Meta’s new AI lab delivers first models internally
Meta’s newly created artificial intelligence lab has delivered its first major AI models internally, signaling early progress in the company’s push to compete at the forefront of the AI race.
Speaking at the World Economic Forum in Davos, Meta Chief Technology Officer Andrew Bosworth said the models developed by the Meta Superintelligence Labs team — formed last year — are showing strong promise just months into their development. He described the early results as “very good,” though he noted the technology is still being refined before public release, News.Az reports, citing Reuters.
The lab was created as part of CEO Mark Zuckerberg’s broader effort to rebuild Meta’s AI leadership, attract top talent with aggressive recruitment offers, and accelerate innovation after criticism over the performance of the company’s Llama 4 model.
Media reports have previously suggested Meta is working on a text-based AI model codenamed “Avocado” and an image and video-focused model called “Mango,” though Bosworth did not confirm which models have now been delivered internally.
Bosworth emphasized that post-training work remains significant, explaining that turning AI models into tools that are reliable for internal teams and consumers requires extensive additional development. However, he said Meta is beginning to see returns on its heavy investments in AI infrastructure and computing power.
Looking ahead, Bosworth predicted that 2026 and 2027 will be key years for consumer AI products, as current models are already capable of handling everyday user needs, with more complex capabilities still in progress.
Meta is also expanding its AI-driven hardware ambitions, currently promoting AI-enabled Ray-Ban smart glasses, while temporarily pausing international expansion to meet strong demand in the U.S. market.
Similar news
