MindGames : A Live Arena for Evaluating Social and Strategic Reasoning in Multi-Agent LLMs

MindGames Organizer & Participation Teams, NeurIPS 2025 Competition · 2026

Read MindGames : A Live Arena for Evaluating Social and Strategic Reasoning in Multi-Agent LLMs by MindGames Organizer & Participation Teams, NeurIPS 2025 Competition (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-29512.txt.

Introduction

Kevin Wang and colleagues present Mindgames, an arena for social and strategic thought in multi-agent language models. Static scenes and single-game tests miss long-run thought in multi-agent settings. Mindgames stages four games with an interface, skill ratings, and play logs. A 2025 cycle drew 944 agents from 76 teams across Colonel Blotto, iterated prisoner's dilemma, Codenames, and Secret Mafia. Analysis finds brittle rule following a bottleneck, with top systems relying on scaffolding. Secret Mafia shows a failure confound, with tens of thousands of games shared plus offline contest rules.

Catalog record

CREATORMindGames Organizer & Participation Teams, NeurIPS 2025 Competition
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE141552
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.29512

Browse the full library · About the Tower of Babel