I am currently working on foundation model pretraining at StepFun. My research interests include large language model pretraining, scaling laws, training stability, muP, and Muon.

I received my M.S. from the Academy of Mathematics and Systems Science, Chinese Academy of Sciences, and my B.S. in Mathematics from Shandong University.

• 2026 ICLR 2026 Oral 🎉 "Can Mixture-of-Experts Surpass Dense LLMs Under Strictly Equal Resources?" accepted.
• 2026 ICLR 2026 🎉 "How Many Code and Test Cases Are Enough?" accepted.
• 2026 Release 🎉 Step-3.5 Flash released. Blog · GitHub · Tech report.
• 2025 NeurIPS 2025 Spotlight 🎉 "Farseer: A Refined Scaling Law in Large Language Models" spotlight.
• 2025 EMNLP 2025 🎉 "Debate-to-Detect: Reformulating Misinformation Detection as a Real-World Debate" accepted.
• 2025.06 Edu 🎉 Obtained M.S. from University of Chinese Academy of Sciences.
• 2025 AAAI 2025 Oral 🎉 "Beyond Detection: Exploring Evidence-based Multi-Agent Debate …" oral.
• 2024.12 Work 🎉 Joined StepFun (foundation model pretraining).
• 2024 EMNLP 2024 🎉 "Breaking Language Barriers: Cross-Lingual Continual Pre-Training at Scale" accepted.

1. Breaking Language Barriers: Cross-Lingual Continual Pre-Training at Scale

   EMNLP 2024
   W Zheng*, W Pan*, X Xu*, L Qin, L Yue, M Zhou
2. StepLaw — Optimal Hyperparameter Scaling Law in Large Language Model Pretraining

   ★ Representative Open-source
   Houyi Li*, Wenzhen Zheng*, Qiufeng Wang*, Hanshan Zhang, Zili Wang, Shijie Xuyang, Yuantao Fan, Shuigeng Zhou, Xiangyu Zhang, Daxin Jiang
   Project page · arXiv · Code
3. Farseer: A Refined Scaling Law in Large Language Models

   NeurIPS 2025 Spotlight
   Houyi Li*, Wenzhen Zheng*, Qiufeng Wang, Zhenyu Ding, Haoying Wang, Zili Wang, Shijie Xuyang, Ning Ding, Shuigeng Zhou, Xiangyu Zhang, Daxin Jiang
   Project page · arXiv · Code
4. Scaling Laws for Code: A More Data-Hungry Regime

   Preprint
   Xianzhen Luo*, Wenzhen Zheng*, Qingfu Zhu, Rongyi Zhang, Houyi Li, Siming Huang, YuanTao Fan, Wanxiang Che
5. Debate-to-Detect: Reformulating Misinformation Detection as a Real-World Debate with Large Language Models

   EMNLP 2025
   C Han*, W Zheng*, X Tang
6. Beyond Detection: Exploring Evidence-based Multi-Agent Debate for Misinformation Intervention & Persuasion

   AAAI 2025 Oral
   C Han, Y Ma, J Tan, W Zheng, X Tang
7. How Many Code and Test Cases Are Enough? Evaluating Test Cases Generation from a Binary-Matrix Perspective

   ICLR 2026
   X Luo, J Huang, W Zheng, Q Zhu, M Xu, Y Xu, Y Fan, L Qin, W Che
8. Can Mixture-of-Experts Surpass Dense LLMs Under Strictly Equal Resources?

   ICLR 2026 Oral
   H Li, KM Lo, Z Wang, Z Wang, W Zheng, S Zhou, X Zhang, D Jiang
9. Step-3 is Large yet Affordable: Model-system Co-design for Cost-effective Decoding

   Model report Open-source
   StepFun Team
   arXiv:2507.19427
10. Step 3.5 Flash: Fast, Sharp & Reliable Agentic Intelligence

    ★ Representative Model report Open-source
    StepFun Team
    Tech report · GitHub · Blog
11. Simulating social network with LLM agents: An analysis of information propagation and echo chambers

    KSS 2024 Oral
    W Zheng, X Tang

• MCM/ICM F Award (Finalist, top 1%)
• National College Math Competition (Math Major Category A) Second Prize
• Hua Luogeng Scholarship, CAS Institute of Mathematics
• Shandong Province College Physics Competition First Prize
• National High School Math League (Anhui Province) First Prize