Meta Delivers New ‘Avocado’ and ‘Mango’ AI Models Internally: Public Release Planned for H1 2026
Meta’s newly formed Superintelligence Labs has delivered its first major AI models internally just six months after formation, CTO Andrew Bosworth announced at the World Economic Forum in Davos.
Two New Models
According to Reuters, Meta is developing two distinct AI models:
- Avocado: A text-focused large language model positioned as Llama’s successor, targeting improved code generation and reasoning capabilities. Q1 2026 release expected
- Mango: A multimodal model specialized in image and video generation, designed to compete with ChatGPT and Gemini
Bosworth described early results as “very good,” indicating significant progress toward production readiness.
Six Months of Rapid Development
Superintelligence Labs emerged from CEO Mark Zuckerberg’s aggressive moves to shake up Meta’s AI leadership and poach talent from competitors with sky-high offers.
This acceleration followed the departure of former Meta chief AI scientist Yann LeCun to found a “world models” company. Some executives reportedly considered switching from Llama to competing models before recommitting to in-house innovation.
Market Implications
Industry analysts suggest the success of Mango and Avocado will heavily influence how quickly Meta can reclaim a leadership position in AI:
- Mango: Potential step-change in visual AI capabilities
- Avocado: Direct competition with OpenAI and Google in coding and reasoning
2026 is predicted to be a decisive year when multimodal AI, agent-based systems, and world models enter the mainstream. These models are expected to accelerate AI integration across Meta’s Facebook, Instagram, and WhatsApp products.