Meta Building Three New AI Models: Image, Video, and Coding Generation

What Happened

Meta is reportedly developing three new artificial intelligence models that could significantly expand the company’s presence in generative AI, according to a Wall Street Journal report published on December 19, 2025.

The models are being built by Meta Superintelligence Labs (MSL), a specialized AI division created earlier this year. During an internal Q&A session, MSL head and Chief AI Officer Alexandr Wang revealed details about the company’s ambitious AI development roadmap.

The Three Models in Development

1. “Mango” — Image and Video Generation

The most significant announcement is an image and video generation model codenamed “Mango”. This model is expected to launch in the first half of 2026 and will compete directly with:

• Google’s Imagen and Veo
• OpenAI’s ChatGPT Image and Sora
• Runway’s Gen-4.5

Model	Codename	Focus	Expected Release
Image/Video	Mango	Visual generation	H1 2026
Coding LLM	Avocado	Code generation	H1 2026
World Models	—	Physical simulation	Research phase

2. “Avocado” — Coding-Focused LLM

Meta is also developing a text-based large language model (LLM) codenamed “Avocado” with the same tentative H1 2026 release timeline. This model focuses specifically on coding assistance, entering an increasingly competitive market that includes:

• Google’s Antigravity
• OpenAI’s Codex
• Anthropic’s Claude Code

The coding focus suggests Meta may be preparing to either develop its own integrated development environment (IDE) or create tools that integrate with existing IDEs.

3. World Models — A Path to AGI

Perhaps the most technically ambitious development is MSL’s reported breakthrough in world models research.

World models are AI systems that can:

• Learn about the real world based on visual input
• Act as internal simulators
• Predict the physical consequences of actions before taking them

This technology is considered a crucial stepping stone toward artificial general intelligence (AGI) and is particularly important for robotics applications—both humanoid and non-humanoid general-purpose robots.

Why This Matters

Competition Intensifies

Meta’s move into image and video generation comes at a pivotal moment in the AI industry:

• OpenAI recently launched Sora 2 and signed a $1 billion deal with Disney
• Google launched Veo 2 and Imagen 3 earlier this month
• Adobe just partnered with Runway for professional AI video workflows
• Microsoft continues to invest heavily in AI through its OpenAI partnership

Meta entering this space with dedicated models suggests the company believes there’s still significant market share to capture, despite being later to market than competitors.

The Social Media Advantage

Unlike pure AI companies, Meta has immediate distribution through its platforms:

• Facebook (3+ billion users)
• Instagram (2+ billion users)
• WhatsApp (2+ billion users)
• Threads

If Mango integrates directly into these platforms, Meta could instantly make AI image and video generation accessible to billions of users—a scale no competitor can match.

Open Source Strategy

Meta has historically taken an open-source approach with its AI models (LLaMA, Segment Anything). Whether Mango and Avocado follow this pattern could significantly impact the broader AI ecosystem.

The Bigger Picture

Superintelligence Labs’ Role

The creation of Meta Superintelligence Labs earlier this year signaled the company’s serious commitment to advanced AI research. With these three models in development, MSL is tackling:

1. Creative AI (Mango) — Competing with consumer and enterprise video/image tools
2. Developer AI (Avocado) — Targeting the lucrative enterprise coding assistance market
3. Foundational AI (World Models) — Pursuing longer-term AGI research

Timeline Considerations

The H1 2026 target for Mango and Avocado gives Meta approximately 6 months of development time. Given the rapid pace of AI advancement, Meta will need to ensure these models offer compelling advantages over established competitors to gain market traction.

What we’re watching: Whether Meta’s models will be open-source like LLaMA, and how they’ll integrate with Meta’s existing social platforms.

---

Sources

• The Wall Street Journal: Meta Developing New AI Image and Video Model
• Gadgets 360: Meta Reportedly Building Three New Generative AI Models

Related on GenMediaLab

• Adobe and Runway Partner for Next-Gen AI Video
• Disney Invests $1 Billion in OpenAI for Sora
• Google Launches Veo 2 and Imagen 3
• OpenAI Launches Sora 2