Meta Loses Top AI Talent as Llama Team Members Move to Rivals

Meta is losing many of the researchers who helped build its powerful Llama AI models. Of the 14 authors behind the 2023 Llama paper, only three still work at Meta. The others have joined rival companies, with many now working at Mistral, a growing AI startup in Paris.

Mistral was co-founded by Guillaume Lample and Timothée Lacroix, two key figures in the original Llama project. Other former Meta employees have also joined top AI firms like Anthropic, DeepMind, Cohere, and Microsoft AI.

This wave of departures raises questions about Meta’s ability to keep its best AI talent. Despite spending billions on AI, Meta is falling behind newer open-source competitors such as Qwen and DeepSeek.

The Llama models were once seen as a major success. Meta shared them openly with the public, making it easier for researchers to build advanced AI tools. But now, the team that made it all happen is mostly gone.

Inside Meta, there has also been a leadership change. Joelle Pineau, who led the FAIR (Fundamental AI Research) group for eight years, has stepped down. She has been replaced by Robert Fergus, who co-founded FAIR and recently returned from DeepMind.

Meta is also facing delays in releasing Behemoth, its biggest AI model. Reports say the model has performance issues and leadership concerns. Meanwhile, the reception to Llama 4 has been mild, and developers are turning to faster-growing rivals.

Some former Meta researchers had worked at the company for more than five years. Their quiet exit marks a major shift for Meta’s AI division, which once hoped to lead the open-source movement.