Elamur Research⌕
☰
home / papers / 2026 / GLM-5: from Vibe Coding to Agentic Engineering

GLM-5: from Vibe Coding to Agentic Engineering

Summary

GLM-5, from Zhipu AI and Tsinghua University, is an open-weight flagship model. It targets two obstacles to using LLMs as autonomous software engineers: high compute cost and weak reliability on long, real-world tasks. The model is a Mixture-of-Experts with 744B total and 40B active parameters. Its base was trained on 28.5T tokens, and a mid-training stage gradually extended the context window from 4K to 200K using long-context agentic data. DeepSeek Sparse Attention (DSA) was added through continued pre-training from a dense checkpoint, which cuts long-context cost without retraining from scratch. Post-training runs three RL stages in order (reasoning, agentic, general), with on-policy cross-stage distillation to limit forgetting. A new asynchronous RL system separates rollout generation from training, and new agent RL algorithms support learning from long-horizon interactions. The authors report that GLM-5 roughly matches Claude Opus 4.5 and GPT-5.2 on agentic, coding and reasoning benchmarks. This suggests open models are approaching proprietary systems for practical agent work.

Key points

  • Architecture: 256 experts and 80 layers, about double GLM-4.5's 355B total parameters.
    • MLA with a "Muon Split" optimizer tweak (head-wise orthogonalization) closes the gap with GQA-8.
    • MLA-256 uses larger heads and fewer of them to make decoding cheaper.
  • Speculative decoding: three MTP layers share parameters during training. This gives an accept length of 2.76, compared with 2.55 for DeepSeek-V3.2.
  • Leaderboards:
    • Reaches a score of 50 on version 4.0 of the Artificial Analysis Intelligence Index, an improvement over GLM-4.7's 42. It is the first open-weights model to reach that score.
    • Ranks #1 among open models in both LMArena Text and Code.
  • Long-horizon tasks:
    • Ends Vending-Bench 2 with a $4,432 balance, the best among open models.
    • Shows a large gain over GLM-4.7 on the internal CC-Bench-V2.
    • Averages about 20% better than GLM-4.7 across 8 benchmarks, including SWE-bench Verified and Terminal-Bench 2.0.
  • Hardware support: optimized for seven Chinese chip platforms, including Huawei Ascend, Cambricon and Kunlunxin.
  • Release: code and weights are available at github.com/zai-org/GLM-5.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Natural language processing

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2602.15763 · arXiv 2602.15763

