Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LongCat-Flash-Thinking-2601 Technical Report

Meituan LongCat Team, Anchun Gui, Bei Li, Bingyang Tao, Bole Zhou, Borun Chen, Chao Zhang, Chao Zhang, Chen Gao, Chen Zhang, Chengcheng Han, Chenhui Yang, Chuyu Zhang, Cong Chen, Cunguang Wang, Daoru Pan, Defei Bu, Dengchang Zhao, Di Xiu, Dishan Liu, Dongyu Ru, Dunwei Tu, Fan Wu, Fengcheng Yuan, Fengcun Li, Gang Xu, Guanyu Wu, Guoyuan Lin, Haibin Wang, Hansi Yang, Hao Yang, Haonan Yan, Haoxiang Ma, Haoxing Wen, Hongyan Hao, Hongyin Tang, Hongyu Zang, Hongzhi Ni, Hui Su, Jiacheng Zhang, Jiahong Zhou, Jiahuan Li, Jiaming Wang, Jian Yang, Jianfei Zhang, Jianhao Xu, Jianing Wang, Jiapeng Zhu, Jiaqi Sun, Jiarong Shi, Jiarui Zhao, Jingang Wang, Jinluan Yang, Jinrui Ding, Jinwei Xiao, Jiyuan He, Juncan Xu, Kefeng Zhang, Keheng Wang, Li Wei, Lianhui Ma, Lin Qiu, Lingbing Kong, Lingchuan Liu, Linsen Guo, Mengshen Zhu, Mengxia Shen, Mingyang Zhu, Peiguang Li, Peng Pei, Pengcheng Jia, Pengtao Zhang, Peng Zhao, Qi Gu, Qiong Huang, Qiyuan Duan, Quanchi Weng, Rongxiang Weng, Rongzhi Zhang, Rumei Li, Shanglin Lei, Shengnan An, Shijun Dai, Shuaikang Liu, Shuang Zhou, Shuo Wang, Songyuan Zhao, Tao Liang, Tianhao Hu, Tianze Chen, Wei Liu, Wei Shi, Wei Wang, Weifeng Tang, Wenjie Shi, Wenlong Zhu, Wentao Chen, Wentao Shi, Xi Su, Xiangcheng Liu, Xiandi Ma, Xiangyu Xi, Xiangyuan Liu, Xiangzhou Huang, Xiao Liu, Xiaodong Cai, Xiaolong Chen, Xiaowei Shi, Xiaoyu Li, Xin Chen, Xingchen Liu, Xuan Huang, Xuezhi Cao, Xunliang Cai, Yan Chen, Yang Bai, Yang Liu, Yang Yang, Yang Zheng, Yaoming Wang, Yaoming Zhu, Yaqi Huo, Yanyu Chen, Yaorui Shi, Yerui Sun, Yi Zhang, Yihao Chen, Yi-Kai Zhang, Yifan Lu, Yifan Zhao, Yitao Zhai, Yongjing Yin, Yongwei Zhou, Youshao Xiao, Yuchuan Dai, Yuchen Xie, Yuchen Yu, Yufei Zhang, Yuhuai Wei, Yulei Qian, Yunfan Liang, Yunke Zhao, Yuwei Jiang, Yuxin Bian, Yuxin Chen, Yuxin Liu, Yue Xu, Yueqing Sun, Zeyang Yu, Zhao Yang, Zhengsheng Huang, Zhengyu Chen, Zhijian Liu, Zhikang Xia, Zhimin Lin, Zhiyuan Yao, Zhuofan Chen, Zhuowen Han, Zijian Zhang, Ziran Li, Ziwen Wang, Ziyuan Zhuang

TL;DR

LongCat-Flash-Thinking-2601 tackles enabling agentic reasoning in large language models by allowing structured interaction with external environments across long-horizon tasks. It introduces a 560B open-source Mixture-of-Experts model trained with a unified pipeline that combines text-driven and environment-grounded data synthesis, planning-oriented augmentation, and scalable multi-domain RL via the DORA framework, plus a Heavy Thinking mode for test-time scaling. The system constructs over 20 domains of executable environments with automated tool graphs, implements robust RL training with noise curricula, and uses PD disaggregation to sustain throughput on large hardware. Empirically, it achieves state-of-the-art open-source performance on agentic benchmarks such as BrowseComp and RWSearch, while maintaining strong math, coding, and general QA capabilities; it also demonstrates efficient long-context inference via ZigZag Attention that scales to up to 1,000,000 tokens. Collectively, the work advances practical, robust agentic reasoning for real-world tasks and provides open-source ZigZag variants for ultra-long contexts.

