Meta Platforms’ newly established artificial intelligence research unit has produced its first major AI models for internal use, signalling early progress in the company’s renewed push to lead in advanced AI development. Speaking at a press briefing on the sidelines of the World Economic Forum in Davos, Meta Chief Technology Officer Andrew Bosworth said the output from Meta Superintelligence Labs—set up last year—has been encouraging and validates the company’s aggressive investments in talent and infrastructure.
Providing a timeline for the effort, Bosworth noted that the lab is still in its early stages. “They’re basically six months into the work, not quite even,” he said, adding that the team’s AI models were “very good”. While Bosworth did not disclose specific details about the models, their internal delivery marks an important milestone for a group created as part of CEO Mark Zuckerberg’s broader overhaul of Meta’s AI strategy.
According to a December report by The Wall Street Journal, Meta has been developing multiple AI systems, including a text-based model internally referred to as ‘Avocado’ and an image- and video-focused model codenamed ‘Mango’, both expected to debut in 2025. Bosworth declined to confirm whether either of these models was among those recently delivered to internal teams.
Meta’s AI progress is being closely scrutinised after Zuckerberg reorganised leadership, launched the Superintelligence Labs, and moved swiftly to recruit top researchers with highly competitive compensation packages. The goal is to place Meta at the forefront of foundational and consumer-facing AI systems amid intense competition from rivals such as OpenAI, Google, and Anthropic.
Bosworth also stressed that developing impactful AI goes beyond training models. “There’s a tremendous amount of work to do post-training to actually deliver the model in a way that’s usable internally and by consumers,” he said, pointing to deployment, integration, and reliability as equally critical challenges.
He described 2025 as a “tremendously chaotic year” for Meta, driven by rapid data centre expansion, rising compute demands, and the need to secure long-term power capacity to sustain AI workloads. Despite this turbulence, Bosworth said the company is now beginning to see tangible returns from those investments.
Looking ahead, he identified 2026 and 2027 as defining years for consumer AI, as systems mature from answering everyday questions to becoming widely adopted, mass-market products. Meta has already begun moving in that direction with AI-powered smart glasses developed with Ray-Ban, though it has temporarily paused international expansion to meet strong demand in the US.
