EdgeBench：衡量真实环境学习并发现新扩展定律
原标题：EdgeBench: Measuring Real-World Environment Learning and Discovering a New Scaling Law
AI 摘要
字节跳动Seed Research发布了EdgeBench，一个超长时程基准，用于衡量智能体在真实环境中的持续学习能力，包含134个任务，每个任务支持至少12小时交互。研究发现，智能体的环境学习性能遵循log-sigmoid曲线，拟合优度R²=0.998，且前沿模型的学习速度约每三个月翻倍。EdgeBench已开源51个任务和完整评估框架，以促进社区研究。
正文节选
EdgeBench: Measuring Real-World Environment Learning and Discovering a New Scaling Law EdgeBench: Measuring Real-World Environment Learning and Discovering a New Scaling Law Date 2026-07-07 Category Frontier research Over the past few years, pretraining scaling laws have led to a broad consensus: model capabilities improve in a relatively predictable way as data and compute scale. But once large models enter real-world settings, a more practical question comes into focus: can they continue learn