Meta’s New AI Model for Images and Video Set for 2026
▼ Summary
– Meta is developing new AI models, including an image/video model called “Mango” and a text-based model called “Avocado,” under its superintelligence lab led by Alexandr Wang.
– The company plans to release these new models in the first half of 2026, as announced in an internal roadmap presentation.
– The goal for the text model is to improve its coding capabilities and to develop world models that can understand visual information and reason autonomously.
– Meta is currently behind rivals like OpenAI and Google, and its AI division has faced recent restructurings, leadership changes, and researcher departures.
– Meta lacks a standalone winning AI product, instead relying on its existing social apps to distribute its AI assistant, putting significant pressure on the first models from its new lab.
Meta is intensifying its efforts in artificial intelligence with the development of two major new models, aiming for a significant release in early 2026. According to reports, the company is building an advanced image and video model currently referred to as “Mango” and a sophisticated text-based system internally called “Avocado.” These projects are being spearheaded by the Meta Superintelligence Lab, led by Scale AI co-founder Alexandr Wang, and were detailed in a recent internal presentation with chief product officer Chris Cox.
The strategic goal for the text model, Avocado, is to enhance its capabilities in computer programming and coding. Furthermore, Meta is exploring the creation of novel “world models” designed to comprehend visual data. These systems would possess the ability to reason, formulate plans, and take action without requiring exhaustive training on every conceivable scenario, representing a leap toward more generalized and autonomous AI.
This ambitious roadmap comes at a crucial time for Meta, which has faced challenges in keeping pace with leading AI competitors such as OpenAI, Anthropic, and Google. The company’s AI division has undergone considerable reorganization this year, involving shifts in leadership and the recruitment of researchers from other prominent firms. However, this restructuring has seen some instability, with several researchers who joined the Superintelligence Labs having already departed. In a notable recent development, Meta’s chief AI scientist, Yann LeCun, also announced his exit to launch an independent startup.
Presently, Meta lacks a standalone, market-leading AI product. The adoption metrics for its Meta AI assistant are largely supported by its massive existing user base across platforms like Facebook, Instagram, and WhatsApp, where the assistant is integrated directly into search functions. This context places substantial pressure on the initial outputs from the Superintelligence Lab. The success of the Mango and Avocado models is seen as pivotal for Meta to establish a truly competitive and innovative foothold in the rapidly advancing AI landscape, moving beyond reliance on its social media ecosystem to deliver groundbreaking technology.
(Source: TechCrunch)
