BlueLM-2.5-3B Technical Report
Abstract
We present BlueLM-2.5-3B, a compact and unified dense Multimodal Large Language Model (MLLM) designed for efficient edge-device deployment, offering strong general-purpose and reasoning capabilities. To the best of our knowledge, this is the first 3B-scale MLLM to support both thinking and non-thinking modes, while also enabling explicit control over thinking token budget. BlueLM-2.5-3B is developed through diversified data curation, key data resampling, hybrid heterogeneous reinforcement learning, and a high-performance training infrastructure. Our model achieves superior multimodal capacity while preserving competitive pure-text performance with only 2.9 billion parameters. We conduct comprehensive evaluations across a broad range of multimodal and text-only benchmarks. In thinking mode, BlueLM-2.5-3B achieves comparable performance to Qwen3-4B on text-only benchmarks, and trails the larger Kimi-VL-A3B-16B by only about 5% on average across multimodal evaluations. In non-thinking mode, it outperforms Qwen2.5-VL-3B on the majority of multimodal benchmarks. Additionally, BlueLM-2.5-3B exhibits exceptional data efficiency. All of the aforementioned performance is achieved with substantially less total training data than Qwen2.5-VL-3B and Qwen3-4B. We hope our work contributes to the advancement of high-performance, on-device MLLMs and provides meaningful insights to the research community.
Keywords
Cite
@article{arxiv.2507.05934,
title = {BlueLM-2.5-3B Technical Report},
author = {Baojiao Xiong and Boheng Chen and Chengzhi Wang and Daxiong Luo and Dongsheng Xu and Dongyang Liu and Fan Yang and Fangyuan Li and Fei Teng and Feng Wang and Fukang Qin and Fuquan Peng and Guanxin Tan and Guozhi Wang and Haibo Yu and Haohao Gao and Heng Liu and Hongbo Yang and Hongjian Zou and Houzheng Shen and Hu Meng and Huan Li and Hui Tan and Jiali Chen and Jianzhao Chen and Jinliang Zhu and Kai Wang and Lei Wu and Liangbing Liu and Liuyang Bian and Liyan He and Long Liu and Peiwen Li and Penggang Shi and Qi Ding and Rui Hu and Shuai Cao and Shuai Ren and Shuang Peng and Teng Xie and Weiji Chen and Weilin Xiang and Weixin Wu and Xi Yin and Xiaoxin Chen and Xu Chen and Yafei Wen and Yan Hu and Yanzhou Yang and Yina Xie and Yinghao Chen and Yixuan Liao and Yu Geng and Yuanjiang Ouyang and Yuanzhuo Yang and Yuehua He and Yushuai Peng and Zhaoxiong Wang and Zheng Wang and Zhibo Zhou and Ziyang Wu},
journal= {arXiv preprint arXiv:2507.05934},
year = {2025}
}