- 2026.08.25 🎉 "FinGuard: Detecting Financial Regulatory Non-Compliance in LLM Interactions" has been accepted to Findings of EMNLP 2026!
- 2026.08.17 🎉 DianJin-RED is released — an action-grounded red-teaming benchmark for complete agent systems, with 1,661 executable cases, 15 intervention strategies (user input, agent-platform state, and external tools/data), and isolated service worlds.
- 2026.05.20 🎉 DianJin-SKILLS (百技图) is open-sourced — an AI Agent skill library for finance covering banking, insurance, and securities/asset management: 10 professional roles, 130+ standardized skills, ready to plug into Agent frameworks.
- 2026.05.10 🎉 "Fin-PRM: A Domain-Specialized Process Reward Model for Financial Reasoning in Large Language Models" has been accepted by IJCAI 2026!
- 2026.04.07 🎉 "Benchmarking Large Vision-Language Models on CFMME: A Comprehensive Chinese Financial Multimodal Evaluation Dataset" has been accepted by ACL 2026 main conference!
- 2026.02.27 🎉 Our papers FinMCP-Bench and CARE have been accepted by ICASSP 2026!
- 2025.11.15 "Evaluating, Synthesizing, and Enhancing for Customer Support Conversation" has been officially accepted by AAAI 2026!
- 2025.10.11 "FinMCP-Bench: Benchmarking LLM Agents for Real-World Financial Tool Use under the Model Context Protocol" jointly released by Yingmi Fund and partners — the first benchmark dataset and evaluation framework for real-world financial tool use by LLM agents, built on MCP.
- 2025.10.11 "CARE: Cognitive-reasoning Augmented Reinforcement for Emotional Support Conversation" is now published!
- 2025.08.25 "Fin-PRM: A Domain-Specialized Process Reward Model for Financial Reasoning in Large Language Models" is now published and open-sourced!
- 2025.08.18 "DianJin-OCR-R1: Enhancing OCR Capabilities via a Reasoning-and-Tool Interleaved Vision-Language Model" is now published and open-sourced!
- 2025.08.08 "Evaluating, Synthesizing, and Enhancing for Customer Support Conversation" is now published and open-sourced!
Earlier news
- 2025.05.22 "M3FinMeeting: A Multilingual, Multi-Sector, and Multi-Task Financial Meeting Understanding Evaluation Dataset" has been officially accepted by ACL 2025!
- 2025.04.23 DianJin-R1 series open-sourced! Includes the DianJin-R1-Data dataset and two powerful models, DianJin-R1-7B and DianJin-R1-13B. See our technical report "DianJin-R1: Evaluating and Enhancing Financial Reasoning in Large Language Models" for details.
- 2025.01.06 The CFLUE dataset has been fully open-sourced and is now available for download! 🚀🚀🚀
- 2024.05.16 "Benchmarking Large Language Models on CFLUE - A Chinese Financial Language Understanding Evaluation Dataset" has been officially accepted by ACL 2024! 🚀🚀🚀
This repository serves as the open-source hub for our financial AI research. The released code, models, and data are summarized below:
| Project | Code | ModelScope | HuggingFace | Paper |
|---|---|---|---|---|
| DianJin-SKILLS (百技图) |
DianJin-SKILLS | — | — | — |
| DianJin-RED | DianJin-RED | — | — | — |
| Fin-PRM | DianJin-PRM | Fin-PRM | Fin-PRM | IJCAI 2026 |
| FinMCP-Bench | DianJin-TIR | — | — | ICASSP 2026 |
| CARE | — | — | — | ICASSP 2026 |
| DianJin-OCR-R1 | DianJin-OCR-R1 | DianJin-OCR-R1 | DianJin-OCR-R1 | arXiv 2508.13238 |
| CSC | DianJin-CSC | CSC | CSC | AAAI 2026 |
| M3FinMeeting | — | Application Required | ACL 2025 | |
| DianJin-R1 | DianJin-R1 | DianJin-R1-32B | DianJin-R1-32B | arXiv 2504.15716 |
| DianJin-R1-7B | DianJin-R1-7B | |||
| DianJin-R1-Data | DianJin-R1-Data | |||
| CFLUE | cflue | CFLUE | CFLUE | ACL 2024 |
Welcome to Qwen DianJin 👋
We are the Tongyi DianJin team at Alibaba Cloud's Financial Services Industry, focused on the exploration, research, and deployment of large models in finance.
Our work follows a closed-loop methodology — Evaluation → Data Synthesis → Post-Training — where each cycle produces reusable financial capabilities that feed the next:
- Evaluation sets quantifiable capability targets via industry-grade benchmarks.
- Data synthesis turns the chronic "data scarcity" problem in finance into "data producibility" through controllable, scalable trajectory data.
- Post-training (SFT, RL) converts that data into stable, transferable task capabilities, which feed back into evaluation and data generation.
- Financial benchmarks & evaluation — a three-tier capability framework spanning foundational (CFLUE), scenario (M3FinMeeting), and agentic (FinMCP-Bench), with FinGDPVal in co-construction.
- Data synthesis — agent-based and workflow-based trajectory distillation, MCP environment interaction, and multi-agent self-play; powering datasets such as DianJin-R1-Data and the CSC corpus.
- Post-training — DianJin-R1 reasoning models, and Fin-PRM, the first financial-domain process reward model.
