Meta is developing new image, video, and text AI models for 2026 as it tries to regain ground in an increasingly crowded AI race.
Meta is developing a new generation of artificial intelligence models designed for image, video, and text applications, with plans to release them in the first half of 2026, according to a report by The Wall Street Journal.
The effort is being led by Meta’s recently formed Superintelligence Lab, overseen by Alexandr Wang, the co-founder of Scale AI, whom Meta recruited to help accelerate its AI ambitions. Internally, the company is working on an image-and-video model codenamed Mango, alongside a text-based system known as Avocado, the report said.
The roadmap was outlined during an internal question-and-answer session at Meta this week, where Wang appeared alongside Chris Cox, the company’s chief product officer. According to the report, Meta intends to improve its text models’ coding capabilities while also developing so-called world models—systems designed to understand visual information and reason, plan, and act without being explicitly trained on every possible scenario.
Also Read: GenAI Search Is Rewriting the Shopper’s Playbook
The push comes as Meta attempts to reassert itself in an AI race increasingly dominated by rivals such as OpenAI, Anthropic, and Google. While Meta was once seen as a leader in open research, it has struggled to translate that work into widely recognized AI products.
The company’s AI organization has undergone repeated restructurings over the past year, including leadership changes and aggressive recruitment from competing labs. Some of those hires, however, have already departed Meta’s Superintelligence Lab, underscoring the difficulty of stabilizing a top-tier research team.
Adding to the uncertainty, Yann LeCun, Meta’s longtime chief AI scientist, announced last month that he plans to leave the company to start his own venture—raising further questions about continuity at a critical moment.
So far, Meta has yet to produce a breakout AI product. Its Meta AI assistant reaches a massive audience largely because it is embedded directly into the company’s social platforms, rather than through independent demand. That distribution advantage masks a more basic challenge: proving that Meta can deliver AI systems that compete on capability, not just reach.
Also Read: Edition 3: Sharp Minds Decoding Tech’s Next Wave
As a result, the first models to emerge from the Superintelligence Lab will carry unusually high expectations. For Meta, they are not just another product cycle—they are a test of whether the company can still shape the direction of artificial intelligence, rather than merely follow it.
