( * : equal contribution, † : corresponding author )

• DataOrchestra: Learning to Orchestrate Per-Example Curation of Pretraining Data

  Zhen Huang, Yikun Wang, Shijie Xia, Pengfei Liu†

  arXiv 2026

  [Paper][Code]

  Summary: A learned orchestrator tailors data curation to each pretraining example, improving data quality with less processing compute.

• AdaCodec: A Predictive Visual Code for Video MLLMs

  Haowen Hou, Zhen Huang, Zheming Liang, Qingyi Si, Chenglin Li, Shuai Dong, Kele Shao, Ruilin Li, Dianyi Wang, Nan Duan, Jiaqi Wang†

  NeurIPS 2026 (spotlight)

  [Paper][Code]

  Summary: Compact representations of inter-frame changes reduce redundant visual tokens and improve video understanding.

• DaVinci-Dev: Agent-native Mid-training for Software Engineering

  Ji Zeng, Dayuan Fu, Tiantian Mi, Yumin Zhuang, Yaxing Huang, Xuefeng Li, Lyumanshan Ye, Muhang Xie, Qishuo Hua, Zhen Huang, Mohan Jiang, Hanning Wang, Jifan Lin, Yang Xiao, Jie Sun, Yunze Wu, Pengfei Liu†

  ICML 2026 (Oral)

  [Paper][Code][🤗 Dataset][🤗 Model (32B)][🤗 Model (72B)]

  Summary: Agent-native mid-training uses realistic development trajectories to strengthen software agents’ repository-level problem solving.

• SciPedia: Unlocking the Value of Scientific Data for Pre-training

  Yiwei Qin*, Zhen Huang*, Tiantian Mi*, Weiye Si, Qipeng Guo, Siyuan Feng, Pengfei Liu†

  ACL 2026 (main)

  [Paper][Code][🤗 Dataset][🤗 Model (3B)][🤗 Model (7B)][🤗 Benchmark]

  Summary: SciPedia is a 900B-token scientific corpus enriched through cleaning and pedagogical augmentation for more effective pretraining.

• InnovatorBench: Evaluating Agents’ Ability to Conduct Innovative LLM Research

  Yunze Wu, Dayuan Fu, Weiye Si, Zhen Huang, Mohan Jiang, Keyu Li, Shijie Xia, Jie Sun, Tianze Xu, Xiangkun Hu, Pengrui Lu, Xiaojie Cai, Lyumanshan Ye, Wenhong Zhu, Yang Xiao, Pengfei Liu†

  ICLR 2026

  [Paper][Code][🤗 Dataset][OpenReview]

  Summary: InnovatorBench evaluates the full LLM research workflow across 20 tasks in an executable research environment.

• LIMO: Less is More for Reasoning

  Yixin Ye*, Zhen Huang*, Yang Xiao, Ethan Chern, Shijie Xia, Pengfei Liu†

  COLM 2025

  [Paper][Code][Model][🤗 Datasets][Featured by AK][机器之心][Talk (Chinese)]

  Summary: Strong mathematical reasoning can emerge from supervised fine-tuning on a small set of carefully curated examples.

• OlympicArena: Benchmarking Multi-discipline Cognitive Reasoning for Superintelligent AI

  Zhen Huang, Zengzhi Wang, Shijie Xia, Xuefeng Li, Haoyang Zou, Ruijie Xu, Run-Ze Fan, Lyumanshan Ye, Ethan Chern, Yixin Ye, Yikai Zhang, Yuqing Yang, Ting Wu, Binjie Wang, Shichao Sun, Yang Xiao, Yiyuan Li, Fan Zhou, Steffi Chern, Yiwei Qin, Yan Ma, Jiadi Su, Yixiu Liu, Yuxiang Zheng, Shaoting Zhang†, Dahua Lin†, Yu Qiao†, Pengfei Liu†

  NeurIPS 2024

  [Paper][Code][Homepage][🤗 Datasets][ 🤗 Competition][Featured by AK][机器之心][量子位]

  Summary: OlympicArena is a multimodal reasoning benchmark with 11,163 bilingual Olympiad problems across seven disciplines.