C2C: Cooperate to Compete — AI vs AI Games This dataset contains 972 fully logged AI vs AI games from the Cooperate to Compete (C2C) benchmark — a long horizon, mixed motive multi agent negotiation environment based on a four player conquest game with private regional objectives, fog of war, and non binding cheap talk negotiation. Project page: Paper: Play against AI agents: Github: Note: This is a raw structured log dataset (nested JSON files grouped by game, not a tabular Parquet/CSV dataset). The HuggingFace dataset viewer is disabled because the per turn game state JSON files contain heterogeneous nested schemas that cannot be unified into a single Arrow table. To use the dataset, download the files directly (see "Loading" below). Dataset Structure The dataset is organized into 6 intervention conditions, each containing 162 matched games played on the same starting board positions: Folder Description baseline/ Standard prompt — no intervention. deceit strategy/ One target agent is prompted to deceive opponents. aggressive ask strategy/ One target agent is prompted to negotiate more aggressively. support strategy/ One target agent is prompted to ask explicitly for support. one p…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy