Shaobo Wang (王少博)

Shaobo Wang at Xitan Hotel in Beijing on his 27th birthday

Xitan Hotel, Beijing
On my 27th birthday

I lead the Midtraining team at MiniMax, working on scaling agentic data and beyond, with a focus on data quality and training efficiency.

I study data scaling: how to use data effectively at scale, and how to turn better data into stronger model capabilities with efficient use of compute. Most of my work focuses on the pretraining and midtraining stages of foundation models, with three closely connected directions:

  • Data selection and mixture: Making the most of data quality, diversity, and relevance to improve training efficiency.
  • Data distillation and synthesis: Compressing useful information and generating new training data to expand what models can learn.
  • Data evaluation: Building efficient, faithful evaluations that reflect real-world model capabilities and guide data decisions.

Previously, I worked on midtraining for Kimi K3 at Moonshot AI, and on pretraining and midtraining for Qwen3–3.7 with the Qwen Team. Meanwhile, I am a Ph.D. student at Shanghai Jiao Tong University, where I am fortunate to be supervised by Prof. Linfeng Zhang.

Opportunities at MiniMax. If you are interested in positions at MiniMax, especially in data scaling and midtraining, feel free to contact me at gszfwsb@gmail.com. I am happy to recruit people who believe in scaling laws and the power of data, and are committed to rigorous, hands-on data work, regardless of their background.

Selected Publications

* denotes equal contribution.

  1. Tech Report
    Kimi K3: Open Frontier Intelligence
    Kimi Team
    arXiv preprint arXiv:2607.24653, 2026
    arXiv PDF
  2. Tech Report
    Qwen3.5: Towards Native Multimodal Agents
    Qwen Team
    Feb 2026
    Website Link
  3. ICML oral
    OPUS: Towards Efficient and Principled Data Selection in Large Language Model Pre-training in Every Iteration
    Shaobo Wang, Xuan Ouyang, Tianyi Xu, Yuzheng Hu, Jialin Liu, Guo Chen, Tianyu Zhang, Junhao Zheng, and
    In International Conference on Machine Learning, 2026
    arXiv HTML PDF Link
  4. ICLR
    Rethinking LLM Evaluation: Can We Evaluate LLMs with 200x Less Data?
    Shaobo Wang*, Cong Wang*, Wenjie Fu*, Yue Min, Mingquan Feng, Isabel Guan, Xuming Hu, Conghui He, and
    In The Fourteenth International Conference on Learning Representations, 2026
    arXiv PDF Code Link
  5. CVPR highlight
    Dataset Distillation with Neural Characteristic Function: A Minmax Perspective
    Shaobo Wang, Yicun Yang, Zhiyuan Liu, Chenghao Sun, Xuming Hu, Conghui He, and Linfeng Zhang
    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025
    arXiv HTML PDF Code News (Chinese)