Zhaochen Su

Hong Kong SAR, China

Hi, welcome! I am Zhaochen Su (苏肇辰), a Ph.D. student at the Hong Kong University of Science and Technology (HKUST), co-advised by Prof. Junxian He and Prof. Yi R. (May) Fung.

I am currently an Agent Research Intern at Kimi (Moonshot AI), contributing to Kimi K2 Thinking, K2.5, K2.6, and K2.7. Before that, I worked with the post-training team at MiniMax, contributing to MiniMax-M2, and was a research intern at Shanghai AI Laboratory.

My recent research focuses on long-horizon agents and AI for quant. If you are interested in collaboration, feel free to reach out!

Experience

Kimi (Moonshot AI) · Agent Research Intern 2025.08 – present
Kimi K2 Thinking, K2.5, K2.6, and K2.7
MiniMax · Post-training Team Intern 2025.06 – 2025.08
MiniMax-M2
Shanghai AI Laboratory · Research Intern 2023.11 – 2025.04

Service

Area Chair: NeurIPS 2026, ACL 2026, ACL 2025, EMNLP 2025
Reviewer: ICLR, NeurIPS, COLM, ACL, EMNLP, ARR, MM, ACM Computing Surveys

News

Jul 08, 2026	Iterative Value Refinement (IVR) receives the Outstanding Paper Award at ACL 2026! 🥇
May 11, 2026	XSkill is accepted by ICML 2026, and two papers are accepted by ACL 2026!
Mar 12, 2026	We released XSkill, a dual-stream framework enabling multimodal agents to continually learn from experience and skills without parameter updates.
Feb 24, 2026	We released AgentVista, the first benchmark for multimodal agents on realistic, ultra-challenging visual scenarios with long-horizon hybrid tool use.
Feb 03, 2026	The Kimi K2.5 tech report has been released.
Jan 22, 2026	Three papers are accepted by ICLR 2026, and one paper is accepted by TACL!
Aug 04, 2025	I joined Kimi (Moonshot AI) as an Agent Research Intern.
Jul 01, 2025	Check out our survey on the reasoning paradigm shift from “Think with Text” to “Think with Image” (1.4k+ on GitHub)!

Selected Publications

Tech Report

Kimi K2.5: Visual Agentic Intelligence

Kimi Team

arXiv preprint arXiv:2602.02276, 2026

arXiv Bib Code

@article{kimiteam2026k25,
  title = {Kimi K2.5: Visual Agentic Intelligence},
  author = {{Kimi Team}},
  journal = {arXiv preprint arXiv:2602.02276},
  year = {2026},
  github_repo = {MoonshotAI/Kimi-K2.5},
}

Tech Report

The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence

MiniMax

arXiv preprint arXiv:2605.26494, 2026

arXiv Bib Code

@article{minimax2026m2,
  title = {The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence},
  author = {{MiniMax}},
  journal = {arXiv preprint arXiv:2605.26494},
  year = {2026},
  github_repo = {MiniMax-AI/MiniMax-M2},
}

ICML 2026

XSkill: Continual Learning from Experience and Skills in Multimodal Agents

Guanyu Jiang^*, Zhaochen Su^*, Xiaoye Qu, and Yi R. (May) Fung

In International Conference on Machine Learning (ICML). Guanyu Jiang and Zhaochen Su contributed equally , 2026

arXiv Bib Code

@inproceedings{jiang2026xskill,
  title = {XSkill: Continual Learning from Experience and Skills in Multimodal Agents},
  author = {Jiang, Guanyu and Su, Zhaochen and Qu, Xiaoye and Fung, Yi R. (May)},
  booktitle = {International Conference on Machine Learning (ICML)},
  year = {2026},
  github_repo = {XSkill-Agent/XSkill},
}

arXiv

AgentVista: Evaluating Multimodal Agents in Ultra-Challenging Realistic Visual Scenarios

Zhaochen Su, Jincheng Gao, Hangyu Guo, Zhenhua Liu, Lueyang Zhang, Xinyu Geng, Shijue Huang, Peng Xia, Guanyu Jiang, Cheng Wang, Yue Zhang, Yi R. (May) Fung, and Junxian He

arXiv preprint arXiv:2602.23166, 2026

arXiv Bib Code

@article{su2026agentvista,
  title = {AgentVista: Evaluating Multimodal Agents in Ultra-Challenging Realistic Visual Scenarios},
  author = {Su, Zhaochen and Gao, Jincheng and Guo, Hangyu and Liu, Zhenhua and Zhang, Lueyang and Geng, Xinyu and Huang, Shijue and Xia, Peng and Jiang, Guanyu and Wang, Cheng and Zhang, Yue and Fung, Yi R. (May) and He, Junxian},
  journal = {arXiv preprint arXiv:2602.23166},
  year = {2026},
  github_repo = {hkust-nlp/AgentVista},
}

