FCMBench-V1.0

**FCMBench** is a multimodal benchmark for credit-risk–oriented workflows. It aims to provide a standard playground to promote collaborative development between academia and industry and provides standardized datasets, prompts, and evaluation scripts across multiple tracks (image, video, speech, agents, etc.) **FCMBench-V1.0** focuses on financial document understanding. #### Paper & Project Links - [**Paper (arXiv)**](https://arxiv.org/abs/2601.00150) - [**Paper (PDF)**](https://arxiv.org/pdf/2601.00150) - [**Project Page**](https://github.com/QFIN-tech/FCMBench/tree/main/vision_language) - [**Leaderboard**](https://github.com/QFIN-tech/FCMBench/blob/main/vision_language/LEADERBOARD.md) - [**Dataset (ModelScope)**](https://modelscope.cn/datasets/QFIN/FCMBench-V1.0) - [**Dataset (Hugging Face)**](https://huggingface.co/datasets/QFIN/FCMBench-V1.0) #### DOWNLOAD METHOD :modelscope-code[]{type="sdk"} :modelscope-code[]{type="git"}**FCMBench** 是一款面向信贷风险工作流的多模态基准测试集（multimodal benchmark）。其旨在打造标准化实验平台，以推动学术界与产业界的协同研发，并为图像、视频、语音、AI智能体（AI Agent）等多个赛道提供标准化数据集、提示词（prompt）与评估脚本。 **FCMBench-V1.0** 聚焦于金融文档理解任务。 #### 论文与项目链接 - [**论文（arXiv）**](https://arxiv.org/abs/2601.00150) - [**论文（PDF版）**](https://arxiv.org/pdf/2601.00150) - [**项目主页**](https://github.com/QFIN-tech/FCMBench/tree/main/vision_language) - [**排行榜**](https://github.com/QFIN-tech/FCMBench/blob/main/vision_language/LEADERBOARD.md) - [**数据集（ModelScope）**](https://modelscope.cn/datasets/QFIN/FCMBench-V1.0) - [**数据集（Hugging Face）**](https://huggingface.co/datasets/QFIN/FCMBench-V1.0) #### 下载方式 :modelscope-code[]{type="sdk"} :modelscope-code[]{type="git"}

搜集汇总

背景与挑战

背景概述

该数据集名为FCMBench-V1.0，采用Apache License 2.0许可证，于2026年3月16日更新，但当前内容尚未完善。