@misc{zeng2026glm,
  title = {{GLM-5: from Vibe Coding to Agentic Engineering}},
  author = {Zeng, Aohan and Lv, Xin and Hou, Zhenyu and Du, Zhengxiao and Zheng, Qinkai and Chen, Bin and Yin, Da and Ge, Chendi and Huang, Chenghua and Xie, Chengxing and Zhu, Chenzheng and Yin, Congfeng and Wang, Cunxiang and Pan, Gengzheng and Zeng, Hao and Zhang, Haoke and Wang, Haoran and Chen, Huilong and Zhang, Jiajie and Jiao, Jian and Guo, Jiaqi and Wang, Jingsen and Du, Jingzhao and Wu, Jinzhu and Wang, Kedong and Li, Lei and Fan, Lin and Zhong, Lucen and Liu, Mingdao and Zhao, Mingming and Du, Pengfan and Dong, Qian and Lu, Rui and Shuang-Li and Cao, Shulin and Liu, Song and Jiang, Ting and Chen, Xiaodong and Zhang, Xiaohan and Huang, Xuancheng and Dong, Xuezhen and Xu, Yabo and Wei, Yao and An, Yifan and Niu, Yilin and Zhu, Yitong and Wen, Yuanhao and Cen, Yukuo and Bai, Yushi and Qiao, Zhongpei and Wang, Zihan and Wang, Zikang and Zhu, Zilin and Liu, Ziqiang and Li, Zixuan and Wang, Bojie and Wen, Bosi and Huang, Can and Cai, Changpeng and Yu, Chao and Li, Chen and Hu, Chengwei and Zhang, Chenhui and Zhang, Dan and Lin, Daoyan and Yang, Dayong and Wang, Di and Ai, Ding and Zhu, Erle and Yi, Fangzhou and Chen, Feiyu and Wen, Guohong and Sun, Hailong and Zhao, Haisha and Hu, Haiyi and Zhang, Hanchen and Liu, Hanrui and Zhang, Hanyu and Peng, Hao and Tai, Hao and Zhang, Haobo and Liu, He and Wang, Hongwei and Yan, Hongxi and Ge, Hongyu and Liu, Huan and Chu, Huanpeng and Zhao, Jia'ni and Wang, Jiachen and Zhao, Jiajing and Ren, Jiamin and Wang, Jiapeng and Zhang, Jiaxin and Gui, Jiayi and Zhao, Jiayue and Li, Jijie and An, Jing and Li, Jing and Yuan, Jingwei and Du, Jinhua and Liu, Jinxin and Zhi, Junkai and Duan, Junwen and Zhou, Kaiyue and Wei, Kangjian and Wang, Ke and Luo, Keyun and Zhang, Laiqiang and Sha, Leigang and Xu, Liang and Wu, Lindong and Ding, Lintao and Chen, Lu and Li, Minghao and Lin, Nianyi and Ta, Pan and Zou, Qiang and Song, Rongjun and Yang, Ruiqi and Tu, Shangqing and Yang, Shangtong and Wu, Shaoxiang and Zhang, Shengyan and Li, Shijie and Li, Shuang and Fan, Shuyi and Qin, Wei and Tian, Wei and Zhang, Weining and Yu, Wenbo and Liang, Wenjie and Kuang, Xiang and Cheng, Xiangmeng and Li, Xiangyang and Yan, Xiaoquan and Hu, Xiaowei and Ling, Xiaoying and Fan, Xing and Xia, Xingye and Zhang, Xinyuan and Zhang, Xinze and Pan, Xirui and Zou, Xu and Zhang, Xunkai and Liu, Yadi and Wu, Yandong and Li, Yanfu and Wang, Yidong and Zhu, Yifan and Tan, Yijun and Zhou, Yilin and Pan, Yiming and Zhang, Ying and Su, Yinpei and Geng, Yipeng and Yan, Yong and Tan, Yonglin and Bi, Yuean and Shen, Yuhan and Yang, Yuhao and Li, Yujiang and Liu, Yunan and Wang, Yunqing and Li, Yuntao and Wu, Yurong and Zhang, Yutao and Duan, Yuxi and Zhang, Yuxuan and Liu, Zezhen and Jiang, Zhengtao and Yan, Zhenhe and Zhang, Zheyu and Wei, Zhixiang and Chen, Zhuo and Feng, Zhuoer and Yao, Zijun and Chai, Ziwei and Wang, Ziyuan and Zhang, Zuzhou and Xu, Bin and Huang, Minlie and Wang, Hongning and Li, Juanzi and Dong, Yuxiao and Tang, Jie},
  year = {2026},
  doi = {10.48550/arxiv.2602.15763},
  eprint = {2602.15763},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2602.15763}
}

