Meta Rethinks Open Source Strategy as ‘Avocado’ AI Model Faces Potential Shift to Closed Source
Table of Contents
Meta is reportedly considering a dramatic shift in its artificial intelligence strategy, potentially converting its next-generation AI model, ‘Avocado,’ to a closed-source system. This move comes after billions of dollars in investment and a major personnel overhaul, triggered by a disappointing reception to the open-source Llama 4 model and concerns over intellectual property.
The decision, first reported by The Deep View on December 11, 2025, signals a significant departure from Meta’s previously staunch commitment to open-source AI development. CEO Mark Zuckerberg initiated a comprehensive review of the company’s AI approach following the lukewarm response from developers to Llama 4, released last April.
Aggressive Investment and a New AI Leadership
In an effort to bolster its AI capabilities, Meta embarked on an aggressive recruitment drive, aiming to build what Zuckerberg described as “an elite organization with a large number of the industry’s best AI experts.” This included a substantial $14.3 billion (approximately 20 trillion won) investment in data labeling company Scale AI last June. Simultaneously, the company appointed Alexandr Wang, 28, founder and CEO of Scale AI, as Meta’s inaugural Chief AI Officer.
Wang now leads a team internally known as ‘TBD Lab,’ and reports suggest the team is attracting talent with compensation packages exceeding $100 million. “Meta is clearly signaling its commitment to becoming a leader in AI, and is willing to invest heavily to achieve that goal,” one analyst noted.
The Double-Edged Sword of Open Source
The impetus for considering a closed-source model for ‘Avocado’ reportedly stemmed from the discovery that Chinese AI firm DeepSeek’s R1 model incorporated elements of Meta’s Rama architecture. According to a CNBC report, this revelation sparked internal complaints within Meta. DeepSeek had developed its R1 model based on the Rama 3 series.
However, Meta also benefits from open-source contributions. Bloomberg reported that TBD Lab is currently leveraging China’s Alibaba’s Qwen model in the training process for ‘Avocado,’ highlighting the complex interplay of open and closed systems in the AI landscape.
Internal Friction and Key Departures
Meta’s rapid AI reorganization has not been without internal strife. Yann LeCun, Meta’s chief AI scientist and a pioneering figure in deep learning, announced his departure last November to launch his own startup. The Turing Award winner reportedly expressed dissatisfaction with Meta’s decision to lay off 600 employees at the Meta Superintelligence Laboratory (MSL) in October.
This week saw further departures from MSL, with Sang Michael Xie and Vitaliy Chiley also leaving the company. According to The New York Times, these departures were fueled by disagreements over strategic direction. Some senior executives advocated for focusing AI efforts on enhancing Meta’s core social media and advertising businesses, while Wang championed a strategy of catching up to cutting-edge models from OpenAI and Google before prioritizing product applications.
An Uncertain Future in a Competitive Landscape
Industry observers suggest Meta’s move towards a closed-source approach reflects its struggle to compete with other open-source AI models. Maintaining Llama’s open-source status offered Meta a unique differentiator, but the current situation presents Wang with the challenge of surpassing the proprietary models offered by industry giants like OpenAI, Google, and Anthropic.
Deep View assessed the situation, stating, “Like the name of the team led by Wang (TBD stands for ‘To Be Determined’), its chances of success are still uncertain.” The release of the ‘Avocado’ model, initially slated for the end of 2025, has now been postponed to the first quarter of 2026.
As Meta navigates this pivotal moment, the future of its AI strategy – and its position in the rapidly evolving AI landscape – remains to be seen.
