Meta has internally delivered the first AI models from its new Superintelligence Labs. CTO Andrew Bosworth called the results promising during the World Economic Forum in Davos. The team has been working on the development for six months.
Bosworth emphasized during a press conference that the models are performing “very good,” although they are not yet ready for use. “They’re basically six months into the work, not quite even,” said the CTO. Meta is reportedly working on a text model codenamed Avocado and an image-and-video model called Mango.
Developments at Meta have been closely monitored since CEO Mark Zuckerberg launched a large-scale recruitment drive with packages of up to $100 million. The company has attracted more than 50 engineers and researchers, including four former OpenAI scientists.
Criticism of Llama 4 performance
Meta has been criticized for the performance of its Llama 4 model, while competitors have gained momentum in the AI race. The new models should help the company regain ground in the competitive AI market.
Bosworth acknowledged that there is still “a tremendous amount of work” to be done after training. “There’s a tremendous amount of work to do post-training” for AI, “to actually deliver the model in a way that’s usable internally and by consumers.” the CTO explained.
Still, Meta is seeing positive results from its investments. Bosworth called 2025 a “tremendously chaotic year” for building the lab and infrastructure and for acquiring energy.
According to Bosworth, 2026 and 2027 will solidify consumer AI trends. Recent advances have produced models that already answer “the kinds of things that you ask every day with your family, your kids.” For more complex questions, technological developments can further improve results. That’s why the next two years are important for bringing consumer products to market, he said.