References

63 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Anthropic. System card: Claude opus 4.5, 2025. find
  2. S. Ashkboos, A. Mohtashami, M. L. Croci, B. Li, P. Cameron, M. Jaggi, D. Alistarh, T. Hoefler, and J. Hensman. Quarot: Outlier-free 4-bit inference in rotated llms, 2024. find our card
  3. A. Backlund and L. Petersson. Vending-bench: A benchmark for long-term coherence of autonomous agents. arXiv preprint arXiv:2502.15840, 2025. — [arXiv:2502.15840](https://arxiv.org/abs/2502.15840) find
  4. I. Badertdinov, A. Golubev, M. Nekrashevich, A. Shevtsov, S. Karasik, A. Andriushchenko, M. Trofimova, D. Litvintseva, and B. Yangel. Swe-rebench: An automated pipeline for task collection and decontaminated evaluation of software engineering agents. arXiv preprint arXiv:2505.20411, 2025. — [arXiv:2505.20411](https://arxiv.org/abs/2505.20411) find
  5. Y. Bai, S. Tu, J. Zhang, H. Peng, X. Wang, X. Lv, S. Cao, J. Xu, L. Hou, Y. Dong, J. Tang, and J. Li. LongBench v2: Towards deeper understanding and reasoning on realistic long-context multitasks. In ACL’25, pages 3639–3664, 2025. find our card
  6. C. Bandi, B. Hertzberg, G. Boo, T. Polakam, J. Da, S. Hassaan, M. Sharma, A. Park, E. Hernandez, D. Rambado, et al. Mcp-atlas: A large-scale benchmark for tool-use competency with real mcp servers. arXiv preprint arXiv:2602.00933, 2026. — [arXiv:2602.00933](https://arxiv.org/abs/2602.00933) find
  7. V. Barres, H. Dong, S. Ray, X. Si, and K. Narasimhan. τ 2 -bench: Evaluating conversational agents in a dual-control environment. arXiv preprint arXiv:2506.07982, 2025. find
  8. G. DeepMind. Gemini 3 pro model card, 2025. find
  9. DeepSeek-AI, A. Liu, A. Mei, and et al. Deepseek-v3.2: Pushing the frontier of open large language models, 2025. find
  10. W. Du, S. Toshniwal, B. Kisacanin, S. Mahdavi, I. Moshkov, G. Armstrong, S. Ge, E. Minasyan, F. Chen, and I. Gitman. Nemotron-math: Efficient long-context distillation of mathematical reasoning from multi-mode supervision. arXiv preprint arXiv:2512.15489, 2025. find
  11. C. Gao, X. Wu, Z. Lin, D. Zhang, and S. Hu. Nextlong: Toward effective long-context training without long documents, 2025. find
  12. H. Ge, J. Feng, Q. Huang, F. Fu, X. Nie, L. Zuo, H. Lin, B. Cui, and X. Liu. Bytescale: Efficient scaling of llm training with a 2048k context length on more than 12,000 gpus. arXiv preprint arXiv:2502.21231, 2025. find
  13. F. Gloeckle, B. Y. Idrissi, B. Rozière, D. Lopez-Paz, and G. Synnaeve. Better & faster large language models via multi-token prediction. arXiv preprint arXiv:2404.19737, 2024. find
  14. Y. Gu, L. Dong, F. Wei, and M. Huang. Minillm: Knowledge distillation of large language models. In ICLR’23, 2025. find
  15. Y. Gu, Q. Hu, S. Yang, H. Xi, J. Chen, S. Han, and H. Cai. Jet-nemotron: Efficient language model with post neural architecture search. arXiv preprint arXiv:2508.15884, 2025. — [arXiv:2508.15884](https://arxiv.org/abs/2508.15884) find
  16. Y. He, S. Li, J. Liu, Y. Tan, W. Wang, H. Huang, X. Bu, H. Guo, C. Hu, B. Zheng, Z. Lin, X. Liu, D. Sun, S. Lin, Z. Zheng, X. Zhu, W. Su, and B. Zheng. Chinese simpleqa: A chinese factuality evaluation for large language models, 2024. find
  17. C.-P. Hsieh, S. Sun, S. Kriman, S. Acharya, D. Rekesh, F. Jia, and B. Ginsburg. Ruler: What’s the real context size of your long-context language models? In COLM’24, 2024. find our card
  18. J. Jia, Z. Chen, X. Wu, C. Gao, Z. Lin, D. Zhang, S. Hu, and B. Guo. Entropylong: Effective long-context training via predictive uncertainty, 2025. find
  19. C. E. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. Narasimhan. Swe-bench: Can language models resolve real-world github issues? arXiv preprint arXiv:2310.06770, 2023. find our card
  20. Y. Leviathan, M. Kalman, and Y. Matias. Fast inference from transformers via speculative decoding. In ICML’23, pages 19274–19286, 2023. find
  21. J. Li, A. Fang, G. Smyrnis, M. Ivgi, and et al. Datacomp-lm: In search of the next generation of training sets for language models, 2025. find our card