Abstract

We introduce LongCat-Flash-Thinking-2601, a 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model with superior agentic reasoning capability. LongCat-Flash-Thinking-2601 achieves state-of-the-art performance among open-source models on a wide range of agentic benchmarks, including agentic search, agentic tool use, and tool-integrated reasoning. Beyond benchmark performance, the model demonstrates strong generalization to complex tool interactions and robust behavior under noisy real-world environments. Its advanced capability stems from a unified training framework that combines domain-parallel expert training with subsequent fusion, together with an end-to-end co-design of data construction, environments, algorithms, and infrastructure spanning from pre-training to post-training. In particular, the model's strong generalization capability in complex tool-use are driven by our in-depth exploration of environment scaling and principled task construction. To optimize long-tailed, skewed generation and multi-turn agentic interactions, and to enable stable training across over 10,000 environments spanning more than 20 domains, we systematically extend our asynchronous reinforcement learning framework, DORA, for stable and efficient large-scale multi-environment training. Furthermore, recognizing that real-world tasks are inherently noisy, we conduct a systematic analysis and decomposition of real-world noise patterns, and design targeted training procedures to explicitly incorporate such imperfections into the training process, resulting in improved robustness for real-world applications. To further enhance performance on complex reasoning tasks, we introduce a Heavy Thinking mode that enables effective test-time scaling by jointly expanding reasoning depth and width through intensive parallel thinking.

LongCat-Flash-Thinking-2601 Technical Report

TL;DR

LongCat-Flash-Thinking-2601 tackles enabling agentic reasoning in large language models by allowing structured interaction with external environments across long-horizon tasks. It introduces a 560B open-source Mixture-of-Experts model trained with a unified pipeline that combines text-driven and environment-grounded data synthesis, planning-oriented augmentation, and scalable multi-domain RL via the DORA framework, plus a Heavy Thinking mode for test-time scaling. The system constructs over 20 domains of executable environments with automated tool graphs, implements robust RL training with noise curricula, and uses PD disaggregation to sustain throughput on large hardware. Empirically, it achieves state-of-the-art open-source performance on agentic benchmarks such as BrowseComp and RWSearch, while maintaining strong math, coding, and general QA capabilities; it also demonstrates efficient long-context inference via ZigZag Attention that scales to up to 1,000,000 tokens. Collectively, the work advances practical, robust agentic reasoning for real-world tasks and provides open-source ZigZag variants for ultra-long contexts.

Abstract

We introduce LongCat-Flash-Thinking-2601, a 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model with superior agentic reasoning capability. LongCat-Flash-Thinking-2601 achieves state-of-the-art performance among open-source models on a wide range of agentic benchmarks, including agentic search, agentic tool use, and tool-integrated reasoning. Beyond benchmark performance, the model demonstrates strong generalization to complex tool interactions and robust behavior under noisy real-world environments. Its advanced capability stems from a unified training framework that combines domain-parallel expert training with subsequent fusion, together with an end-to-end co-design of data construction, environments, algorithms, and infrastructure spanning from pre-training to post-training. In particular, the model's strong generalization capability in complex tool-use are driven by our in-depth exploration of environment scaling and principled task construction. To optimize long-tailed, skewed generation and multi-turn agentic interactions, and to enable stable training across over 10,000 environments spanning more than 20 domains, we systematically extend our asynchronous reinforcement learning framework, DORA, for stable and efficient large-scale multi-environment training. Furthermore, recognizing that real-world tasks are inherently noisy, we conduct a systematic analysis and decomposition of real-world noise patterns, and design targeted training procedures to explicitly incorporate such imperfections into the training process, resulting in improved robustness for real-world applications. To further enhance performance on complex reasoning tasks, we introduce a Heavy Thinking mode that enables effective test-time scaling by jointly expanding reasoning depth and width through intensive parallel thinking.
Paper Structure (48 sections, 6 equations, 15 figures, 2 tables)

This paper contains 48 sections, 6 equations, 15 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Pre-Training
  3. Text-driven synthesis
  4. Environment-grounded synthesis
  5. Planning-Oriented Data Augmentation
  6. Scaling Reinforcement Learning
  7. RL Preparation
  8. Environment
  9. Code Sandbox
  10. Agentic Tool-Use Environment
  11. Initial Policy
  12. General Thinking
  13. Agentic Coding
  14. Agentic Search
  15. Agentic Tool-Use
  16. ...and 33 more sections

Figures (15)

  • Figure 1: Benchmark performance of LongCat-Flash-Thinking-2601.
  • Figure 2: The comparison of agentic capability (pass@k. "Baseline" represents the model undergoing the old mid-training recipe, while "Enhanced" represents the model undergoing the agentic-enhanced mid-training recipe.
  • Figure 3: Automated construction of executable domain graph. Starting from a high-level domain specification, the pipeline synthesizes domain-specific tools, generates corresponding databases schema and tool schema implementations. Building on these we construct a verified tool dependency graph that serves as the foundation for environment scaling.
  • Figure 4: Verifiability-preserving environment expansion. Environment is progressively expanded from a seed executable tool chain through controlled graph growth.
  • Figure 5: The Execution Workflow of Our Scalable Asynchronous Agentic RL Framework.
  • ...and 10 more figures