Meta wasn’t dead, but she wasn’t partying either. He was working hard on a new AI model for which there are huge expectations. Now we know a little more about that project, but one thing doesn’t change: it better not fail.
what has happened. Andrew Bossworth, CTO of Meta, has confirmed during the World Economic Forum in Davos that the Superintelligence Labs division already has a first internal version of its new AI models. This is an important and long-awaited milestone that they have been working on for six months.
“Very good”, but not ready yet. Bossworth did not want to give too many details, but he did indicate that preliminary tests show that the models perform very well. These models will still take time to come to market: Meta is currently in a critical post-training phase for these models to be truly useful for both internal developers and end users.
Two great models. Although the names of those models were not specified, rumors and leaks point to two major developments. On the one hand, Avocado, focused on text and which could be launched in the first quarter of 2026. On the other, Mango, focused on image and video generation.
A 2025 of transition. The manager described the year 2025 as “tremendously chaotic”, and it was certainly a very complicated year for Meta. He Llama 4 failure It made the company completely change its philosophy and strategy. Zuckerberg did not stop hire talent with a exorbitant costespecially when it acquired Scale AI and signed its CEO, Alexandr Wangnow head of Superintelligence Labs. That investment has also been allocated to acquire companies like Manusthat could become another key component of Meta’s strategy going forward.
Glasses as an AI device. If they behave as expected, these models will probably also end up being used in smart glasses from Meta, which has been collaborating with Ray-Ban for years and which you have just seen its second version accompanied, of course, by the striking and even more versatile Meta Ray-Ban Display. Interaction with AI models is one of the most striking features, and these models could take it to new limits.
The mystery of Open Source. In July 2024 Mark Zuckerberg indicated that “Open Source AI is the way to go.” Llama was at that time the clear reference, but the disappointing launch of Llama 4 and above all the push for open models from various Chinese companies has made this panorama change significantly. It is not clear that Meta will launch its new models with open versions, and if it did not do so, Chinese hegemony would be even more notable.
Will it be worth the investment? Meta is one of the companies that has spent (or bet) the most money on the future of AI. Mark Zuckerberg said that I was willing to lose “hundreds of billions of dollars in AI” because not investing them would be even more dangerous for Meta. He has been consistent with that statement, but It remains to be seen if it ends up working.. The company certainly has the resources to be a protagonist in this market, but today its solutions—with Meta AI at the head—have a very reduced role compared to that of their competitors.
Image | Goal
In Xataka | China’s best kept weapon in AI is not Qwen: it is the more than 100,000 variables created by other companies
GIPHY App Key not set. Please check settings