Cover Image

Saved in:
Bibliographic Details
Main Authors: Huang, Ailin, Yao, Chengyuan, Han, Chunrui, Wan, Fanqi, Guo, Hangyu, Lv, Haoran, Zhou, Hongyu, Wang, Jia, Zhou, Jian, Sun, Jianjian, Hu, Jingcheng, Lin, Kangheng, Zhao, Liang, Huang, Mitt, Yuan, Song, Qu, Wenwen, Wang, Xiangfeng, Lai, Yanlin, Zhao, Yingxiu, Zhang, Yinmin, Shi, Yukang, Chen, Yuyang, Weng, Zejia, Meng, Ziyang, Li, Ang, Kong, Aobo, Dong, Bo, Wan, Changyi, Wang, David, Qi, Di, Li, Dingming, Yu, En, Li, Guopeng, Yin, Haiquan, Zhou, Han, Zhang, Hanshan, Yan, Haolong, Zhou, Hebin, Peng, Hongbo, Zhang, Jiaran, Lv, Jiashu, Fu, Jiayi, Cheng, Jie, Zhou, Jie, Yin, Jisheng, Xie, Jingjing, Wu, Jingwei, Zhang, Jun, Liu, Junfeng, Tan, Kaijun, Yan, Kaiwen, Chen, Liangyu, Chen, Lina, Li, Mingliang, Zhao, Qian, Sun, Quan, Pang, Shaoliang, Fan, Shengjie, Shang, Shijie, Zhang, Siyuan, You, Tianhao, Ji, Wei, Xie, Wuxun, Yang, Xiaobo, Hou, Xiaojie, Jiao, Xiaoran, Ren, Xiaoxiao, Kong, Xiangwen, Huang, Xin, Wu, Xin, Chen, Xing, Wang, Xinran, Zhang, Xuelin, Wei, Yana, Li, Yang, Xu, Yanming, Shen, Yeqing, Peng, Yuang, Peng, Yue, Zhou, Yu, Li, Yusheng, Yang, Yuxiang, Zhang, Yuyang, Xie, Zhe, Huang, Zhewei, Lu, Zhenyi, Fan, Zhimin, Cheng, Zihui, Jiang, Daxin, Han, Qi, Zhang, Xiangyu, Zhu, Yibo, Ge, Zheng
Format: Preprint
Published: 2026
Subjects:
Computer Vision and Pattern Recognition
Online Access:https://arxiv.org/abs/2601.09668
Tags: Add Tag
No Tags, Be the first to tag this record!
Table of Contents:
  • We present STEP3-VL-10B, a lightweight open-source foundation model designed to redefine the trade-off between compact efficiency and frontier-level multimodal intelligence. STEP3-VL-10B is realized through two strategic shifts: first, a unified, fully unfrozen pre-training strategy on 1.2T multimodal tokens that integrates a language-aligned Perception Encoder with a Qwen3-8B decoder to establish intrinsic vision-language synergy; and second, a scaled post-training pipeline featuring over 1k iterations of reinforcement learning. Crucially, we implement Parallel Coordinated Reasoning (PaCoRe) to scale test-time compute, allocating resources to scalable perceptual reasoning that explores and synthesizes diverse visual hypotheses. Consequently, despite its compact 10B footprint, STEP3-VL-10B rivals or surpasses models 10$\times$-20$\times$ larger (e.g., GLM-4.6V-106B, Qwen3-VL-235B) and top-tier proprietary flagships like Gemini 2.5 Pro and Seed-1.5-VL. Delivering best-in-class performance, it records 92.2% on MMBench and 80.11% on MMMU, while excelling in complex reasoning with 94.43% on AIME2025 and 75.95% on MathVision. We release the full model suite to provide the community with a powerful, efficient, and reproducible baseline.

Similar Items