Introduction#
Meta's shares dipped in premarket trading following a report from the New York Times indicating that the company's latest AI model, Avocado, has not performed as well as competing models from other tech giants. This underperformance has led to a delay in its launch.
Performance Issues#
The Avocado model, which is intended to be a foundational AI system, has struggled in internal tests, particularly in tasks related to reasoning, coding, and writing. While it has shown improvements over Meta’s previous AI models, it still falls short when compared to Google's latest model, Gemini 3.0, which was released in November. As a result, Meta has postponed the launch of Avocado to at least May, pushing it back from an initial target of this month.
Strategic Decisions#
In light of these challenges, Meta's AI division is considering temporarily licensing Google’s Gemini models to enhance some of its AI products. This decision is still under discussion, and no final agreements have been reached. Meta has been heavily investing in AI development, especially after its earlier model, Llama 4, did not meet expectations last year. Recently, the company invested $14.3 billion in Scale AI and appointed its CEO, Alexandr Wang, as its new chief AI officer.
Future Directions#
Wang has been instrumental in forming a new research group, TBD Lab, which is working on Avocado and another model called Mango, aimed at generating images and videos. The team has completed the initial training phase for Avocado and is now in the post-training stage. However, some researchers have left the group ahead of the model's release. Additionally, there is ongoing debate within Meta about whether the new AI system should be open source or kept closed, with recent tendencies leaning toward a closed model despite the company's historical support for open-source initiatives.