From Davos to the front lines, Meta frames 2026–27 as the window to translate research into consumer-facing products, citing encouraging returns from prior investments.

Meta emphasizes translating technical progress into usable consumer products and stresses extensive post-training work to ensure reliability and scalability for both internal teams and customers.

Early work has included text-focused Avocado and image/video-focused Mango, two models linked to a Q1 2026 launch, though specifics remain unconfirmed.

Bosworth says the team is about six months into development; the delivered models are promising but require further post-training before practical use.

The push relies on aggressive talent recruitment and expanded compute capacity to convert early results into scalable features across Meta’s platforms.

Meta has formed the Superintelligence Labs to accelerate its generative AI push, delivering its first high-profile internal models this month, as leadership reshuffles and heavy investment press the effort forward.

The move comes as Meta competes with Google and others, with ongoing talent recruitment and post-training work seen as crucial to turning research progress into reliable, consumer-ready products.

CEO Andrew Bosworth forecasts 2026 and 2027 as pivotal years for consumer AI, with current models already answering everyday questions while more advanced capabilities are still in development.

Analysts expect 2026–27 to solidify consumer AI trends, as Meta aims to bring broader products to market and scale more complex features over the next two years.

The internal models have not been named publicly, but executives describe them as strong and promising, with work ongoing after post-training reliability and scalability steps.

Open-source momentum around the LLaMA ecosystem could accelerate scalability once new models launch.

The updates were shared at a Davos press briefing during the World Economic Forum on January 21, 2026.