avatar

Longguang Zhong

Member of Technical Staff
Moonshot AI
2813409972@qq.com


About Me

Hello, Iā€™m Longguang Zhong, a Member of Technical Staff at Moonshot AI, working on large language models. I received my M.S. in Computer Technology from Sun Yat-sen University, advised by Prof. Xiaojun Quan, and my B.E. in Software Engineering from Xidian University.

Research Interests

My research focuses on large language models, with a current emphasis on agents and reinforcement learning.

News

[February 2026] šŸ”„šŸ”„ We release Kimi K2.5, an open-source visual agentic intelligence model. Check out the tech report here.

[October 2025] šŸ”„šŸ”„ We release Kimi Linear, an expressive and efficient attention architecture. Check out the tech report here.

[July 2025] šŸ”„šŸ”„ We release Kimi K2, an open agentic intelligence model. Check out the tech report here.

[August 2025] šŸ”„šŸ”„ ThinkSwitcher, our work on adaptive thinking strategies for language reasoning models, is accepted to EMNLP 2025 Findings! Check out the paper here.

[August 2025] šŸ”„šŸ”„ FuseChat, our work on knowledge fusion of chat models, is accepted to EMNLP 2025 Main! Check out the paper here and the code on GitHub.

[May 2025] šŸ”„ BlockPruner, a fine-grained block pruning framework for large language models, is accepted to ACL 2025 Findings! Check out the paper here and the code on GitHub.

[April 2025] šŸ”„ We release FuseRL, a dense preference optimization framework for heterogeneous model fusion. Check out the tech report here.

[Jan 2025] šŸ”„ We release FuseO1-Preview, an advanced fusion model that enhances System-II reasoning by integrating multiple O1-like models using SCE merging, excelling in mathematics, coding, and science.

[Dec 2024] šŸ”„ We release FuseChat-3.0 and Blog Post. FuseChat-3.0 contains a series of models crafted to enhance performance by integrating the strengths of multiple source LLMs into more compact target LLMs.

[Aug 2024] šŸ”„ We update the FuseChat tech report and release FuseChat-7B-v2.0, which is the fusion of six prominent chat LLMs with diverse architectures and scales. FuseChat-7B-v2.0 achieves an average performance of 7.38 on MT-Bench (GPT-4-0125-Preview as judge LLM), which is comparable to Mixtral-8x7B-Instruct and approaches GPT-3.5-Turbo-1106.

Publications

  1. Tech Report
    Kimi K2.5: Visual Agentic Intelligence
    Kimi Team (including Longguang Zhong)
    arXiv preprint arXiv:2602.02276, 2026.
    PDF Tech Report

  2. Tech Report
    Kimi Linear: An Expressive, Efficient Attention Architecture
    Kimi Team (including Longguang Zhong)
    arXiv preprint arXiv:2510.26692, 2025.
    PDF Tech Report

  3. Tech Report
    Kimi K2: Open Agentic Intelligence
    Kimi Team (including Longguang Zhong)
    arXiv preprint arXiv:2507.20534, 2025.
    PDF Tech Report

  4. Tech Report
    FuseRL: Dense Preference Optimization for Heterogeneous Model Fusion
    Longguang Zhong, Fanqi Wan, Ziyi Yang, Guosheng Liang, Tianyuan Shi, Xiaojun Quan* (*Corresponding authors)
    arXiv preprint arXiv:2504.06562, 2025.
    PDF Tech Report

  5. ACL
    Mutual-Taught for Co-adapting Policy and Reward Models
    Tianyuan Shi, Canbin Huang, Fanqi Wan, Longguang Zhong, Ziyi Yang, Wei Shen, Xiaojun Quan*, Ming Yan (*Corresponding authors)
    The 63rd Annual Meeting of the Association for Computational Linguistics (ACL), 2025.
    PDF Main

  6. ACL
    BlockPruner: Fine-grained Pruning for Large Language Models
    Longguang Zhong, Fanqi Wan, Ruijun Chen, Xiaojun Quan*, Liangzhi Li (*Corresponding authors)
    The 63rd Annual Meeting of the Association for Computational Linguistics (ACL), 2025.
    PDF Code Findings

  7. EMNLP
    FUSECHAT: Knowledge Fusion of Chat Models
    Fanqi Wan, Longguang Zhong, Ziyi Yang, Ruijun Chen, Xiaojun Quan* (*Corresponding authors)
    The 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2025.
    PDF Code Main

  8. EMNLP
    ThinkSwitcher: When to Think Hard, When to Think Fast
    Guosheng Liang, Longguang Zhong, Ziyi Yang, Xiaojun Quan* (*Corresponding authors)
    The 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2025.
    PDF Findings

  9. ICLR
    WEIGHTED-REWARD PREFERENCE OPTIMIZATION FOR IMPLICIT MODEL FUSION
    Ziyi Yang, Fanqi Wan, Longguang Zhong, Tianyuan Shi, Xiaojun Quan* (*Corresponding authors)
    The Thirteenth International Conference on Learning Representations (ICLR), 2025.
    PDF Code BibTex Poster

  10. ICLR Workshop
    FuseChat-3.0: Preference Optimization Meets Heterogeneous Model Fusion
    Ziyi Yang, Fanqi Wan, Longguang Zhong, Canbin Huang, Guosheng Liang, Xiaojun Quan* (*Corresponding authors)
    ICLR 2025 First Workshop on Open Science for Foundation Models (ICLR WorkShop), 2025.
    PDF Code BibTex Poster