Meta's Llama AI team has been bleeding talent. Many top researchers have joined French AI startup Mistral.

Meta is experiencing a significant loss of talent within its Llama AI team, as many of the leading researchers who contributed to the development of its open-source models have departed for new opportunities, particularly at the French startup Mistral. Of the 14 authors credited in the pivotal 2023 paper that introduced the Llama models, only three remain at Meta. The majority have moved on to join or establish competing firms, raising concerns about Meta's ability to retain top AI talent amid increasing internal and external challenges. Mistral, co-founded by former Meta researchers Guillaume Lample and Timothée Lacroix, is now at the forefront of developing powerful open-source AI models that directly rival Meta's offerings. This shift is particularly notable as Meta is grappling with delays in releasing its largest AI model, Behemoth, due to internal worries regarding its effectiveness and leadership. The reception of Llama 4, Meta's latest AI model, has been tepid among developers, many of whom are now exploring faster-moving alternatives like DeepSeek and Qwen. The shake-up at Meta has been exacerbated by the recent resignation of Joelle Pineau, who led the Fundamental AI Research group for eight years. She is set to be succeeded by Robert Fergus, a co-founder of the group who recently returned to Meta after a stint at Google’s DeepMind. As Meta's research team undergoes this transformation, the departure of its original architects raises critical questions about the company’s ability to maintain its lead in the competitive AI landscape. The 2023 Llama paper not only marked a significant technical achievement but also played a crucial role in legitimizing open-weight large language models, which offered an accessible alternative to proprietary systems. Despite substantial investments in AI, Meta has yet to develop a dedicated reasoning model capable of handling complex tasks requiring multi-step thinking. This gap is becoming increasingly evident as rivals like Google and OpenAI prioritize these capabilities in their latest models. The average duration of employment for the 11 researchers who left was over five years, indicating their deep involvement in Meta’s AI endeavors. The exodus of talent from Meta underscores a critical phase in the company’s AI journey, as it strives to defend its early advancements in an industry that is rapidly evolving. The future of Meta's AI strategy now hangs in the balance with many of its original innovators no longer in the picture.