Meta Platforms’ new AI lab has internally released its first major AI models this month, according to CTO Andrew Bosworth. Speaking at the World Economic Forum in Davos, he highlighted the progress of the Meta Superintelligence Labs team, which was formed last year, noting that their models—only six months into development—are already “very good.”

Bosworth did not reveal which models were delivered, though reports from December suggest Meta is developing a text-based AI model called Avocado, expected in the first quarter, along with an image and video-focused model codenamed Mango. These initiatives come as CEO Mark Zuckerberg reshapes the company’s AI efforts, forming a new lab and recruiting top talent with high-profile offers, aiming to compete with rivals such as Google. Meta previously faced criticism for the performance of its Llama 4 model while competitors advanced.

Bosworth stressed that the technology is still evolving, explaining that post-training work is crucial to make models usable internally and for consumers. Despite this, he said the company is already seeing promising returns from its 2025 investments, which he described as a “tremendously chaotic year” in building infrastructure, the lab, and securing computing power.

Bosworth said 2026 and 2027 will be important years for consumer AI adoption, as current models can already handle everyday questions, with further improvements expected for complex queries. Meanwhile, Meta is focusing on AI-powered consumer products like Ray-Ban Display glasses, pausing international expansion to prioritize U.S. orders.