Meta Develops Mango and Avocado AI Models for 2026

Introduction

Meta Platforms is preparing a major technological offensive in the AI race, developing two next-generation artificial intelligence models designed to challenge the dominance of OpenAI and Google. According to a report from The Wall Street Journal on December 18, 2025, the social media giant is working on Mango, a high-stakes image and video-focused AI model, and Avocado, its next major text-based large language model (LLM).

This development represents Meta's first major output from Meta Superintelligence Labs (MSL), a specialized division created after a major restructuring in the summer of 2025. The ambitious roadmap was detailed during an internal Q&A session featuring Meta's Chief AI Officer Alexandr Wang and Chief Product Officer Chris Cox, signaling Meta's commitment to moving beyond the incremental updates of the current Llama family.

The Strategic Vision

Beyond Llama: A New Generation

Meta's development of Mango and Avocado marks a significant strategic shift. While the Llama family has been successful in the open-source AI community, these new models represent Meta's attempt to compete directly with closed-source frontier models from OpenAI and Google.

Key Strategic Elements:

First-half 2026 timeline: Aggressive development schedule targeting competitive release
Dual-model approach: Specialized models for different modalities (image/video vs. text)
World models research: Early-stage development of AI systems that understand physical reality
Coding focus: Addressing a traditional weakness in Meta's model portfolio

The Investment Behind the Strategy

Meta's commitment to this initiative is substantial. The company invested over $14 billion to acquire a near-majority stake in Scale AI, cementing Alexandr Wang's role as the primary architect of Meta's post-Llama strategy. This investment demonstrates CEO Mark Zuckerberg's willingness to spend aggressively to close the gap with rivals.

Meta Superintelligence Labs

Organizational Structure

Meta Superintelligence Labs was created following a major restructuring in the summer of 2025. This specialized division represents Meta's focused effort to develop next-generation AI capabilities:

Leadership:

Alexandr Wang: Chief AI Officer, 28-year-old founder of Scale AI
Chris Cox: Chief Product Officer, involved in strategic planning
Mark Zuckerberg: CEO, personally orchestrated high-profile recruiting efforts

Talent Acquisition:

Poached more than 20 researchers from OpenAI
Hired Wang to oversee the effort
Significant investment in building a world-class AI research team

First Major Output

Mango and Avocado represent the first major output from MSL, establishing the division as a key player in Meta's AI strategy. The development timeline and capabilities suggest these models will be significantly more advanced than previous Meta offerings.

Mango: Image and Video AI Model

Capabilities and Focus

Mango is Meta's next-generation model focused on image and video generation and understanding:

Expected Capabilities:

Advanced image generation and editing
Video creation and analysis
Multimodal understanding combining visual and textual information
Integration with world models research

Strategic Positioning:

Designed to compete with models like OpenAI's Sora and Google's video generation capabilities
High-stakes development indicating Meta's commitment to visual AI
Part of broader multimodal AI strategy

Market Context

The image and video AI market has seen rapid growth, with models like:

OpenAI's Sora: Advanced video generation
Google's Gemini: Strong multimodal capabilities
Stable Diffusion: Open-source image generation

Mango represents Meta's attempt to establish a competitive position in this space, moving beyond the text-focused Llama models.

Avocado: Advanced Text LLM

Coding-Focused Development

Avocado is Meta's next major text-based large language model, with a specific focus on addressing previous weaknesses:

Key Focus Areas:

Advanced coding capabilities: Traditional weak point for Meta models
Competitive performance: Designed to match or exceed OpenAI and Google models
Production readiness: Built for real-world applications

Development Priorities:

Coding performance comparable to GPT-4 and Gemini
Enhanced reasoning capabilities
Better tool use and agentic workflows

Competitive Landscape

Avocado enters a competitive market dominated by:

OpenAI's GPT models: Industry-leading coding capabilities
Google's Gemini: Strong reasoning and multimodal understanding
Anthropic's Claude: Advanced safety and coding features

Meta's focus on coding capabilities suggests Avocado will prioritize developer use cases and agentic applications.

World Models: The Future Vision

Understanding Physical Reality

During the internal session, Wang noted that Meta is in the early stages of developing world models. This represents a significant shift in AI development:

Current LLM Approach:

Predict the next word in a sequence
Text-based understanding
Limited physical world comprehension

World Models Vision:

Understand physical reality through visual information
Process vast amounts of visual data
Comprehend how the physical world works

Implications

World models could enable:

Robotics applications: AI systems that understand physical environments
Autonomous systems: Better navigation and interaction with the real world
Scientific research: Understanding physical processes and phenomena
Augmented reality: Enhanced AR experiences with better world understanding

This research direction aligns with Meta's focus on the metaverse and AR/VR technologies, suggesting potential integration with Meta's hardware and software ecosystem.

