Meta to launch new image and video AI models to take on Google Nano Banana

Meta AI

Meta is quietly preparing a major reset of its artificial intelligence strategy, with a new set of image, video and text models now in development and targeted for release in the first half of 2026. The effort is being led by the company’s Superintelligence Lab, which has emerged as Meta’s most important internal AI unit after a turbulent year marked by restructurings, leadership changes and shifting priorities.

Internally, Meta is working on an image and video model codenamed Mango, alongside a separate text-based system referred to as Avocado. The two projects were recently outlined during an internal company-wide session where senior leadership shared a revised roadmap for Meta’s AI ambitions. The message was clear. These models are expected to do more than close gaps with competitors. They are meant to redefine how Meta approaches artificial intelligence across products, research and long-term platform bets.

The text-based model is being positioned as a meaningful upgrade in coding-related tasks, an area where Meta has struggled to match the pace set by rivals. Coding capability has increasingly become a benchmark for evaluating the practical usefulness of large language models, especially among developers and enterprises. Meta’s leadership believes improving performance here is essential if it wants its AI tools to be taken seriously outside its own ecosystem.

Beyond text, Meta is also doubling down on so-called world models. These systems are designed to process and understand visual information in a more holistic way. Instead of being trained exhaustively on every possible scenario, they aim to reason, plan and act within dynamic environments. This shift reflects a broader belief inside Meta that future AI systems must move beyond pattern prediction and towards contextual understanding that can support more autonomous behaviour.

This renewed focus comes at a time when Meta is widely viewed as lagging behind in the AI race. While competitors have managed to turn their models into clear platforms for developers and businesses, Meta has yet to produce a standalone AI product that people actively choose to use. Its AI assistant reaches a massive audience, but largely because it is placed by default inside the company’s social apps. Scale, in this case, has not translated into clear preference or trust.

The internal environment has also been unsettled. Over the past year, Meta’s AI division has gone through repeated reorganisations, with teams reshaped and leadership roles redefined. The company has aggressively recruited researchers from other leading AI labs, but retention has been uneven. Several high-profile hires associated with the Superintelligence Lab have already moved on, adding to concerns about continuity and execution.

That is why Mango and Avocado carry outsized importance internally. These models are not incremental updates designed to keep pace with industry trends. They are intended to form the technical foundation for Meta’s next phase in AI, influencing everything from developer tools to future consumer products.