On January 21, 2026, at a press briefing on the sidelines of the World Economic Forum in Davos, Meta CTO Andrew Bosworth confirmed that Meta Superintelligence Labs — the unit assembled after Mark Zuckerberg’s 2025 overhaul of the company’s AI organization — has delivered its first high-profile AI models internally this month. That is roughly six months into the lab’s existence. Bosworth called the models “very good” and said they had “shown significant promise,” while declining to name them and cautioning that “there’s a tremendous amount of work to do” in post-training before the technology can be rolled out across the company and to consumers.
The context is what makes this worth attention. Llama 4 drew criticism for lagging rivals while Google and OpenAI kept raising the bar, and MSL — built through aggressive, expensive talent raids — is Meta’s answer. First internal models at the six-month mark is the first verifiable evidence that the reorganization actually produces output.
First Models in Six Months: Progress and Caveats
Per December media reports, Meta had two codenames in flight: a text model dubbed Avocado, targeted for a Q1 launch, and a multimodal image-and-video model dubbed Mango. Bosworth’s timeline lines up with that reporting, but he did not confirm which models had landed — only that the team had delivered high-profile results. The caveat matters as much as the milestone: post-training is the phase that turns a pretrained checkpoint into something usable, and by his own account there is still “a tremendous amount of work” left there.
Bosworth framed 2025 as a “tremendously chaotic year” but argued the bets — lab-building, infrastructure, power procurement — are already showing favorable returns. Read that as both a progress report and a defense of the spend: the money is committed, the models are starting to appear, but the accounting comes later.
Glasses and Neural Bands: The Superintelligence Vehicle
Elsewhere at Meta’s Davos pavilion, the delivery vehicle got sharper definition. Derya Matras, Meta’s VP for Europe, the Middle East and Africa, argued that wearables hold an edge phones and computers cannot match: “They see what you see, they hear what you hear” — and make sense of your surroundings. “That’s how we get closer to superintelligence.” She pointed to Ray-Ban smart glasses and neural bands, with the stated goal of AI that knows users’ goals, interests, and dreams, framed as “giving that power to billions of people.”
The operational facts back the narrative. Nicola Mendelsohn, head of global business, said 3 billion people use Meta’s apps daily and that AI is “the backbone our recommendation systems.” Meanwhile Meta is marketing the Ray-Ban Display glasses and, earlier in January, paused their international expansion to prioritize fulfilling US orders. Wearables are not a side business in this story; they are the primary input surface for the superintelligence pitch.
The Route Question After LeCun
Former chief scientist Yann LeCun has called LLMs a “dead end” on the road to superintelligence, and his departure followed Meta’s hiring of Scale AI founder Alexandr Wang — an LLM believer. Asked about him in Davos, Matras replied: “Yann has a particular view and he is a brilliant scientist,” adding that Meta now has “perhaps the most talented team in the industry.”
The disagreement was never resolved; it was reorganized away. MSL’s bet is the LLM-centric scaling route. Six months to internal delivery proves execution speed, but there is still no public evidence about whether the route reaches the destination.
2026–2027: The Years Consumer AI Sets
Bosworth’s timing call: consumer AI patterns will firm up across 2026 and 2027, because current models already handle “the kinds of things that you ask every day with your family, your kids,” while development continues on harder queries. In practice, the next two years are Meta’s window to put its models into glasses and apps and lock in daily-use scenarios.
Worth pairing with the year’s opening signals: January’s 2026 opening outlook flagged reports that Meta’s flagship might go closed-weights. With first models now landing internally, both the schedule pressure and the route question get sharper at the same time.
What It Means for Developers and Product Teams
Three practical judgments. First, keep Meta’s frontier models off your near-term roadmap — when the CTO himself keeps saying “tremendous amount of work” on post-training, the external availability date is genuinely unknown. Second, wearables are a new interaction surface: once input comes from continuous glasses and wristband sensing, the default assumptions of product design — privacy, notifications, context management — get rewritten. Third, for competitors, a reorganized lab delivering internal models in six months resets expectations for how fast a frontier team can be stood up; the catch-up window is shorter than most planners assume.
Sources
- Meta’s new AI team delivered first key models internally this month, CTO says — Reuters
- Meta says wearables are central to AI superintelligence — The National
- Meta Superintelligence Lab delivers first AI models — The Jerusalem Post
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.