  22. J. Li, W. Zhao, J. Zhao, W. Zeng, H. Wu, X. Wang, R. Ge, Y. Cao, Y. Huang, W. Liu, et al. The tool decathlon: Benchmarking language agents for diverse, realistic, and long-horizon task execution. arXiv preprint arXiv:2510.25726, 2025. find
  23. R. Li, J. Fu, B.-W. Zhang, T. Huang, Z. Sun, C. Lyu, G. Liu, Z. Jin, and G. Li. Taco: Topics in algorithmic code generation dataset. arXiv preprint arXiv:2312.14852, 2023. find
  24. A. Liu, B. Feng, B. Wang, B. Wang, B. Liu, C. Zhao, C. Dengr, C. Ruan, D. Dai, D. Guo, et al. Deepseek-v2: A strong, economical, and efficient mixture-of-experts language model. arXiv preprint arXiv:2405.04434, 2024. find
  25. A. Liu, B. Feng, B. Xue, B. Wang, B. Wu, C. Lu, C. Zhao, C. Deng, C. Zhang, C. Ruan, et al. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437, 2024. find
  26. A. Liu, A. Mei, B. Lin, B. Xue, B. Wang, B. Xu, B. Wu, B. Zhang, C. Lin, C. Dong, et al. Deepseek-v3. 2: Pushing the frontier of open large language models. arXiv preprint arXiv:2512.02556, 2025. find our card
  27. J. Liu, J. Le Tian, V. Daita, Y. Wei, Y. Ding, Y. K. Wang, J. Yang, and L. ZHANG. Repoqa: Evaluating long context code understanding. In First Workshop on Long-Context Foundation Models@ ICML 2024. find
  28. K. Lu and T. M. Lab. On-policy distillation. Thinking Machines Lab: Connectionism, 2025. https://thinkingmachines.ai/blog/on-policy-distillation. ↗
  29. M.-T. Luong, D. Hwang, H. H. Nguyen, G. Ghiasi, Y. Chervonyi, I. Seo, J. Kim, G. Bingham, J. Lee, S. Mishra, et al. Towards robust mathematical reasoning. In EMNLP’25, pages 35406– 35430, 2025. find
  30. I. Moshkov, D. Hanley, I. Sorokin, S. Toshniwal, C. Henkel, B. Schifferer, W. Du, and I. Gitman. Aimo-2 winning solution: Building state-of-the-art mathematical reasoning models with openmathreasoning dataset. arXiv preprint arXiv:2504.16891, 2025. find
  31. D. Narayanan, M. Shoeybi, J. Casper, P. LeGresley, M. Patwary, V. A. Korthikanti, D. Vainbrand, P. Kashinkunti, J. Bernauer, B. Catanzaro, A. Phanishayee, and M. Zaharia. Efficient large-scale language model training on gpu clusters using megatron-lm, 2021. find
  32. OpenAI. Introducing gpt 5.2, 2025. find
  33. T. Patwardhan, R. Dias, E. Proehl, G. Kim, M. Wang, O. Watkins, S. P. Fishman, M. Aljubeh, P. Thacker, L. Fauconnet, et al. Gdpval: Evaluating ai model performance on real-world economically valuable tasks. arXiv preprint arXiv:2510.04374, 2025. find
  34. L. Phan, A. Gatti, Z. Han, N. Li, J. Hu, H. Zhang, C. B. C. Zhang, M. Shaaban, J. Ling, S. Shi, et al. Humanity’s last exam. arXiv preprint arXiv:2501.14249, 2025. find our card
  35. Prime Intellect. Synthetic-2 release: Four million collaboratively generated reasoning traces, 2025. Blog post. find
  36. V. Pyatkin, S. Malik, V. Graf, H. Ivison, S. Huang, P. Dasigi, N. Lambert, and H. Hajishirzi. Generalizing verifiable instruction following, 2025. find
  37. P. Qi, X. Wan, G. Huang, and M. Lin. Zero bubble pipeline parallelism. arXiv preprint arXiv:2401.10241, 2023. find
  38. S. Rajbhandari, J. Rasley, O. Ruwase, and Y. He. Zero: Memory optimizations toward training trillion parameter models, 2020. find
  39. D. Rein, B. L. Hou, A. C. Stickland, J. Petty, R. Y. Pang, J. Dirani, J. Michael, and S. R. Bowman. Gpqa: A graduate-level google-proof q&a benchmark. In CoLM’24, 2024. find our card
  40. Z. Shao, P. Wang, Q. Zhu, R. Xu, J. Song, X. Bi, H. Zhang, M. Zhang, Y. Li, Y. Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300, 2024. find our card
  41. V. Sirdeshmukh, K. Deshpande, J. Mols, L. Jin, E.-Y. Cardona, D. Lee, J. Kritz, W. Primack, S. Yue, and C. Xing. Multichallenge: A realistic multi-turn conversation evaluation benchmark challenging to frontier llms, 2025. find
  42. H. F. Team. Harbor: A framework for evaluating and optimizing agents and models in container environments., 2026. find
  43. K. Team, T. Bai, Y. Bai, Y. Bao, S. Cai, Y. Cao, Y. Charles, H. Che, C. Chen, G. Chen, et al. Kimi k2. 5: Visual agentic intelligence. arXiv preprint arXiv:2602.02276, 2026. find
