GameHorizon Suite: Multi-Horizon Data and Evaluation in Gameplay

AI 导读

GameHorizon Suite通过多时间尺度数据和评估框架，解决现有游戏AI数据集的局限性，为不同模型家族提供标准化评估工具。

GameHorizon Suite addresses limitations in existing game AI datasets by offering multi-horizon data and evaluation frameworks for diverse models.

重点速览

• GameHorizon Suite解决现有数据集覆盖窄、缺乏指令和高方差在线回滚的问题 Addresses narrow coverage, lack of instructions, and high-variance online rollouts in existing datasets
• 包含注释器、大规模游戏数据集和可复现的离线/在线基准测试三个核心组件 Comprises annotator, large-scale game dataset, and reproducible offline/online benchmarking components
• 提供5000小时专业玩家录制数据和跨任务诊断评估体系 Provides 5,000-hour expert player recordings and cross-task diagnostic evaluation system
• 通过百万级模型调用揭示任务难度层级和模型能力差异 Reveals task difficulty hierarchy and model capability differences via million-level invocations
• 建立首个跨时间尺度的游戏能力标准化评估标准 Establishes first standardized cross-horizon game capability evaluation framework

一句话：GameHorizon Suite为AI游戏能力评估提供标准化工具。 / GameHorizon Suite provides standardized tools for AI game capability evaluation.

推荐理由 填补游戏AI评估空白，推动多模型对比研究。