Meta develops Mango AI model for images and videos, targets first half of 2026 launch

Meta Unveils Ambitious AI Roadmap With Mango and Avocado Models

Meta is developing new artificial intelligence models under its superintelligence lab, with plans to release both an image and video model codenamed Mango AI and a text-based large language model called Avocado in the first half of 2026 1

1

2

2

. The tech giant's chief AI officer, Alexandr Wang, and chief product officer Chris Cox unveiled the roadmap during an internal question-and-answer session held on Thursday, according to sources who spoke with The Wall Street Journal 2

2

.

The announcement comes as Meta intensifies its efforts to catch up with rivals in the AI race, particularly OpenAI, Google, and Anthropic 1

1

. Following reports of the new AI initiative, Meta shares rose 2.30 percent in trading, reflecting investor confidence in the company's generative AI strategy 2

2

.

Mango Targets Visual AI Sector With Multimodal Capabilities

The new image and video model represents Meta's push into multimodal AI capabilities, designed to generate advanced images and videos in response to user requests 2

2

3

3

. With this 2026 release, Meta aims to position itself strategically as a competitor to Google's Gemini Nano Banana, which established its position in the creative AI ecosystem after its 2025 release 3

3

.

Image and video generation has become a key battleground for AI companies. Meta previously released an AI video generator called Vibes in September, developed in collaboration with startup Midjourney, while OpenAI launched its own video generator, Sora, less than a week later 4

4

. Google has also seen success in this space, with its Nano Banana image tool helping drive rapid growth for Gemini, increasing monthly users from 450 million in July to more than 650 million by late October 4

4

.

Avocado Model Emphasizes Coding Capabilities and World Models

Alongside Mango, Meta is developing Avocado, a text-based large language model also slated for early 2026 2

2

4

4

. Alexandr Wang stated that one of Avocado's main goals is to become much better at coding tasks 4

4

. Additionally, Wang said Meta aims to explore new world models that understand visual information and can reason, plan, and act without needing to be trained on every possibility 1

1

.

These world models aim to learn about the real world by taking in visual information, helping AI better understand its surroundings 4

4

. The focus on coding capabilities and visual reasoning suggests Meta is targeting multiple fronts in the competitive landscape of generative AI.

Meta Superintelligence Labs Faces High Stakes

Over the summer, Meta reorganized its AI efforts and hired Alexandr Wang, Scale AI co-founder, to lead a new group called Meta Superintelligence Labs 1

1

4

4

. CEO Mark Zuckerberg played a hands-on role in building the team, recruiting more than 20 researchers from OpenAI. The new division now includes over 50 researchers, engineers, and other AI experts 4

4

.

However, the company's AI division has seen significant restructurings this year, including leadership changes and the poaching of researchers from other top companies. Several of the researchers who joined Meta Superintelligence Labs have already left the company 1

1

. Last month, the company's chief AI scientist, Yann Lecun, also announced that he's leaving to create his own startup .

Meta doesn't have a winning AI product as of yet. Instead, Meta AI assistant's numbers are buoyed by the company's existing social networks spanning billions of users, since the company places the assistant in the search bar of its apps . This means the first projects and models coming out of Meta Superintelligence Labs will have a lot riding on them, making the success of Mango and Avocado critical for Meta's competitive position in the visual AI sector and broader multimodal AI landscape.