Meta has watched the bulk of its key artificial intelligence researchers depart in the wake of its influential Llama project, putting fresh pressure on its status as a leader in the fast-changing AI world.
Of the 14 researchers who authored the groundbreaking 2023 Llama paper introducing Meta’s open language model, only three remain on staff. Most of the experts involved have taken roles at direct competitors, with several now working at Mistral, a Paris AI startup launched by former Meta engineers Guillaume Lample and Timothée Lacroix.
The rate of these departures has unsettled observers and left Meta with a significantly reshuffled research base, just as open alternatives to Llama are gaining ground. Rivals like DeepSeek and Qwen have started to attract developer attention, particularly after the tepid reception of Meta’s latest model, Llama 4.
Leadership Changes Hit Meta’s AI Team
Further instability followed last month when Joelle Pineau, who spent eight years leading Meta’s Fundamental Artificial Intelligence Research group, stepped down, with Robert Fergus set to take her place. Fergus, a FAIR cofounder, brings experience from his years at Google’s DeepMind.
Even as the company highlights its open approach to advanced AI, many of those instrumental in making Llama possible have exited over the last two years. While Meta invested heavily to stand out with models made purely from public data, built for efficiency, it now scrambles to maintain its relevance in the open AI space.
The 2023 Llama release was seen as a pivotal moment, validating the case for freely accessible large models that let others experiment with, adapt, and expand them as alternatives to proprietary platforms from competitors like OpenAI and Google. For a short period, this put Meta at the center of developer discussions, but ongoing exits have gradually dulled that advantage.
Meta’s development pipeline has slowed, too, with reports that its largest project yet, Behemoth, is delayed amid performance concerns. Meanwhile, the company has yet to deliver an AI model designed specifically for advanced reasoning and problem solving, a gap as rivals roll out products pushing these boundaries.
The former Llama researchers had spent an average of more than five years with Meta, highlighting their deep ties to the firm’s AI ambitions rather than a pattern of short stays. Their departures date back to early 2023, but have continued through this year, weakening Meta’s original core of talent just as it faces heightened internal and external pressure to deliver.
With original architects gone and the pace of rival releases accelerating, Meta is tasked with defending its openness-inspired legacy without the team that launched its most notable AI achievement. For more on this, see the Meta AI research exodus and read about the Llama model talent loss.
