Zuckerberg has bet everything on artificial intelligence. Meta is investing hundreds of billions of dollars in AI infrastructure, and the moment is approaching when it will show what it got for that money. According to an exclusive report by Axios, Meta's superintelligence division is preparing to release the first AI models developed under the leadership of Alexandr Wang.
Wang did not join Meta by chance. Zuckerberg brought him on as part of the acquisition of Scale AI, for which Meta paid 15 billion dollars. Wang became the director of this project and now leads a team whose goal is to build artificial general intelligence (AGI). In other words, machines that truly think rather than merely predict the most likely words in a sentence.
Open, but not immediately or entirely
From the outset, Meta built a reputation as a company that makes its models freely available. Llama, Llama 2, Llama 3... developers grew accustomed to receiving models from Meta that they could work with freely. But now things are changing.
Although the new models will eventually be released as open source, this will not happen immediately or in full. Meta will initially keep part of the code to itself. The reasons? Two. It wants to ensure that the models do not introduce new security risks. It also wants to maintain a competitive advantage. That is understandable, but it represents a significant change for the developer community.
Wang reportedly expects the largest and most powerful models to remain closed permanently. Meta is therefore shifting toward a hybrid strategy: some models open, others closed.
Why Wang sees Meta differently from OpenAI or Anthropic
Wang has a clear vision. According to him, OpenAI and Anthropic are increasingly focusing on large companies and government contracts. Their models are powerful but expensive and closed. Wang wants to take a different path: make AI available to as many people around the world as possible, either for free or at minimal cost.
In this respect, Meta has an enormous advantage that others can only dream of. WhatsApp, Facebook, and Instagram are used by billions of people every day. If Meta integrates its AI tools directly into these platforms, it will instantly reach an audience that no other technology company can easily replicate. And at no cost.
While Anthropic fine-tunes Claude for corporate customers and OpenAI competes for lucrative government contracts, Meta is betting on sheer accessibility. Its user base is its greatest weapon.
The new models bring high expectations, but also considerable pressure. Llama 4, the previous model series, lagged significantly behind the competition in AI benchmarks. For Meta, which presents itself as a leader in artificial intelligence, this was an uncomfortable finding.
And the competition is not standing still. Both OpenAI and Anthropic are preparing new models of their own that are being described as a major leap forward. Meta knows this. Its executives openly admit that the new models may not be the best in every respect, but they believe they will find areas where they excel specifically for everyday users. It is an approach that makes sense. Not everyone needs the most powerful model possible. Many people simply want a functional AI assistant in an app they already use every day.
Yann LeCun's departure from Meta
The current developments have another interesting dimension. Yann LeCun, Meta's longtime head of AI research and one of the field's most prominent figures, left the company last year. The reason? A fundamental disagreement over the direction AI research should take.
LeCun has long argued that current language models are a dead end. In his view, they are sophisticated tools for finding patterns in data, but they do not represent genuine intelligence. He believes a true breakthrough would require an entirely different approach inspired by how the human brain works.
Meta had been working in this area for some time. The "Brain Decoding" project, first introduced in 2023, focused specifically on simulating neural activity and understanding human thought. We do not yet know where this research stands today or whether Wang's new models reflect it in any way.
Meta is not the only company reconsidering how open it should be. Alibaba recently decided that its most powerful Qwen models would remain closed, even though the company had previously actively promoted an open approach. Elon Musk, meanwhile, has publicly attacked OpenAI for betraying its original promise of openness. The trend is therefore clear: the more powerful the model, the greater the temptation not to release it. Whether for security reasons, competitive advantage, or simply because too much money has been invested in it.
In this environment, Meta is trying to maintain its image as a company that supports developers. Wall Street analysts rate META stock as a strong buy, with an average price target of around 862 dollars per share, which would represent an increase of nearly fifty percent from its current price.
Wang knows what he wants. Give developers enough reason to stay. At the same time, keep the most valuable assets private so that Meta does not lose its lead over the competition.
Sources: tech.yahoo.com and economictimes.indiatimes.com
