Kaggle's Game Arena: LLMs Play to Win

🔥 What happened Kaggle launched Game Arena, a platform where LLMs battle head-to-head in Chess, Poker, and Werewolf. 62 authors, three pilot environments, one mission: kill static benchmarks. 💡 Why it matters Games scale with model capability – no saturation like MMLU. Perfect info, imperfect info, multiplayer: three distinct stress tests for planning, adaptation, and robustness under uncertainty. ⚡ Our take Finally a benchmark that won't be gamed out in two years. If your model can bluff at Werewolf, it can run agents. If it only does multiple choice, it's toast.