A benchmark for coding agents in evolving workspaces, pairing unchanged and changed continuations from the same saved state across 120 scenarios, 12 change families, and nine intervention positions.面向持续变化工作环境的 Coding Agent 评测。从同一保存状态出发，对比保持原样与引入变化后的执行，覆盖 120 个场景、12 类变化及 9 个干预位置。
An open autoregressive model for image generation, editing, and visual prediction.面向图像生成、编辑与视觉预测的开放自回归模型。
A curated map of datasets, methods, and evaluation resources for vision-to-music generation.面向 Vision-to-Music Generation 的数据集、方法与评测资源索引。
A framework for evaluating attacks, defenses, and safety risks across large language models.面向大语言模型攻击、防御与安全风险评测的一体化框架。