Jinyi Han 韩槿一

Ph.D. candidate at East China Normal University · Shanghai, China

I am a third-year Ph.D. candidate advised by Prof. Yanghua Xiao. I also work with Dr. Jiaqing Liang at the Knowledge Works Lab, Fudan University.

My research focuses on language agents, reasoning models, and LLM post-training, with an emphasis on self-evolution, tool use, reasoning efficiency, and reinforcement learning.

Selected for Tencent 2026 Project UP and the ByteDance Soaring Star Talent Program.

Email CV Google Scholar GitHub

Internships

LLM Agent Researcher

Tencent TEG Hunyuan · Tencent 2026 Project UP

May 2026 - Present

Multimodal LLM Algorithm Engineer

ByteDance · Soaring Star Talent Program

Jul 2025 - Apr 2026

LLM Reasoning & Agent Researcher

Knowledge Works Lab · Fudan University

Sep 2023 - Present

News

Recent updates

2026.08
Released SKILL-USE, a benchmark for evaluating whether LLM agents can recognize and correctly apply skills.
2026.05
Joined Tencent TEG Hunyuan as a research intern, working on LLM agent evaluation.
2026.05
Two papers accepted at ACL 2026 and one at ICML 2026.
2026.04
GenericAgent open-sourced and reached 13K+ GitHub stars.
2026.04
Attended ICLR 2026 in Rio de Janeiro, Brazil.
2026.02
Three papers accepted at ICLR 2026.
2025.05
CDS accepted at ACL 2025.
2025.02
Released a compact GRPO implementation, now with 1.6K+ GitHub stars.

Selected Publications

First author, then second author

First Author

Preprint 2026

SKILL-USE: Can LLMs Actually Use Skills in Agentic Harnesses?

Jinyi Han*, Yuanjian Xu*, Ying Liao, Xinyi Wang, Zishang Jiang, Zixiang Di, Fanyang Lu, Zhichao Hu, Yanghua Xiao

Key contribution. Introduces SKILL-USE, a benchmark of 79 real skills and 177 executable tasks that evaluates Trigger, Compliance, and Boundary under progressive disclosure. Paper Code

Preprint 2026

Not All Negative Samples Are Equal: LLMs Learn Better from Plausible Reasoning

Zixiang Di*, Jinyi Han*, Shuo Zhang, Ying Liao, Zhi Li, Xiaofeng Ji, et al.

Key contribution. Synthesizes plausible but ultimately incorrect reasoning traces as higher-quality negative data for preference optimization. Paper

Preprint 2026

Structured Reasoning for Large Language Models

Jinyi Han, Zixiang Di, Zishang Jiang, Ying Liao, Jiaqing Liang, Yongqi Wang, Yanghua Xiao

Key contribution. Decouples reasoning into generation, verification, and revision, with targeted supervision that improves self-verification and reduces redundant reasoning. Paper

ICLR 2026 2025

Your Models Have Thought Enough: Training Large Reasoning Models to Stop Overthinking

Jinyi Han, Ying Huang, Ying Liao, Zishang Jiang, Xikun Lu, Haiquan Zhao, Xinyi Wang, et al.

Key contribution. Proposes JET, which uses truncated rollouts and a quality-controlled length reward to teach reasoning models when to stop without sacrificing accuracy. Paper Code

ICLR 2026 2025

A Stitch in Time Saves Nine: Proactive Self-Refinement for Language Models

Jinyi Han, Xinyi Wang, Haiquan Zhao, Tingyun Li, Zishang Jiang, Sihang Jiang, et al.

Key contribution. Proposes PASR, enabling a model to decide whether, when, and how to refine its output during generation rather than repeatedly rewriting the full response. Paper Code

Preprint 2025

Mind the Generation Process: Fine-Grained Confidence Estimation During LLM Generation

Jinyi Han, Tingyun Li, Shisong Chen, Jie Shi, Xinyi Wang, Guanglei Yue, et al.

Key contribution. Introduces FineCE for continuous, fine-grained confidence estimation during text generation, including backward confidence integration at inference time. Paper

AAAI 2024 2024

Small Language Model Can Self-correct

Haixia Han (Jinyi Han), Jiaqing Liang, Jie Shi, Qianyu He, Yanghua Xiao

Key contribution. Builds intrinsic self-correction data with Partial Answer Masking, enabling small language models to trigger and perform correction without complex prompts. Paper

Second Author

Preprint 2026

GenericAgent: A Token-Efficient Self-Evolving LLM Agent via Contextual Information Density Maximization

Jiaqing Liang, Jinyi Han, Weijia Li, Xinyi Wang, Zhoujia Zhang, Zishang Jiang, et al.

Key contribution. Maximizes decision-relevant information within a fixed context through compact tools, hierarchical memory, experience-to-skill evolution, and context compression. Paper Code

ACL 2026 2025

Don't Tell the Answer, Truly Guide the Reasoning During RL Rollouts

Xinyi Wang, Jinyi Han, Zishang Jiang, Tingyun Li, Jiaqing Liang, Sihang Jiang, et al.

Key contribution. Uses abstract meta-hints and affinity-aware policy optimization to guide difficult RL rollouts while avoiding answer leakage and unstable off-policy updates. Paper

ICLR 2026 2025

Selective Expert Guidance for Effective and Diverse Exploration in Reinforcement Learning of LLMs

Zishang Jiang, Jinyi Han, Tingyun Li, Xinyi Wang, Sihang Jiang, Jiaqing Liang, et al.

Key contribution. Proposes MENTOR, which injects expert guidance only at critical decision points to preserve both effective and diverse exploration in RLVR. Paper

AAAI 2026 2026

Difficulty Is Not Enough: Curriculum Learning for LLMs Fine-tuning Must Consider Utility

Zishang Jiang, Jinyi Han, Tingyun Li, Xinyi Wang, Sihang Jiang, Xiaojun Meng, et al.

Key contribution. Introduces DUCL, a curriculum framework that jointly scores sample difficulty and utility and schedules data with window ordering. Paper

ACL 2025 2025

CDS: Data Synthesis Method Guided by Cognitive Diagnosis Theory

Haokun Zhao, Jinyi Han, Jiaqing Liang, Yanghua Xiao

Key contribution. Applies cognitive diagnosis to identify fine-grained model weaknesses and synthesize targeted training data for those knowledge points. Paper

ICLR 2025 2024

Think Thrice Before You Act: Progressive Thought Refinement in Large Language Models

Chengyu Du, Jinyi Han, Yizhou Ying, Aili Chen, Qianyu He, Haokun Zhao, et al.

Key contribution. Proposes progressive thought refinement with collaborative data selection and thought-mask fine-tuning to improve responses across diverse tasks. Paper

* Equal contribution. Only selected first- and second-author work is shown. See the full publication list.

Projects

GenericAgent Token-efficient, self-evolving agent for long-horizon tasks · 13.6K+ stars GitHub
Autonomous Tool Learning Reinforcement learning for proactive LLM tool use Report
Simple GRPO A compact R1-Zero reasoning reproduction · 1.6K+ stars GitHub

Background

Education

Ph.D. in Computer Science East China Normal University · Joint with Fudan University · 2023-2027 (expected)
M.S. in Computer Science Donghua University · 2020-2023
B.S. in Computer Science Henan University of Economics and Law · 2016-2020

Selected Honors

Huawei Spark Award2024, 2025
Outstanding Graduate of ShanghaiTop 5% · 2023
Outstanding Master's ThesisDonghua University · Top 3% · 2023
Outstanding Bachelor's ThesisHenan Province · Top 3% · 2021