ICLR 2026

The Tool Decathlon: Benchmarking Language Agents for Diverse, Realistic, and Long-Horizon Task Execution

Junlong Li, Wenshuo Zhao, Jian Zhao, Weihao Zeng, Haoze Wu, Xiaochen Wang, Rui Ge, Yuxuan Cao, Yuzhen Huang, Wei Liu, Junteng Liu, Zhaochen Su, Yiyang Guo, Fan Zhou, Lueyang Zhang, Juan Michelini, Xingyao Wang, Xiang Yue, Shuyan Zhou, Graham Neubig, and Junxian He

In International Conference on Learning Representations (ICLR), 2026

arXiv Bib Code

@inproceedings{li2026toolathlon,
  title = {The Tool Decathlon: Benchmarking Language Agents for Diverse, Realistic, and Long-Horizon Task Execution},
  author = {Li, Junlong and Zhao, Wenshuo and Zhao, Jian and Zeng, Weihao and Wu, Haoze and Wang, Xiaochen and Ge, Rui and Cao, Yuxuan and Huang, Yuzhen and Liu, Wei and Liu, Junteng and Su, Zhaochen and Guo, Yiyang and Zhou, Fan and Zhang, Lueyang and Michelini, Juan and Wang, Xingyao and Yue, Xiang and Zhou, Shuyan and Neubig, Graham and He, Junxian},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year = {2026},
  github_repo = {hkust-nlp/Toolathlon},
}

arXiv

Thinking with Images for Multimodal Reasoning: Foundations, Methods, and Future Frontiers

Zhaochen Su, Peng Xia, Hangyu Guo, Zhenhua Liu, Yan Ma, Xiaoye Qu, Jiaqi Liu, Yanshu Li, Kaide Zeng, Zhengyuan Yang, Linjie Li, Yu Cheng, Heng Ji, Junxian He, and Yi R. Fung

arXiv preprint arXiv:2506.23918, 2025

arXiv Bib Code

@article{su2025thinking,
  title = {Thinking with Images for Multimodal Reasoning: Foundations, Methods, and Future Frontiers},
  author = {Su, Zhaochen and Xia, Peng and Guo, Hangyu and Liu, Zhenhua and Ma, Yan and Qu, Xiaoye and Liu, Jiaqi and Li, Yanshu and Zeng, Kaide and Yang, Zhengyuan and Li, Linjie and Cheng, Yu and Ji, Heng and He, Junxian and Fung, Yi R.},
  journal = {arXiv preprint arXiv:2506.23918},
  year = {2025},
  github_repo = {zhaochen0110/Awesome_Think_With_Images},
}

arXiv

OpenThinkIMG: Learning to Think with Images via Visual Tool Reinforcement Learning

Zhaochen Su, Linjie Li, Mingyang Song, Yunzhuo Hao, Zhengyuan Yang, Jun Zhang, Guanjie Chen, Jiawei Gu, Juntao Li, Xiaoye Qu, and Yu Cheng

arXiv preprint arXiv:2505.08617, 2025

arXiv Bib Code

@article{su2025openthinkimg,
  title = {OpenThinkIMG: Learning to Think with Images via Visual Tool Reinforcement Learning},
  author = {Su, Zhaochen and Li, Linjie and Song, Mingyang and Hao, Yunzhuo and Yang, Zhengyuan and Zhang, Jun and Chen, Guanjie and Gu, Jiawei and Li, Juntao and Qu, Xiaoye and Cheng, Yu},
  journal = {arXiv preprint arXiv:2505.08617},
  year = {2025},
  github_repo = {zhaochen0110/OpenThinkIMG},
}

NeurIPS 2024

ConflictBank: A Benchmark for Evaluating Knowledge Conflicts in Large Language Models

Zhaochen Su, Jun Zhang, Xiaoye Qu, Tong Zhu, Yanshu Li, Jiashuo Sun, Juntao Li, Min Zhang, and Yu Cheng

In Advances in Neural Information Processing Systems (NeurIPS), 2024

arXiv Bib Code

@inproceedings{su2024conflictbank,
  title = {ConflictBank: A Benchmark for Evaluating Knowledge Conflicts in Large Language Models},
  author = {Su, Zhaochen and Zhang, Jun and Qu, Xiaoye and Zhu, Tong and Li, Yanshu and Sun, Jiashuo and Li, Juntao and Zhang, Min and Cheng, Yu},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year = {2024},
  github_repo = {zhaochen0110/conflictbank},
}