Otter: A Time-Aware, History-Conditioned Human Chess AI

Otter is a human chess AI that models move prediction as a time-aware, sequential process rather than treating each position independently. It combines two conditioning signals: a move history encoder that conditions on the last 20 moves to capture opening preferences and behavioral drift, and a time control module that accounts for clock pressure. Trained on 6.1 billion positions from 117 million Lichess rapid games over 30 days on a single T4 GPU, Otter achieves 55.23% top-1 and 90.95% top-5 move-prediction accuracy, surpassing the prior state-of-the-art human chess model Maia 2 with significantly fewer parameters and less training data. Performance peaks at 57.38% accuracy in the 1900-1999 Elo bracket. The results demonstrate that modeling chess as a time-aware, sequential activity yields more human-accurate predictions than position-only approaches. Code, trained models, and training logs are publicly released.