The Scale AI Connection

Strategic Acquisition

Meta's investment in Scale AI represents one of the largest AI-related acquisitions:

Investment Details:

Over $14 billion for near-majority stake
Alexandr Wang becomes Meta's Chief AI Officer
Scale AI's expertise in data labeling and AI infrastructure

Strategic Value:

Access to Scale AI's data labeling capabilities
Infrastructure for training large models
Expertise in building production AI systems
Established relationships with AI research community

Leadership Transition

Alexandr Wang's transition from Scale AI founder to Meta's Chief AI Officer represents a significant talent acquisition. At 28 years old, Wang brings:

Experience building AI infrastructure companies
Understanding of production AI systems
Connections in the AI research community
Vision for next-generation AI development

Competitive Positioning

Against OpenAI

Meta's development of Mango and Avocado positions the company to compete with OpenAI across multiple fronts:

Text Models:

Avocado vs. GPT-4 and future GPT models
Focus on coding capabilities to match OpenAI's strengths
Competitive reasoning and tool use

Multimodal Models:

Mango vs. Sora and other OpenAI visual models
Advanced image and video generation
Integration with text capabilities

Against Google

Google's Gemini family represents another key competitor:

Model Comparison:

Avocado vs. Gemini Pro and Ultra models
Mango vs. Gemini's multimodal capabilities
World models research vs. Google's multimodal understanding

Strategic Differences:

Meta's focus on open-source community (via Llama)
Google's integration with search and productivity tools
Different approaches to multimodal AI

Timeline and Expectations

First-Half 2026 Release

Both Mango and Avocado are expected to release in the first half of 2026, according to the internal Q&A session. This timeline suggests:

Development Status:

Active development underway
Significant progress made since MSL formation
Aggressive timeline to compete with rivals

Market Timing:

Potential release alongside or before GPT-5
Competition with Google's Gemini updates
Opportunity to establish market position

What to Expect

Based on the information revealed:

Mango:

Advanced image and video generation
Multimodal understanding capabilities
Integration with Meta's platforms

Avocado:

Strong coding performance
Competitive reasoning capabilities
Production-ready features

World Models:

Early-stage research
Long-term vision for physical world understanding
Potential applications in AR/VR and robotics

Implications for the AI Industry

Market Dynamics

Meta's aggressive investment and development signal:

Increased Competition:

Three major players (OpenAI, Google, Meta) competing for leadership
Significant capital investment in AI development
Talent war intensifying with poaching from competitors

Innovation Acceleration:

Faster model development cycles
Focus on specialized capabilities (coding, multimodal)
Research into next-generation AI approaches (world models)

Open Source vs. Closed Source

Meta's history with Llama suggests potential open-source releases, but Mango and Avocado may follow a different strategy:

Considerations:

Competitive pressure may favor closed-source models
Investment recovery may require commercial licensing
Balance between open-source community and competitive advantage

Technical Challenges and Opportunities

Development Challenges

Building models that compete with OpenAI and Google presents several challenges:

Technical Hurdles:

Matching or exceeding coding capabilities
Advanced multimodal understanding
Efficient training and inference
World models research is still early-stage

Resource Requirements:

Significant compute infrastructure
Large-scale data collection and processing
Talent acquisition and retention
Research and development investment

Opportunities

Meta's unique position offers several advantages:

Platform Integration:

Integration with Facebook, Instagram, WhatsApp
AR/VR hardware ecosystem (Quest, Ray-Ban Meta)
Massive user base for testing and deployment

Research Capabilities:

FAIR (Facebook AI Research) expertise
Access to vast amounts of user-generated content
Infrastructure for large-scale training

Conclusion

Meta's development of Mango and Avocado represents a significant escalation in the AI race, with the company making substantial investments to compete with OpenAI and Google. The strategic acquisition of Scale AI, hiring of Alexandr Wang, and formation of Meta Superintelligence Labs demonstrate Meta's commitment to becoming a leader in next-generation AI.

Key Takeaways

Dual-model strategy: Mango for image/video, Avocado for text with coding focus
Aggressive timeline: First-half 2026 release target
Substantial investment: Over $14 billion in Scale AI acquisition
Talent acquisition: Poached 20+ researchers from OpenAI, hired Scale AI founder
Future vision: Early-stage world models research for physical reality understanding

What This Means

For the AI industry, Meta's entry into high-stakes model development increases competition and accelerates innovation. For developers, the potential release of advanced coding-focused models (Avocado) and multimodal capabilities (Mango) offers new tools and possibilities. For Meta, this represents a critical strategic move to establish leadership in AI and support its metaverse and AR/VR ambitions.

The success of Mango and Avocado will depend on their ability to match or exceed the capabilities of existing frontier models while offering unique advantages. With the first-half 2026 timeline, the AI community will soon see whether Meta's substantial investment and strategic vision translate into competitive AI models that can challenge the current market leaders.

Sources

Interested in learning more about AI models and their capabilities? Explore our AI models section, check out our glossary of AI terms, or discover other AI tools in our comprehensive catalog.