Nathan Lambert departs Allen Institute for AI after leading Olmo and Tülu post-training projects

Nathan Lambert announced his departure from Ai2, where he worked on the Olmo models and post-training projects including Tülu 2, Tülu 3, and Olmo 3. He joined Ai2 in October 2023 after meeting Luca Soldaini at ICML 2023 in Hawaii.

Lambert led major post-training initiatives including RewardBench (reward model evaluation), Tülu 3 (which introduced the term Reinforcement Learning with Verifiable Rewards / RLVR), and Olmo 3. He states that Olmo 3 was originally targeted for release by June or July of 2025 but was completed later after training a bigger model.

Lambert plans to continue working in the open AI ecosystem to improve coordination and usefulness, while remaining based in Seattle for most of the year. He credits Ai2's scientific culture and collaborative environment as central to his work's impact and his personal growth.