Recent advances in vision-language models (VLMs) emphasize long chain-of-thought reasoning; yet, we find that their performance on visual tasks is primarily limited by a lack of visual perception as opposed to reasoning itself. In this work, we systematically study the interplay between perception and reasoning in VLM post-training by decomposing their capabilities into three separate training stages: visual perception, visual reasoning, and textual reasoning, incorporating specialized training data. We demonstrate that visual perception (a) requires targeted optimization with specialized data; (b) serves as a fundamental scaffold that should be solidified through staged training before refining visual reasoning; and (c) is more effectively learned via RL than caption-based SFT. Our experiments across multiple VLMs demonstrate that staged training consistently improves both visual perception and reasoning performance over merged training. Notably, models trained with our approach achieve 1.5% higher reasoning accuracy with 20.8% shorter reasoning traces, suggesting that superior perception reduces the need for excessive reasoning. Furthermore, we show that this capability-based staging represents a new curriculum dimension orthogonal to traditional difficulty-based curricula, and combining both yields further additive gains. Our staged-training models achieve superior performance among open-weight VLMs, establishing advanced results on several visual math and perception (e.g., +5.2% on WeMath and +3.7% on RealWorldQA) tasks compared with the base counterpart.
Seeing Red, Thinking Bad: Color Bias in Vision Language Models
National Institute of Advanced Industrial Science and Technology (AIST)（日本产业技术综合研究所）；University of Tsukuba（筑波大学）；University of Technology Nuremberg（纽伦堡工业大学）；University of Oxford（牛津大学）
Are VLMs Seeing or Just Saying? Uncovering the Illusion of Visual Re-examination
University of Southern California（南加州大学）；University of California San Diego（加州大学圣地亚哥分校）；Carnegie Mellon University（卡内基梅隆大学）；University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）
Prism: A Framework for Decoupling and Assessing the Capabilities of VLMs
Tongji University（同济大学）；Nanjing University of Posts and Telecommunications（南京邮电大学）；University of Science and Technology of China（中国科学技术大学）；Shanghai AI Laboratory（上海人工智能实验室）；The Chinese University of Hong Kong（香港中文大学）；Nanjing University（南京大学）
GeoReason: Aligning Thinking And Answering In Remote Sensing Vision-Language Models Via Logical Consistency Reinforcement Learning
Aerospace Information Research Institute, Chinese Academy of Sciences（中国科学院航空信息研究所）；Key Laboratory of Target Cognition and Application Technology, Chinese Academy of Sciences（中国科学院目标认知与应用技术重点实验室）；University of Chinese Academy of Sciences（中国科学院大学）
Atomic Thinking of LLMs: Decoupling and Exploring Mathematical Reasoning Abilities
Sun Yat-sen University（中山大学）；Tsinghua University（清华大学）；ByteDance Inc（字节跳动公司）；Arizona State University（亚利桑那州立大学）；Tencent Youtu Lab（腾讯优图实验室）；Fudan University（复旦大学）；Harbin Institute of Technology (Shenzhen)（哈尔滨工业大学（深圳））；Guangdong Provincial Key Laboratory of Fire Science and Intelligent Emergency Technology（广东省消防科学与智能应急技术重点实验室）；University of Illinois Chicago（伊利诺伊大学香槟分校）
The Markovian Thinker: Architecture-Agnostic Linear Scaling of Reasoning
Mila；Microsoft Research（微软研究院）；McGill University（麦吉尔大学）；ServiceNow Research（ServiceNow研究院）；Canada CIFAR AI Chair（加拿大CIFAR人工智能主席）；Chandar Research Lab（Chandar研究实验室）；Polytechnique Montréal（蒙特利尔理工学院）；Université de Montréal（蒙特利尔大学）
