Overview
The MindGames Challenge at NeurIPS 2025 invites participants to engage in a groundbreaking competition focused on enhancing theory-of-mind capabilities in large language model (LLM) agents. This event aims to explore critical aspects of AI social intelligence, including belief modeling, deception detection, and strategic cooperation, through dynamic gameplay.
Background & Relevance
As advancements in natural language processing (NLP) continue to reshape the AI landscape, understanding how LLMs can effectively model beliefs and navigate social interactions remains a pivotal area of research. The MindGames Challenge provides a unique platform for researchers and developers to test their models in real-time scenarios, fostering innovation in the field of AI social intelligence. By participating, teams can contribute to the broader understanding of how AI can engage in complex social dynamics, which is essential for developing more sophisticated and capable AI systems.
Key Details
- Competition Website: MindGames Arena
- Registration Link: Register Here
- Competition Timeline:
- Arena Open: July 7 – October 7, 2025
- Final Submissions: October 7, 2025
- Winners Announced: October 15, 2025
- NeurIPS Presentation: December 2025
- Games:
- Mafia
- Three Player IPD
- Colonel Blotto
- Codenames
Eligibility & Participation
The MindGames Challenge is open to all researchers and developers interested in advancing the capabilities of LLM agents. Participants can enter either the Open Division, which has no size restrictions, or the Efficient Agent Division, which limits models to a maximum of 8 billion parameters. This structure ensures a fair competition environment for teams with varying resources.
Submission or Application Guidelines
To participate in the MindGames Challenge, follow these steps:
1. Build your LLM agent using any approach you prefer.
2. Deploy your agent to the provided game framework using the starter kit.
3. Compete against other teams in automated matches.
4. Climb the leaderboard through strategic gameplay.
5. Iterate and improve your agent based on match results.
More Information
The MindGames Challenge not only offers a competitive arena for testing LLMs but also serves as a valuable opportunity for participants to learn from diverse strategies employed by other teams. Engaging in this competition can significantly enhance understanding of AI’s social intelligence capabilities and foster networking with leading researchers in the multi-agent AI domain.
Conclusion
The MindGames Challenge at NeurIPS 2025 presents a unique opportunity for researchers and developers to showcase their LLM agents in a competitive setting. By participating, teams can contribute to the advancement of AI social intelligence and gain insights into the capabilities of their models. Interested individuals are encouraged to explore this exciting opportunity and register for the competition today.
Category: Conferences & Workshops
Tags: llm, theory-of-mind, neurips, multi-agent ai, game theory, social intelligence, competition, natural language processing, strategic coordination, deception detection, belief modeling, ut austin, princeton
