Meta Superintelligence Labs delivers first internal AI models, CTO Andrew Bosworth reveals

Meta Superintelligence Labs Achieves Early Milestone

Meta AI has reached a significant milestone as its newly established Meta Superintelligence Labs delivered its first internal AI models this month, CTO Andrew Bosworth confirmed at the World Economic Forum in Davos, Switzerland1

1

3

3

. Speaking at a press briefing, Bosworth described the first internal AI models as "very good," noting the team had been working for barely six months2

2

. The development marks an important step in Mark Zuckerberg's efforts to position the company at the forefront of artificial intelligence, following a major overhaul of Meta's AI leadership and the formation of the specialized lab last year.

Avocado and Mango Models Under Development

While Bosworth did not specify which in-house AI models had been delivered internally, reports from December indicate Meta has been developing two significant projects: a text-based 'Avocado' model scheduled for first-quarter launch and an image/video-focused 'Mango' model1

1

4

4

. The CTO emphasized that building usable systems extends far beyond initial training. "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," Bosworth explained2

2

. This acknowledgment comes as Meta faces scrutiny over the performance of its Llama 4 model amid competitive launches from rivals like Google, OpenAI, and Microsoft.

Aggressive AI Talent Recruitment and Infrastructure Push

Meta's progress reflects months of intensive AI talent recruitment, with the company offering sky-high compensation packages to attract top researchers to Meta Superintelligence Labs3

3

4

4

. Bosworth characterized 2025 as a "tremendously chaotic year" for the company, marked by rapid AI infrastructure investment, expanded computing capacity, and efforts to secure sufficient power to support its AI ambitions1

1

. Despite the challenges, he said Meta is beginning to see returns on those investments, with the infrastructure buildout creating a foundation for future development.

Consumer AI Products on the Horizon

Looking ahead, Bosworth identified 2026 and 2027 as pivotal years for consumer AI products, suggesting Meta aims to translate its technical capabilities into mass-market offerings1

1

4

4

. He noted that recent advances have produced systems capable of handling everyday questions from users and families, with the next challenge being broader deployment. Meta has already begun rolling out AI-powered hardware, including smart glasses developed with Ray-Ban, though international expansion was recently paused to focus on strong US demand1

1

. However, analysts have questioned Meta's monetization strategy, contrasting it with competitors like Microsoft, which monetizes through Azure and Copilot subscriptions, and Google, which sells Gemini access through its cloud division2

2

. Meta currently uses its AI for recommendation engines, ad ranking, and engagement tools, raising questions about how these capabilities will contribute to revenue growth.