📄 中文摘要
大型语言模型（LLMs）在需要战略决策的互动环境中越来越多地被部署，但对这些能力的系统评估仍然面临挑战。现有的 LLM 基准主要通过孤立任务评估静态推理，未能捕捉动态战略能力。近期的基于游戏的评估采用 LLM 对 LLM 的比赛，产生依赖于瞬时模型池的相对排名，导致计算成本呈二次增长，并缺乏稳定的性能锚点以进行长期跟踪。核心挑战在于建立一个可扩展的评估框架，该框架能够根据一致且可解释的标准测量 LLM 的战略推理，而不是依赖于波动的同行模型。
BotzoneBench：通过分级 AI 锚点实现可扩展的 LLM 评估
出处: BotzoneBench: Scalable LLM Evaluation via Graded AI Anchors
发布: 2026年2月17日
📄 English Summary
BotzoneBench: Scalable LLM Evaluation via Graded AI Anchors
Large Language Models (LLMs) are increasingly deployed in interactive environments that require strategic decision-making, yet systematic evaluation of these capabilities remains a challenge. Existing benchmarks primarily assess static reasoning through isolated tasks and fail to capture dynamic strategic abilities. Recent game-based evaluations utilize LLM-vs-LLM tournaments, producing relative rankings that depend on transient model pools, resulting in quadratic computational costs and lacking stable performance anchors for longitudinal tracking. The central challenge lies in establishing a scalable evaluation framework that measures LLM strategic reasoning against consistent and interpretable standards rather than volatile peer models.