VibeThinker-1.5B: Tiny model, big logic
A compact reasoning model showing how diversity-driven post-training can elicit strong reasoning ability without relying on model scale alone.
Large language models · Post-training · Agents
Sen Xu
(徐森)About Me
I am a Senior Research Engineer at the AI R&D Department of Sina Weibo. I received my B.S. and Ph.D. degrees in Computer Science and Technology from Beijing Jiaotong University in 2018 and 2024, where I was advised by Prof. Shikui Wei.
During my Ph.D., I worked primarily on computer vision and visual representations. After joining Weibo, I shifted my research toward foundation models. I am currently the primary technical contributor to the open-source VibeThinker series, coordinating model development and experiments. I now work on reasoning, AI for Science, post-training, and agentic systems for foundation models.
News
-
Preprint
We posted a new preprint introducing CLR, a training-free framework for efficient test-time reasoning.
-
Milestone
VibeThinker-3B ranked #1 on Papers with Code’s Trending Research.
-
Release
We released VibeThinker-3B, which was later featured in X’s Today’s News.
-
Milestone
VibeThinker-1.5B reached #1 on Hugging Face’s trending models list.
-
Release
We released VibeThinker-1.5B, and its technical report achieved the #1 Paper of the Day ranking on Hugging Face.
-
Milestone
Received my Ph.D. from Beijing Jiaotong University.
-
Paper
One paper was accepted to AAAI 2024.
Recent Research
View all
Experience
Senior Research Engineer
AI R&D Department, Sina Weibo Inc. · Beijing
Ph.D. in Computer Science and Technology
Beijing Jiaotong University · advised by Prof. Shikui Wei
B.S. in Computer Science and Technology
Beijing Jiaotong University