Meta is quietly gearing up for its next big AI push.
According to a report from The Wall Street Journal, the company is developing new image, video, and text AI models targeted for release in early 2026.
Inside Meta, the effort is being led by its Superintelligence Lab, now overseen by Alexandr Wang, the co-founder of Scale AI. The roadmap was reportedly shared during an internal Q&A session with Meta leadership.
Two models are at the center of the plan. One focuses on visuals. The other on text.
Meet “Mango” and “Avocado”
Meta’s upcoming models carry playful internal names. But the goals behind them are serious.
Mango: Image and Video AI
The model codenamed Mango is designed to handle both image and video generation. Meta hasn’t shared public demos yet, but the goal is clear. The company wants to improve how AI understands and works with visual information.
This includes not just generating images, but understanding scenes, motion, and context.
Avocado: A Smarter Text Model
The second model, known internally as Avocado, focuses on text. Meta reportedly wants it to perform better at coding tasks. That’s a key area where rivals have pulled ahead.
Alexandr Wang has also hinted at something bigger. Meta wants to explore “world models.” These are systems that can reason, plan, and act without being trained on every single scenario.
In simple terms, Meta wants AI that thinks more like humans do.
Why This Matters for Meta
Meta has struggled to keep pace in the AI race. While competitors like OpenAI, Anthropic, and Google continue to release headline-grabbing models, Meta has yet to deliver a clear breakout AI product.
Yes, Meta AI reaches billions of users. But that reach mostly comes from placement inside Facebook, Instagram, and WhatsApp search bars. Not from people choosing the product on its own.
That difference matters.
A Year of Shakeups Inside Meta
2025 has been turbulent for Meta’s AI teams. The company restructured its AI division, changed leadership, and recruited researchers from rival labs.
However, several of those hires have already left.
Adding to the uncertainty, Meta’s long-time chief AI scientist Yann LeCun recently announced plans to leave and start his own company.
That exit raised fresh questions about Meta’s long-term AI direction.
Big Expectations for the Superintelligence Lab
Because Meta lacks a standout AI product today, the pressure on the Superintelligence Lab is intense. The first models released from the lab won’t just be experiments. They will be signals.
Signals to investors.
Signals to developers.
Signals to the wider AI world.
If Mango and Avocado land well, Meta could re-enter the AI conversation in a serious way. If not, the gap with rivals may grow wider.
The Road to 2026
For now, Meta is staying quiet. No public timelines. No flashy demos. Just internal roadmaps and long-term bets.
But one thing is clear. Meta knows it’s behind. And it’s betting that its next generation of models can help it catch up.
The countdown to 2026 has already started.
