Ziniu Li

Research Scientist, Qingyun Talent Program
Tencent Hunyuan Team, Shenzhen

Ph.D., The Chinese University of Hong Kong, Shenzhen
Email: liziniu1997@gmail.com

Research interests

Reinforcement learning for large language models, including algorithm and infrastructure design, long-horizon agents, frontier science, and theoretical foundations.

Hiring. The Hunyuan Team is recruiting elite interns and full-time members for large-scale RL training and frontier science applications that benefit from strong domain knowledge. Please feel free to reach out.

News

July 2026. Multiple works on long-horizon LLM reinforcement learning and post-training were accepted to ICML 2026.

April 2026. A discussion on X connected Claude Mythos to looped language models, an idea explored in our paper Scaling Latent Reasoning via Looped Language Models.

February 2026. I joined Tencent Hunyuan as a Research Scientist in the Qingyun Talent Program.

May 2025. I joined ByteDance Seed-Edge as an intern, starting frontier research on long-term AI directions.

January 2025. Our work on diversity-preserving supervised fine-tuning was accepted to ICLR 2025 and received Best Paper Runner-up at the NeurIPS 2024 FITML Workshop.

July 2024. ReMax appeared at ICML 2024, introducing a simple and efficient REINFORCE-style method for aligning large language models.

Selected publications

Ziniu Li, Jinbo Wang, Guanhua Huang, Feiyuan Zhang, Pengbo Li, Alex Chen. When Do Larger Batches Help Scale LLM Reinforcement Learning?. arXiv. 2026.
Ziniu Li, Congliang Chen, Tianyun Yang, Tian Ding, Ruoyu Sun, Ge Zhang, Wenhao Huang, Zhi-Quan Luo. Knapsack RL: Compute-Efficient Reinforcement Learning via Heterogeneous Rollout Allocation. ICML. 2026.
Rui-Jie Zhu*, Zixuan Wang*, Kai Hua*, Tianyu Zhang*, Ziniu Li*, et al. Scaling Latent Reasoning via Looped Language Models. arXiv. 2025.
Ziniu Li, Congliang Chen, Tian Xu, Zeyu Qin, Jiancong Xiao, Zhi-Quan Luo, Ruoyu Sun. Preserving Diversity in Supervised Fine-tuning of Large Language Models. ICLR. 2025.
Ziniu Li, Tian Xu, Yushun Zhang, Zhihang Lin, Yang Yu, Ruoyu Sun, Zhi-Quan Luo. ReMax: A Simple, Effective, and Efficient Reinforcement Learning Method for Aligning Large Language Models. ICML. 2024.

View all publications