Speaking on Wednesday at a press briefing on the sidelines of the World Economic Forum in Davos, Meta Chief Technology Officer Andrew Bosworth said the company’s Meta Superintelligence Labs team had delivered several high-profile AI models internally this month. The lab was established last year as part of a broader shake-up of Meta’s AI strategy.
Bosworth said the models, developed in less than six months, showed strong potential, though he declined to identify them. “They’re basically six months into the work, not quite even,” he said, describing the results as “very good” and promising, while stressing that development remains ongoing.
Media reports in December said Meta was working on a text-based AI model codenamed “Avocado,” expected to debut in the first quarter, alongside an image- and video-focused system codenamed “Mango.” Bosworth did not confirm whether either of those projects had been among the models delivered internally.
Meta’s AI efforts have drawn intense scrutiny after Chief Executive Mark Zuckerberg reorganized leadership, launched the new lab and aggressively recruited top AI researchers with lucrative compensation packages. The moves were designed to sharpen Meta’s competitive edge against rivals such as Alphabet’s Google and other major players that have recently gained ground in the race to commercialize advanced AI systems.
The company has faced criticism over the performance of its Llama 4 model, which fell short of expectations at a time when competitors were unveiling increasingly capable and commercially attractive alternatives. Bosworth acknowledged that the work underway at Meta is far from complete.
“There’s a tremendous amount of work to do post-training,” he said, referring to the steps required to make AI models reliable, scalable and usable for both internal teams and consumers.
Bosworth described 2025 as a “tremendously chaotic year” for Meta, marked by heavy investment in talent, computing infrastructure and power capacity. However, he said the company is beginning to see positive returns from those bets.
Looking ahead, Bosworth said 2026 and 2027 are likely to be pivotal years for consumer-facing AI products. He argued that recent advances have already produced models capable of handling everyday questions and interactions, even as more complex tasks still require further breakthroughs.
“That’s why the next couple of years are so important,” he said, pointing to the opportunity to translate technical progress into widely used consumer products.
Meta is already pushing AI-enabled hardware, including its Ray-Ban smart display glasses. Earlier this month, the company paused international expansion of the device to focus on meeting strong demand in the U.S., underscoring its near-term emphasis on consumer markets as its AI strategy evolves.