- Intelligent dialogue — customer-support (CSC) and emotional-support (CARE) conversation with explicit cognition–emotion–strategy reasoning chains.
- Agentic — DianJin-SKILLS (百技图) skill library and DianJin-TIR autonomous planning & tool use.
- Multimodal — reasoning-and-tool interleaved VLMs for financial document understanding (DianJin-OCR-R1).
2026 focus: Agentic RL and meta-harness evolution.
This repository is the open-source home for those efforts.
Our research is productized in the Qwen DianJin Platform — Alibaba Cloud's AI workspace for financial institutions. Visit the platform site for product details and access.
We are also co-building FinGDPVal with industry partners — an evaluation benchmark for whether AI workspaces can end-to-end complete real financial job tasks. The gold-standard evaluation set and toolchain will be open-sourced here.
If you find our work helpful, feel free to cite us.
@inproceedings{finguard,
title = {FinGuard: Detecting Financial Regulatory Non-Compliance in LLM Interactions},
author = {Huaixia Dou and Jie Zhu and Minghao Wu and Shuo Jiang and Junhui Li and Lifan Guo and Feng Chen and Chi Zhang},
booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2026},
year = {2026}
}
@inproceedings{csconv,
title = {Evaluating, Synthesizing, and Enhancing for Customer Support Conversation},
author = {Jie Zhu and Huaixia Dou and Junhui Li and Lifan Guo and Feng Chen and Chi Zhang and Fang Kong},
booktitle = {Proceedings of AAAI},
year = {2026}
}
@inproceedings{fin-prm,
title = {Fin-PRM: A Domain-Specialized Process Reward Model for Financial Reasoning in Large Language Models},
author = {Jie Zhu and Yuanchen Zhou and Shuo Jiang and Junhui Li and Lifan Guo and Feng Chen and Chi Zhang},
booktitle = {Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI)},
year = {2026}
}
@inproceedings{finmcp-bench,
title = {FinMCP-Bench: Benchmarking LLM Agents for Real-World Financial Tool Use under the Model Context Protocol},
author = {Jie Zhu and Yimin Tian and Boyang Li and Kehao Wu and Zhongzhi Liang and Junhui Li and Xianyin Zhang and Lifan Guo and Feng Chen and Yong Liu and Chi Zhang},
booktitle = {Proceedings of ICASSP},
year = {2026},
pages = {19782--19786}
}
@inproceedings{care-esc,
title = {CARE: Cognitive-Reasoning Augmented Reinforcement for Emotional Support Conversation},
author = {Jie Zhu and Yuanchen Zhou and Shuo Jiang and Junhui Li and Lifan Guo and Feng Chen and Chi Zhang and Fang Kong},
booktitle = {Proceedings of ICASSP},
year = {2026},
pages = {17547--17551}
}
@article{dianjin-ocr-r1,
title = {DianJin-OCR-R1: Enhancing OCR Capabilities via a Reasoning-and-Tool Interleaved Vision-Language Model},
author = {Qian Chen and Xianyin Zhang and Lifan Guo and Feng Chen and Chi Zhang},
journal = {arXiv preprint arXiv:2508.13238},
year = {2025}
}
@inproceedings{m3finmeeting,
title = {M$^3$FinMeeting: A Multilingual, Multi-Sector, and Multi-Task Financial Meeting Understanding Evaluation Dataset},
author = {Jie Zhu and Junhui Li and Yalong Wen and Xiandong Li and Lifan Guo and Feng Chen},
booktitle = {Findings of the Association for Computational Linguistics: ACL 2025},
year = {2025},
pages = {244--266}
}
@article{dianjin-r1,
title = {DianJin-R1: Evaluating and Enhancing Financial Reasoning in Large Language Models},
author = {Jie Zhu and Qian Chen and Huaixia Dou and Junhui Li and Lifan Guo and Feng Chen and Chi Zhang},
journal = {arXiv preprint arXiv:2504.15716},
year = {2025}
}
@inproceedings{cflue,
title = {Benchmarking Large Language Models on CFLUE - A Chinese Financial Language Understanding Evaluation Dataset},
author = {Jie Zhu and Junhui Li and Yalong Wen and Lifan Guo},
booktitle = {Findings of the Association for Computational Linguistics: ACL 2024},
year = {2024},
pages = {5673--5693}
}
Qwen DianJin is developed by Alibaba Cloud. Original code and documentation in this repository are Copyright (c) 2026, Alibaba Cloud and its affiliates, and are licensed under the MIT License, except where a file or subdirectory states otherwise.
This repository contains third-party source code and separately licensed data. Existing file-level and component-level copyright and license notices must be retained. In particular, parts of DianJin-PRM and DianJin-OCR-R1 are derived from Apache-2.0 projects, while DianJin-RED benchmark data is subject to CC BY-NC 4.0 and applicable upstream terms. The root MIT License does not override those terms. See NOTICE for details.
Contributions are welcome under the policies in CONTRIBUTING.md and the Code of Conduct.
Thank you for your interest in the Tongyi DianJin series! For research or product inquiries, reach our team via email at [email protected], or scan the QR code below to join our DingTalk group.
We assume no legal liability for the use of the DianJin open-source models and data. Users are responsible for independently evaluating and assuming any potential risks, verifying model outputs, and making informed decisions for their own scenarios. The released data and models are intended for academic research and industry applications, to advance AI in data analysis, financial innovation, and related fields.
