Meta Reportedly Developing Three New AI Models
Meta is reportedly working on three new artificial intelligence models.
The effort is led by its specialised AI unit, Superintelligence Labs.
The division was formed earlier this year.
According to The Wall Street Journal.
The Menlo Park-based company is expanding its AI ambitions.
The models target images, video, coding, and world understanding.
One model is focused on image and video generation.
It is codenamed Mango.
People familiar with the matter say it could launch in the first half of 2026.
Meta’s Chief AI Officer, Alexandr Wang, reportedly discussed the project.
The details were shared during an internal Q&A session.
Another model is a text-based large language model.
It is codenamed Avocado.
This model is said to be focused on coding tasks.
It may follow a similar release timeline.
Meta is also advancing research on world models.
These models learn about the real world from visual input.
They aim to predict outcomes before actions are taken.
The reported plans show Meta’s competitive intent.
The image and video model would rival Google and OpenAI tools.
These include Google’s Nano Banana and Veo.
They also include OpenAI’s ChatGPT Image and Sora.
Image and video generation has driven strong user growth.
Competitors have seen rapid adoption from these features.
The coding-focused model reflects a wider industry trend.
Google, OpenAI, and Anthropic have all invested in coding AI.
These tools help attract enterprise customers.
Meta’s model could later connect to developer tools.
It may integrate with an IDE or evolve into one.
World models are seen as a major long-term bet.
Meta has researched them for years.
The work was led by AI pioneer Yann LeCun.
Such models are viewed as critical for future AI systems.
They are also important for advanced robotics.
Experts consider them a step toward artificial general intelligence.
Tech Newsroom 24/7 