MixEval/MixEval-X

MixEval-X是一个首个any-to-any真实世界模态多样化基准数据集，包含八种不同的输入输出模态组合，具有真实世界的任务分布、一致的高标准跨模态评估和动态性。它旨在为多模态模型提供全面、公平、高效的评估，并能够适应和随时间增长。

MixEval-X is the first any-to-any real-world multimodal benchmark featuring diverse input-output modality combinations, real-world task distributions, consistent high standards across modalities, and dynamism. It is designed to provide comprehensive, fair, and efficient evaluations for multimodal models and can adapt and grow over time.