Biography

• I am a Senior Researcher at Tencent Hunyuan. My research focuses on CUA/Code/Search and long horizon agentic models, reinforcement learning, and synthetic data.
• Previously, I created WizardLM at Microsoft, which contributed the state-of-the-art LLM family WizardLM, WizardCoder, and WizardMath. I also created the widely adopted Evol-Instruct method.
• I received my M.S. degree from the Institute of Computational Linguistics at Peking University, advised by Prof. Houfeng Wang.
• We are hiring research interns. If you have strong experience in LLMs, RL, agents and are excited to work with Hunyuan, feel free to contact me.

News

• [Jul 2026] We introduce TurnOPD, a turn-aware on-policy distillation methodology for efficient long-horizon agent RL training.
• [Jun 2026] We introduce VeriEvol, a verifiable Evol-Instruct methodology for scaling multimodal mathematical reasoning.
• [Mar 2026] Our RL papers RubricBench and Beyond Length Scaling are featured on Hugging Face Daily Papers.
• [Jan 2026] AgentMath is accepted by ICLR 2026.
• [Jan 2026] We release OffSeeker, studying online reinforcement learning for deep research agents.
• [Aug 2025] We release Hunyuan-Large-Vision, which ranks #5 globally on LMArena-Vision and is the #1 VLM of China.
• [May 2025] We release Hunyuan-TurboS, which ranks #7 globally on LMArena-Text and is the #2 LLM of China.
• [Oct 2024] Wizard models achieve 3M+ HF downloads, 9K+ GitHub stars, and #4 globally (#1 open source) on LMSYS Arena.
• [April 2024] Release WizardLM-2, which outperforms GPT-4 on MT-Bench, GPT4-Turbo on AlpacaEval 2.0 and Claude 3 Sonnet on Arena-Hard.
• [June 2023] WizardLM achieves the 1st-rank of the opensource models on Standford AlpacaEval Leaderboard.
• [April 2023] Release WizardLM expertized in following complex instructions. [Github] (Over 9K Stars) [WizardLM Pages] [Huggingface] [Beeboom: 12 Best LLMs in 2024] [华尔街见闻]
• 1 long paper accepted by ICLR 2024!
• 1 long paper about Adversarial Knowledge Stimulated Contrastive Prompting for Few-shot Language Learners accepted by ACL 2023!
• 1 long paper about Self-Supervised Multi-Modal Sequential Recommendation arxiv!
• 1 long paper about Knowledge Stimulated Contrastive Prompting for Low-Resource Stance Detection accepted by EMNLP 2022!
• 1 long paper about Multimodal Dialogue Response Generation accepted by ACL 2022!

Selected Publications [Google Scholar]

(^*: Equal contribution, ^#: The intern I mentored)

Agentic Model

• AgentMath: Empowering Mathematical Reasoning for Large Language Models via Tool-Augmented Agent
  Haipeng Luo, Huawen Feng, Qingfeng Sun, Can Xu, Kai Zheng, Yufei Wang, Tao Yang, Han Hu, Yansong Tang, Di Wang
  ICLR 2026

• TurnOPD: Making On-Policy Distillation Turn-Aware for Efficient Long-Horizon Agent Training
  Yuhang Zhou^#, Kai Zheng, Haoling Li, Dengyun Peng, Can Xu, Jingjing Chen
  arXiv preprint arXiv:2607.05804

• VeriEvol: Scaling Multimodal Mathematical Reasoning via Verifiable Evol-Instruct
  Haoling Li^#*, Kai Zheng^*, Jie Wu, Can Xu, Qingfeng Sun, Han Hu, Yujiu Yang
  arXiv preprint arXiv:2606.23543

• OffSeeker: Online Reinforcement Learning Is Not All You Need for Deep Research Agents
  Yuhang Zhou^#, Kai Zheng, Qiguang Chen, Mengkang Hu, Qingfeng Sun, Can Xu, Jingjing Chen
  arXiv preprint arXiv:2601.18467

RLVR & RLHF

• RubricBench: Aligning Model-Generated Rubrics with Human Standards
  Qiyuan Zhang, Junyi Zhou, Yufei Wang, Fuyuan Lyu, Yidong Ming, Can Xu, Qingfeng Sun, Kai Zheng, Peng Kang, Xue Liu, Chen Ma
  ACL 2026

• Beyond Length Scaling: Synergizing Breadth and Depth for Generative Reward Models
  Qiyuan Zhang, Yufei Wang, Tianhe Wu, Can Xu, Qingfeng Sun, Kai Zheng, Xue Liu, Chen Ma
  ACL 2026 Findings

• WizardLM: Empowering Large Language Models to Follow Complex Instructions
  Can Xu^*, Qingfeng Sun^*, Kai Zheng^*, Xiubo Geng, Pu Zhao, Jiazhan Feng, Chongyang Tao, Daxin Jiang
  ICLR 2024

General LLM&VLM

• Hunyuan-TurboS: Advancing Large Language Models through Mamba-Transformer Synergy and Adaptive Chain-of-Thought
  Tencent Hunyuan Team
  arXiv preprint arXiv:2505.15431

• Adversarial Knowledge Stimulated Contrastive Prompting for Few-shot Language Learners
  Kai Zheng, Qingfeng Sun, Yaming Yang, Tengchao Lv, Yeyong Pi, Changlin Zhao, Fei Xu, Qi Zhang
  ACL 2023 Findings

• Self-Supervised Multi-Modal Sequential Recommendation
  Kunzhe Song, Qingfeng Sun, Can Xu, Kai Zheng, Yaming Yang
  arXiv preprint arXiv:2304.13277

• Knowledge Stimulated Contrastive Prompting for Low-Resource Stance Detection
  Kai Zheng, Qingfeng Sun, Yaming Yang, Fei Xu
  EMNLP 2022 Findings

• Multimodal Dialogue Response Generation
  Qingfeng Sun, Yujing Wang, Can Xu, Kai Zheng, Yaming Yang, Huang Hu, Fei Xu, Jessica Zhang, Xiubo Geng, Daxin Jiang
  ACL 2022

Open-source Projects

• WizardLM: WizardLM, WizardCoder, WizardMath, and Evol-Instruct.
• WizardLM-2: strong open instruction-following models from the WizardLM family.
• Evol-Instruct: automatic instruction evolution for building complex instruction-following data.
• Llama-X: open academic research to achieve sota LLMs.
• OffSeeker: online reinforcement learning for deep research agents.
• VeriEvol: verifiable Evol-Instruct method vlm&&llm.

Experiences

• Now, Senior Researcher, Tencent Hunyuan X.
• Previously, Research Scientist, Microsoft AI.
• ERNIE-LLM team Baidu（百度文心), Research Intern.
• M.S., Institute of Computational Linguistics, Peking University.

Academic Services

Program Committee for

• ICLR 2025
• NeurIPS 2024
• NAACL 2024
• COLM 2024
• EMNLP 2022, 2023, 2024
• ACL 2022, 2023
• KDD 2022, 2023