Google DeepMind has expanded its Kaggle Game Arena, a public benchmarking platform for AI models, by introducing two new game benchmarks: Werewolf and Poker. These additions complement the existing chess benchmark, which assesses strategic reasoning and long-term planning. While chess operates on perfect information, Werewolf and Poker challenge AI models to navigate imperfect information, crucial for real-world applications. Models like Gemini 3 Pro and Gemini 3 Flash currently lead in chess, showcasing advanced strategic reasoning based on classic chess principles. Werewolf tests the ability to communicate and identify deception within a team-based, natural language environment. It evaluates "soft skills" necessary for AI assistants to work alongside humans. This benchmark also provides insights into model behavior regarding manipulation and deception. Poker, on the other hand, introduces risk management, where models must gauge uncertainty and infer opponents' strategies in a game of incomplete information. A new poker tournament will culminate in a leaderboard reveal on February 4. The launch of these benchmarks is accompanied by livestreamed events featuring prominent players like Chess Grandmaster Hikaru Nakamura and poker experts. The events will showcase the capabilities of these AI models across all three games, providing a platform for direct competition and analysis. Kaggle Game Arena aims to reveal the potential of AI through rigorous testing in diverse scenarios, pushing the boundaries of what these models can achieve.