Kaggle Game Arena: Evaluating LLMs via Head-to-Head Chess, Poker, and Werewolf

kaggle · hf · 2026-09-28

Kaggle introduces Game Arena, an open platform that evaluates LLMs through competitive games instead of static benchmarks.

Models play head-to-head in structured environments; difficulty scales naturally as models improve, preventing benchmark saturation
Three pilot environments — Chess (perfect information), Poker (imperfect information), and Werewolf (multiplayer) — probing strategic planning, adaptation, and robustness under uncertainty
Each game ships with environment specs, metrics, and full cross-model competition results
Infrastructure emphasizes reproducibility, transparency, and generalization to new games via large-scale ground-truth evaluation

More from Models