Meta Set to Launch New AI Models Developed by Alexandr Wang

Mark Zuckerberg’s pursuit of artificial intelligence dominance may be approaching a pivotal moment, with Axios reporting that Meta’s Superintelligence division is nearing the release of its first internally developed AI models.

Meta officially launched its Superintelligence initiative in June, with Zuckerberg outlining an ambitious vision to “bring personal superintelligence to everyone,” framing it as a catalyst for accelerating human progress and unlocking a new era of individual empowerment.

To assemble the team, Meta reportedly extended multi-million dollar offers to leading figures in the AI space, including Alexandr Wang. Wang now heads the division after joining Meta from Scale AI in 2025, following Meta’s $15 billion acquisition of the firm, as reported by Business Insider.

With that groundwork laid, attention is now shifting to output.

According to Axios: “Meta is preparing to release the first new AI models developed under Alexandr Wang, with plans to eventually offer versions of those models via an open source license.”

These models are expected to serve as a major demonstration of Meta’s expanding ambitions in artificial intelligence, particularly as the company pours hundreds of billions of dollars into reshaping its business around AI as a foundational technology.

They also carry strategic importance. Meta is widely seen as needing to close the gap with competitors after its most recent Llama model underperformed against rival systems in key AI benchmarks. As such, the upcoming release represents more than just progress—it’s a test of credibility.

At the same time, Meta’s longer-term goal extends beyond current-generation AI tools toward systems capable of simulating human-like reasoning and cognition.

That ambition, however, remains highly contested.

Critics argue that today’s AI models—despite their fluency—do not possess true intelligence. Instead, they function as advanced pattern-matching systems, drawing from vast datasets to generate responses that appear human-like but lack genuine understanding or awareness.

From this perspective, the current trajectory may not lead to artificial general intelligence (AGI), often described as the ultimate objective of AI research.

Even within Meta, there has been internal debate. Yann LeCun, the company’s long-time head of AI research, departed last year amid disagreements over the direction of development.

LeCun, widely regarded as one of the field’s most influential voices, has argued that generative AI models may be a distraction rather than a breakthrough path.

“A lot of people in the AI research and development community are perceiving the idea that perhaps we have a shot within the next decade or so of building machines that might eventually reach human-level intelligence,” LeCun said at the Joint Mathematics Meetings conference in March 2025. “Some people claim it’s imminent—but I don’t believe that.”

He suggested that a combination of open-source collaboration and new model architectures could potentially lead to systems capable of reasoning and planning within the next five to ten years—but not through current approaches alone.

Meta, for its part, has been exploring alternative pathways. Among them is its “Brain Decoding” research, first previewed in 2023, which seeks to better understand neural activity and replicate aspects of human cognition.

Such work may ultimately prove more aligned with the goal of building systems that genuinely think, rather than simply simulate conversation through large-scale data correlations.

Where Meta’s Superintelligence initiative fits within this broader landscape remains an open question—but its next release could offer an early signal of how the company intends to bridge the gap between today’s AI and the more ambitious vision of tomorrow.