• by Laurie Sullivan , Yesterday

As Meta Platforms prepares to release its first new AI models developed under Scale AI CEO and founder Alexandr Wang, it could initially offer a few versions with an open-source license.

Meta has been the largest U.S. company to allow others to modify its frontier models -- the most advanced general-purpose models. But there has been growing speculation that the company could reverse that strategy.

Wang drives AI decisions at Meta, which invested in Scale AI in June 2025. The company acquired a 49% non-voting stake for more than $14 billion in a deal that valued the company at $29 billion. Wang joined to head a new team focused on artificial general intelligence.

The New York Times in March reported delays in releasing the new foundational AI model, code-named “Avocado,” which the company had originally planned to release in March 2026, citing three people with knowledge of the change.

“Avocado” reportedly did not perform as well as offerings from Google and others on coding, reasoning, and writing. The NYT even suggested Meta would temporarily license Gemini to power its AI products.

Avocado represents a significant shift in Meta's approach, moving from the broad accessibility of the open-source Llama series toward a more powerful, potentially proprietary "frontier" model — still with aspirations of being open-sourced.

While Meta intends to release open-source versions of its next models, it will initially release closed models to manage safety risks and protect specifications.

Meta’s models developed under Wang will prioritize consumer-facing features like shopping tools and close the performance gap with industry rivals.

Axios reported that Meta’s first family of models is designed to help it catch up to rivals after its last Llama 4 family fell significantly behind.

“Meta knows its new models may not be competitive across the board with the coming ones from those labs, but believes it will have areas of strength that appeal to consumers,” reported Axios, citing sources.

Wang founded Scale at 19 after dropping out of MIT, and now serves as chief AI officer at Meta. His vision centers on personal superintelligence -- AI that knows a user’s goals and interests -- and aligns with the idea that collecting data for specific goals like planning an event or ad campaign is the future, rather than gathering data just to collect it in case someone may need it in the future.