Introducing LongCat-Flash-Thinking: A Technical Report
Abstract
We present LongCat-Flash-Thinking, an efficient 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model. Its advanced capabilities are cultivated through a meticulously crafted training process, beginning with long Chain-of-Thought (CoT) data cold-start and culminating in large-scale Reinforcement Learning (RL). We first employ a well-designed cold-start training strategy, which significantly enhances the reasoning potential and equips the model with specialized skills in both formal and agentic reasoning. Then, a core innovation is our domain-parallel training scheme, which decouples optimization across distinct domains (e.g., STEM, Code, Agentic) and subsequently fuses the resulting expert models into a single, nearly Pareto-optimal model. This entire process is powered by our Dynamic ORchestration for Asynchronous rollout (DORA) system, a large-scale RL framework that delivers a greater than threefold training speedup over synchronous methods on tens of thousands of accelerators. As a result, LongCat-Flash-Thinking achieves state-of-the-art performance among open-source models on a suite of complex reasoning tasks. The model exhibits exceptional efficiency in agentic reasoning, reducing average token consumption by 64.5% (from 19, 653 to 6, 965) on AIME-25, without degrading task accuracy. We release LongCat-Flash-Thinking to promote further advances in reasoning systems and agentic AI research.
Keywords
Cite
@article{arxiv.2509.18883,
title = {Introducing LongCat-Flash-Thinking: A Technical Report},
author = {Meituan LongCat Team and Anchun Gui and Bei Li and Bingyang Tao and Bole Zhou and Borun Chen and Chao Zhang and Chao Zhang and Chengcheng Han and Chenhui Yang and Chi Zhang and Chong Peng and Chuyu Zhang and Cong Chen and Fengcun Li and Gang Xu and Guoyuan Lin and Hao Jiang and Hao Liang and Haomin Fu and Haoxiang Ma and Hong Liu and Hongyan Hao and Hongyin Tang and Hongyu Zang and Hongzhi Ni and Hui Su and Jiahao Liu and Jiahuan Li and Jialin Liu and Jianfei Zhang and Jianhao Xu and Jianing Wang and Jiaqi Sun and Jiaqi Zhang and Jiarong Shi and Jiawei Yang and Jingang Wang and Jinrui Ding and Jun Kuang and Jun Xu and Ke He and Kefeng Zhang and Keheng Wang and Keqing He and Li Wei and Liang Shi and Lin Qiu and Lingbin Kong and Lingchuan Liu and Linsen Guo and Longfei An and Mai Xia and Meng Zhou and Mengshen Zhu and Peng Pei and Pengcheng Jia and Qi Gu and Qi Guo and Qiong Huang and Quan Chen and Quanchi Weng and Rongxiang Weng and Ruichen Shao and Rumei Li and Shanglin Lei and Shuai Du and Shuaikang Liu and Shuang Zhou and Shuhao Hu and Siyu Xu and Songshan Gong and Tao Liang and Tianhao Hu and Wei He and Wei Shi and Wei Wang and Wei Wu and Wei Zhuo and Weifeng Tang and Wenjie Shi and Wenlong Zhu and Xi Su and Xiangcheng Liu and Xiangyu Xi and Xiangzhou Huang and Xiao Liu and Xiaochen Jiang and Xiaowei Shi and Xiaowen Shi and Xiaoyu Li and Xin Chen and Xinyue Zhao and Xuan Huang and Xuemiao Zhang and Xuezhi Cao and Xunliang Cai and Yajie Zhang and Yang Chen and Yang Liu and Yang Liu and Yang Zheng and Yaoming Wang and Yaqi Huo and Yerui Sun and Yifan Lu and Yiyang Li and Youshao Xiao and Yuanzhe Lei and Yuchen Xie and Yueqing Sun and Yufei Zhang and Yuhuai Wei and Yulei Qian and Yunke Zhao and Yuqing Ding and Yuwei Jiang and Zhaohua Yang and Zhengyu Chen and Zhijian Liu and Zhikang Xia and Zhongda Su and Ziran Li and Ziwen Wang and Ziyuan Zhuang and Zongyu Wang and Zunyuan Yang},
journal= {arXiv preprint arXiv:2509.18883},
year = {2025}
}