Elamur Research⌕
☰
home / papers / 2026 / GOAT: A Training Framework for Goal-Oriented Agent with Tools

GOAT: A Training Framework for Goal-Oriented Agent with Tools

Paper · 2026 · Hyunji Min, Sangwon Jung, Junyoung Sung et al.

Summary

The paper tackles goal-oriented tool use, where a user states only a high-level aim and an LLM agent must break it into steps, choose APIs, and chain calls so that earlier outputs supply later arguments. Annotating such interdependent call sequences by hand is costly, so prior work mostly evaluates models zero-shot, and small open-source models largely fail.

The authors propose GOAT, a framework that creates training data automatically from API documentation. It builds an API dependency graph, prunes it to keep only invocation links that actually work, and samples connected subgraphs. It then uses a "call-first" approach: it runs the API calls first, and only afterwards writes a goal-level user query and final answer that fit the executed sequence. The LLM and the retriever are fine-tuned jointly on this data.

GOAT-trained open-source models reach state-of-the-art results on RestBench, API-Bank, and the new GOATBench, and sometimes beat closed-source models. This offers a practical way to build capable on-premise agents without manual labeling.

Key points

  • Data pipeline. Each synthetic example contains:

    • a goal-oriented query,
    • a set of call units (sub-query, API call, output),
    • a final response.

    The examples come from executed real APIs, not from instruction-first generation, which the authors argue produces data biased toward easy cases.

  • RestBench. Untuned open-source baselines score near zero. After GOAT training, Llama2-13B and Vicuna-13B surpass RestGPT, the previous open-source state of the art.

  • API-Bank. GOAT improves Llama-7B over zero-shot and over the original API-Bank training setup, although ROUGE drops slightly.

  • GOATBench. This new large-scale benchmark was built with the GOAT pipeline plus lightweight human verification. GOAT gains hold under both ReACT and baseline prompting, and GOAT outperforms models fine-tuned on ToolLLM data.

  • Ablation and limitations. On RestBench with a Llama-2-13B backbone, call-first data consistently beats instruction-first data. The method is mainly for in-domain training, and its output depends on the quality of the API documentation. Code and data are released at github.com/KU-MIIL/GOAT.

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

 Institutions

🎓Korea UniversitySeoul · South Korea26 works · 9,490 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

Cite

Cite as Hyunji Min, Sangwon Jung, Junyoung Sung, Dosung Lee, Leekyeung Han and Paul Hongsuck Seo. 2026. GOAT: A Training Framework for Goal-Oriented Agent with Tools. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1150

@inproceedings{min2026goat,
  title = {{GOAT: A Training Framework for Goal-Oriented Agent with Tools}},
  author = {Hyunji Min, Sangwon Jung, Junyoung Sung, Dosung Lee, Leekyeung Han and Paul Hongsuck Seo},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1150},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1150}
}

References

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

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  2. AI@Meta. 2024. Llama 3 model card. ArXiv. find
  3. Leekyeung Han, