According to the Wall Street Journal, Meta is accelerating the development of a new generation of artificial intelligence models within its "Super Intelligence Lab" headed by Scale AI co-founder Alexandr Wang, including an image and video model code-named "Mango" and a new text model code-named "Avocado." These two models are scheduled to be released externally in the first half of 2026 and are core projects in Meta’s new round of AI strategic roadmap.
According to reports, in an internal Meta Q&A session on Thursday, Wang and chief product officer Chris Cox introduced the new route to employees, making it clear that the company was "all hands on deck" to develop research and development around the new model. Wang said that Meta hopes to significantly improve the coding capabilities of the new generation of text models, while exploring new "world models" that can understand visual information and perform reasoning, planning and action without the need to train separately for every possible scenario.
In the generative AI race, Meta has recently been seen lagging behind competitors such as OpenAI, Anthropic and Google. Since this year, Meta’s AI department has undergone several rounds of major reorganizations, including organizational structure and management adjustments, as well as “poaching” researchers from many top technology companies. However, some researchers who joined Meta Super Intelligence Laboratory (MSL) have left one after another, putting pressure on the stability and external expectations of this new department.
Last month, Meta’s chief AI scientist Yann LeCun also announced that he would leave the company to start his own startup, which was seen as another important loss in the Meta AI lineup. Currently, Meta is still considered to have yet to create a truly “killer” AI product. The usage data of its Meta AI assistant relies more on the scale of billions of users on existing social platforms such as Facebook and Instagram, as well as the company’s distribution strategy of embedding the assistant into the application search bar.
This also means that the first batch of products and models from MSL will carry greater internal and external expectations, and will be regarded as a key node for whether Meta can narrow the gap or even achieve a "comeback" in the new round of AI arms race. For Meta, which hopes to re-prove its technical strength in multiple dimensions such as visual understanding, code generation and general reasoning, the two new models "Mango" and "Avocado" have high hopes.