Meta Reportedly Building Three New Generative AI Models With Focus on Image and Video Generation

Meta is reportedly developing three new artificial intelligence (AI) models. As per the report, the new AI models are being built by the Menlo Park-based tech giant's specialised AI division, Superintelligence Labs, which was created earlier this year. Among the three, one is said to be specifically focused on generating images and videos, likely to compete with the recent offerings from Google and OpenAI. The company is reportedly also developing a new coding-focused text model as well as world models.

Meta Reportedly Building Multiple AI Models

According to The Wall Street Journal, Meta has codenamed its image and video generation model “Mango.” Citing unnamed people familiar with the conversations, the publication claimed that the Meta Superintelligence Labs' (MSL) head and the Chief AI Officer, Alexandr Wang, spoke about AI models currently in development during an internal Q&A, and revealed that Mango could be released in the first half of 2026.

Another text-based large language model (LLM), codenamed Avocado, is reportedly also being developed by the company with the same tentative release timeline. It is said to be focused on coding. Additionally, MSL has reportedly made a breakthrough in its research on world models, the AI models that can learn about the real world based on visual input.

Based on the reported information, it is clear that Superintelligence Labs' focus is on competing with the established giants in the AI space. The new image and video generation model will directly compete with Google's Nano Banana and Veo, as well as OpenAI's ChatGPT Image and Sora. Image and video generation features have proven to be crowd pleasers, with both companies gaining a substantial number of users on the back of these capabilities.

Even the focus on coding is something that Google (with Antigravity), OpenAI (with Codex), and Anthropic (Claude Code) have actively explored to gain new enterprise clients. Meta's coding-focused model could also be the first steps before it either develops an integrated development environment (IDE) or a tool that can be integrated with an IDE.

Finally, the most interesting reported development is world models. This was something Meta has been researching for years under the leadership of Yann LeCun, who recently announced his departure from the company. These models act as internal simulators to allow the AI system to see a real-world condition and predict the physical consequences of actions before taking them. This is said to be an important step in the path towards artificial general intelligence (AGI), and is considered a key technology for humanoid and non-humanoid general-purpose robots.