README.md

---
license: apache-2.0
task_categories:
- text-generation
- question-answering
language:
- en
- zh
size_categories:
- n<1K
tags:
- deep-research
- benchmark
- evaluation
- research-agent
- long-form-generation
- RACE
pretty_name: ICBCBench Dataset
configs:
  - config_name: objective_questions
    data_files:
      - data/objective_questions_public_80.json

  - config_name: subjective_questions
    data_files:
      - data/subjective_questions_public_40.json
---

<h1 align="center">ICBCBench: An Industry Consortium Benchmark for Financial Deep Research</h1>

<div align="center">
<a href="https://arxiv.org/abs/2606.17458" target="_blank"><img src=https://img.shields.io/badge/arXiv-b5212f.svg?logo=arxiv></a>
<a href="https://github.com/DeepFin-Intelligence/ICBCBench"><img alt="Github" src="https://img.shields.io/badge/GitHub-181717?logo=github&logoColor=white"></a>
<a href="https://huggingface.co/spaces/DeepFin-Intelligence/ICBCBench-Leaderboard"><img alt="Leaderboard" src="https://img.shields.io/badge/🏆%20Leaderboard-yellow?color=FFD700"></a>
</div>

## Overview

ICBCBench is an industry consortium benchmark for evaluating financial Deep Research Agents in real-world research scenarios. It consists of bilingual objective and subjective tasks across major financial sectors, including capital markets, banking, insurance, and related financial services. Developed with over 50 contributors from more than 40 financial and academic organizations, ICBCBench combines verifiable question answering with expert-aligned long-form report evaluation to assess retrieval–reasoning accuracy, factual grounding, citation reliability, and end-to-end report quality.

 <p align="center">
  <img src="https://raw.githubusercontent.com/DeepFin-Intelligence/ICBCBench/main/images/overview.png" alt="ICBCBench Overview" width="90%"/>
</p>

## Dataset

ICBCBench dataset consists of 120 questions with clear and unambiguous answers, covering both Chinese and English. It includes 40 subjective questions and 80 objective questions. The questions are primarily text-based, with a small number of multimodal examples. The dataset is divided into a public validation set and a test set with private answers and metadata.

## Directory Structure

```
data/
|-- objective_questions_public_80.json              # 80 objective questions
|-- subjective_questions_public_40.json             # 40 subjective report questions
```

## Contact

For questions or feedback about ICBCBench, please contact:

- **Weiya Li**: [weiyali126@outlook.com](mailto:weiyali126@outlook.com)
- **Zhiwei Tang**: [zwtang1220@gmail.com](mailto:zwtang1220@gmail.com)
- **Yizhou He**: [jonah_he@163.com](mailto:jonah_he@163.com)
- **Li Guo**: [guo_li@fudan.edu.cn](mailto:guo_li@fudan.edu.cn)
- **Linfeng Zhang**: [zhanglinfeng@sjtu.edu.cn](mailto:zhanglinfeng@sjtu.edu.cn)

## Citation

```bibtex
@misc{li2026icbcbenchindustryconsortiumbenchmark,
      title={ICBCBench: An Industry Consortium Benchmark for Financial Deep Research}, 
      author={Weiya Li and Zhiwei Tang and Yizhou He and Chenghao Wang and Liang Feng and Xiao Sun and Dongrui Liu and Zichen Wen and Hu Wei and Jinghang Wang and Yi Luo and Li Guo and Linfeng Zhang},
      year={2026},
      eprint={2606.17458},
      archivePrefix={arXiv},
      primaryClass={cs.CE},
      url={https://arxiv.org/abs/2606.17458}, 
}
```