A specialized reinforcement-learning model for competitive 1v1 Catan, designed to make strategic decisions under uncertainty and partial information.
VINCO is Distributed Cognition’s specialist model for competitive one-on-one CATAN. The publisher says it was trained without human games, using offline reinforcement learning on simulated games followed by self-play fine-tuning. In the publisher’s account, the approximately 0.5-million-parameter model reached the top of Colonist.io’s 1v1 leaderboard, and it also reports benchmark matches against several general-purpose models. The report describes play under partial information, randomized dice outcomes and a 15-point win condition.