  44. L. Team, A. Shen, B. Li, B. Hu, B. Jing, C. Chen, C. Huang, C. Zhang, C. Yang, C. Lin, et al. Every step evolves: Scaling reinforcement learning for trillion-scale thinking model. arXiv preprint arXiv:2510.18855, 2025. — [arXiv:2510.18855](https://arxiv.org/abs/2510.18855) find
  45. T. T.-B. Team. Terminal-bench: A benchmark for ai agents in terminal environments, Apr 2025. find
  46. Y. Tian, C. Wang, Z. Liu, H. Huang, W. Yu, D. Song, J. Tang, and Y. Guo. Beyond literal mapping: Benchmarking and improving non-literal translation evaluation, 2026. find our card
  47. Y. Wang, S. Wang, S. Zhu, F. Fu, X. Liu, X. Xiao, H. Li, J. Li, F. Wu, and B. Cui. Flexsp: Accelerating large language model training via flexible sequence parallelism. In ASPLOS’25, pages 421–436, 2025. find
  48. Z. Wang, T. Shi, J. He, M. Cai, J. Zhang, and D. Song. Cybergym: Evaluating ai agents’ cybersecurity capabilities with real-world vulnerabilities at scale. arXiv preprint arXiv:2506.02548, 2025. — [arXiv:2506.02548](https://arxiv.org/abs/2506.02548) find
  49. J. Wei, N. Karina, H. W. Chung, Y. J. Jiao, S. Papay, A. Glaese, J. Schulman, and W. Fedus. Measuring short-form factuality in large language models, 2024. find our card
  50. J. Wei, Z. Sun, S. Papay, S. McKinney, J. Han, I. Fulford, H. W. Chung, A. T. Passos, W. Fedus, and A. Glaese. Browsecomp: A simple yet challenging benchmark for browsing agents. arXiv preprint arXiv:2504.12516, 2025. find
  51. L.-C. Xiaomi. Mimo-v2-flash technical report, 2026. find
  52. A. Yang, A. Li, B. Yang, B. Zhang, and et al. Qwen3 technical report. arXiv preprint arXiv:2505.09388, 2025. find our card
  53. J. Yang, K. Lieret, C. E. Jimenez, A. Wettig, K. Khandpur, Y. Zhang, B. Hui, O. Press, L. Schmidt, and D. Yang. Swe-smith: Scaling data for software engineering agents. arXiv preprint arXiv:2504.21798, 2025. — [arXiv:2504.21798](https://arxiv.org/abs/2504.21798) find
  54. S. Yang, J. Kautz, and A. Hatamizadeh. Gated delta networks: Improving mamba2 with delta rule. In ICLR’24, 2024. find
  55. S. Yao, N. Shinn, P. Razavi, and K. Narasimhan. tau-bench: A benchmark for tool-agent-user interaction in real-world domains. arXiv preprint arXiv:2406.12045, 2024. find our card
  56. H. Yen, T. Gao, M. Hou, K. Ding, D. Fleischer, P. Izsak, M. Wasserblat, and D. Chen. Helmet: How to evaluate long-context language models effectively and thoroughly. arXiv preprint arXiv:2410.02694, 2024. find
  57. Q. Yu, Z. Zhang, R. Zhu, Y. Yuan, X. Zuo, Y. Yue, W. Dai, T. Fan, G. Liu, L. Liu, et al. Dapo: An open-source llm reinforcement learning system at scale. arXiv preprint arXiv:2503.14476, 2025. find our card
  58. T. Yuan, Y. Liu, X. Ye, S. Zhang, J. Tan, B. Chen, C. Song, and D. Zhang. Accelerating the training of large language models using efficient activation rematerialization and optimal hybrid parallelism. In USENIX ATC’24, pages 545–561, 2024. find
  59. L. Zhang, S. He, C. Zhang, Y. Kang, B. Li, C. Xie, J. Wang, M. Wang, Y. Huang, S. Fu, E. Nallipogu, Q. Lin, Y. Dang, S. Rajmohan, and D. Zhang. Swe-bench goes live! arXiv preprint arXiv:2505.23419, 2025. — [arXiv:2505.23419](https://arxiv.org/abs/2505.23419) find
  60. C. Zhao, C. Deng, C. Ruan, D. Dai, H. Gao, J. Li, L. Zhang, P. Huang, S. Zhou, S. Ma, et al. Insights into deepseek-v3: Scaling challenges and reflections on hardware for ai architectures. In ISCA’25, pages 1731–1745, 2025. find
  61. X. Zhao, Y. Liu, K. Xu, J. Guo, Z. Wang, Y. Sun, X. Kong, Q. Cao, L. Jiang, Z. Wen, Z. Zhang, and J. Zhou. Small leak can sink a great ship–boost rl training on moe with icepop!, Sep 2025. find
  62. C. Zheng, S. Liu, M. Li, X.-H. Chen, B. Yu, C. Gao, K. Dang, Y. Liu, R. Men, A. Yang, et al. Group sequence policy optimization. arXiv preprint arXiv:2507.18071, 2025. find our card
  63. P. Zhou, B. Leon, X. Ying, C. Zhang, Y. Shao, Q. Ye, D. Chong, Z. Jin, C. Xie, M. Cao, et al. Browsecomp-zh: Benchmarking web browsing ability of large language models in chinese. arXiv preprint arXiv:2504.19314, 2025. — [arXiv:2504.19314](https://arxiv.org/abs/2504.19314) find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/glm-2026-5-glm-5-vibe-coding/