Meta prepares first AI models from Alexandr Wang, with plans for open-source versions

The move

It has been reported that Meta is preparing to release the first family of AI models developed under Alexandr Wang, and that the company plans to eventually offer versions of those models under an open source license. Short and sweet: Meta wants to show up to the party. Wang, it has been reported, joined Meta last year as part of a roughly $15 billion deal with Scale AI — a splashy hire meant to shake things up.

What’s different

The new models are described as an attempt to close the gap after Llama 4, which allegedly fell behind competitors, and to build future versions that could lead the industry. Don’t expect a full throwback to earlier days of radical openness. It has been reported that some of the largest new models will remain proprietary — a hybrid approach, open in parts and closed in others. Hedge or strategy? You decide.

Why it matters

Meta has been the largest U.S. player to let outside developers tinker with its frontier models; a retreat would have signaled a shift in the open AI ecosystem. This plan looks like a middle path: open enough to win developer mindshare and shape standards, closed enough to protect whatever it thinks gives it an edge. Meanwhile, rivals aren’t standing still — OpenAI and Anthropic are teasing major model advances of their own. Competition, meet compromise.

A lot is at stake here: democratizing access versus safeguarding IP. Will this gamble buy Meta influence among developers and policymakers? Or will the hybrid stance leave everyone scratching their heads? Either way, the industry’s tug-of-war over openness just got louder.

Sources: axios.com