Zhengxiao Du

Also known as 杜正晓

Zhengxiao Du is a tech lead at Zhipu AI, where he co-leads the base model group for the GLM language models. He studied computer science at Tsinghua University, worked in Jie Tang's Knowledge Engineering Group, and received his PhD in 2025. While still a doctoral student, he was first author of the ACL 2022 paper that introduced GLM, a pretraining method based on autoregressive blank infilling, and a coauthor of the GLM-130B model in 2022. He joined Zhipu AI as a tech lead in January 2023, was first author of a NeurIPS 2024 paper on emergent abilities, and was a core contributor to GLM-4.5 in 2025. In February 2026, he was among the seven tech leads of GLM-5.

Now at: Zhipu AI
Badges: Bridge
Links: Google Scholar and GitHub

Timeline

2016–2020

Computer science at Tsinghua

He studies for a B.E. in computer science at Tsinghua University from 2016 to 2020. From 2017 to 2019, he is a research assistant in Jie Tang's Knowledge Engineering Group, and in 2019 he interns at Cornell University under Thorsten Joachims.
2020–2025

PhD under Jie Tang

He starts a PhD in computer science at Tsinghua in 2020 and completes it in the Department of Computer Science and Technology in 2025, advised by Jie Tang. He interns at the Beijing Academy of Artificial Intelligence from 2020 to 2022 and at Alibaba's DAMO Academy in 2020.
2021–2022

The GLM pretraining framework

In March 2021, he posts GLM, a General Language Model pretrained by autoregressive blank infilling, as first author with Yujie Qian, Xiao Liu, Ming Ding, Jiezhong Qiu, Zhilin Yang and Jie Tang. The paper appears at ACL 2022 and reports that GLM outperforms BERT, T5 and GPT at the same model sizes.
October 2022

GLM-130B, an open bilingual model

In October 2022, he is among the authors of GLM-130B, a bilingual English and Chinese language model with 130 billion parameters, presented as an open-source model at least as good as GPT-3 (davinci). The paper is accepted to ICLR 2023.
January 2023

Tech lead at Zhipu AI

He joins ZhipuAI in Beijing as a tech lead in January 2023, co-leading the pre-training team of ChatGLM.
2024

Emergent abilities and GLM-4

In 2024, he is first author of "Understanding Emergent Abilities of Language Models from the Loss Perspective", accepted at NeurIPS 2024, which finds that a model shows emergent abilities on certain tasks once its pre-training loss falls below a specific threshold. He is also among the authors of the June 2024 ChatGLM report on the GLM-4 series, which includes GLM-4, GLM-4-Air and GLM-4-9B.
August 2025

Core contributor to GLM-4.5

He is a core contributor to GLM-4.5, an open-source Mixture-of-Experts model with 355B total parameters and 32B activated parameters, described in a technical report in August 2025. Its hybrid reasoning method supports both thinking and direct response modes.
As of October 2026

Base model group at Zhipu

He is a tech lead at Zhipu AI, where he co-leads the base model group for GLM. In February 2026, he is among the seven tech leads of GLM-5, a foundation model for agentic engineering that adopts DSA to reduce training and inference costs.