Meta is mobilizing its resources to develop new artificial intelligence models. This effort is centered within the company’s superintelligence lab, which is led by Scale AI co-founder Alexandr Wang. According to a report, the lab is currently working on two key projects: an image and video model with the internal codename “Mango” and a new text-based model known internally as “Avocado.”
The company’s plan is to release these new models in the first half of 2026. This timeline was shared during an internal question-and-answer session where Wang and Chief Product Officer Chris Cox outlined the new roadmap. During that session, Wang explained that the goal for the text model is to improve its coding capabilities. The team is also exploring advanced world models that can understand visual information and learn to reason, plan, and act without exhaustive training on every possible scenario.
In the competitive landscape of artificial intelligence, Meta has recently trailed behind rivals like OpenAI, Anthropic, and Google. This year, Meta’s AI division underwent several significant restructurings. These changes included shifts in leadership and the strategic hiring of researchers from other leading technology firms. However, a number of the researchers who joined the Meta Superintelligence Labs have since departed the company.
Adding to the turbulence, the company’s chief AI scientist, Yann LeCun, announced last month that he is leaving Meta to start his own venture.
To date, Meta has not launched a breakout standalone AI product. The usage metrics for its Meta AI assistant are primarily supported by the assistant’s integration into the search bars of the company’s existing social networks, which have billions of users. This context means that the initial projects and models produced by the superintelligence lab will carry substantial weight for the company’s future in artificial intelligence.