Meta is developing new artificial intelligence models for images, video, and text as it tries to regain ground in the fast-moving AI race, according to a report by The Wall Street Journal.
The company is working on an image-and-video AI system internally called “Mango”, alongside a new text-based model known as “Avocado.” Both projects are being developed under Meta’s Superintelligence Lab, which is led by Scale AI co-founder Alexandr Wang. Meta reportedly plans to launch the new models in the first half of 2026.
During an internal company session on Thursday, Wang and Meta’s Chief Product Officer Chris Cox outlined the new AI roadmap. Wang said the upcoming text model is expected to be significantly better at coding, while the visual models will focus on understanding images and videos in more advanced ways, including reasoning, planning, and acting without being trained on every possible scenario.
Meta has recently struggled to keep up with AI competitors such as OpenAI, Google, and Anthropic. Over the past year, the company has restructured its AI division several times, made leadership changes, and recruited researchers from rival firms. However, reports indicate that a number of those researchers have already left Meta’s Superintelligence Lab.
Adding to the uncertainty, Meta’s longtime chief AI scientist Yann LeCun announced last month that he is leaving the company to start his own venture.
Despite placing its Meta AI assistant across Facebook, Instagram, and WhatsApp, the company is still seen as lacking a standout AI product. Much of the assistant’s usage comes from Meta’s massive social media user base rather than strong demand for the AI itself.
As a result, expectations are high for the first wave of models coming out of Meta’s Superintelligence Lab, with the company hoping they can help restore its position as a leader in artificial intelligence.
