Meta is preparing to debut its first AI models under Alexandr Wang, in a pivotal moment after its $15 billion Scale AI deal.
This release is a major test for Wang, who was brought in to lead the Superintelligence group and fix the momentum issues Meta faced after its previous Llama 4 family failed to keep up with the competition. The outcome could determine whether Meta can reestablish itself as a serious contender in the rapidly evolving AI race.
While Meta has built a reputation as the most open among the major US AI players, this new rollout comes with a twist.
The company isn’t giving everything away for free immediately. Instead, they are adopting a hybrid strategy; some parts will be open for developers to tinker with, while the most powerful “frontier” components will stay behind closed doors. Meta’s goal is to ensure these tools don’t create new safety risks before they are shared with the public.
Axios reports that “Wang has indicated that some of its largest new models will remain proprietary — a shift toward a more hybrid strategy, according to sources.”
Meet ‘Avocado’ and ‘Mango’
Internally, these new projects have been given some fruity codenames. Reports suggest Meta is working on a high-end large language model known as “Avocado,” along with a multimedia generator called “Mango.”
While rivals like OpenAI and Anthropic are focusing heavily on selling their tech to big corporations and governments, Meta is doubling down on the everyday user. The company believes it has an advantage because it can bake these AI tools directly into apps that billions of people already use, like WhatsApp, Instagram, and Facebook.
Per Axios, “Meta wants its models distributed as widely and as broadly as possible around the world.”
The pressure to catch up
Meta is feeling the heat. Industry leaders are moving fast, and even companies that used to be open are starting to lock their doors. For example, the Chinese giant Alibaba recently kept its most powerful Qwen models private, a move that mirrors Meta’s new cautiousness.
The stakes are particularly high for Wang. After Meta spent billions on AI development and faced internal shakeups, including the departure of longtime AI head Yann LeCun last year, these new models need to prove that Meta can still compete at the top level.
While Meta admits its new models might not beat the competition in every category, it is betting that being “open enough” will be enough to win over developers and consumers alike.
Also read: China’s intervention in the Manus founders’ Meta AI deal is a reminder that AI competition is increasingly being shaped by geopolitics, not just technology.