Meta delays Llama 4 Behemoth as 11 of 14 researchers leave and internal performance disappoints

Summary

Meta is delaying Llama 4 Behemoth due to poor internal performance and significant leadership disruption. Eleven of the fourteen researchers on the Llama team have left the company. Former team members are publicly noting on LinkedIn that they worked on Llama 1, 2, and 3 but not 4, signaling a sharp break in continuity. The departures follow a major reorganization of Meta's AI leadership and come after Meta's earlier admission of benchmark gaming in its model development. The combination raises concerns about the rigor of Meta's model evaluation and the stability of its AI research organization as competitors advance in frontier model performance.