon” is already planned as a successor to “Avocado.” But even the best Llama 4 model, referred to as “Behemoth,” has been delayed for months.
Freemium strategy as likely future model
It is quite possible that Meta will pursue a freemium strategy in the future: the latest, most powerful AI models would be offered as closed source via API for a fee, while older, weaker models would be made freely available via open source. Similar strategies have already been established by Google and OpenAI.
The delay of “Avocado” thus marks not only a technical setback but symbolizes a fundamental shift in Meta’s AI philosophy: from open-source pioneer to commercial competitor in the race for artificial superintelligence.
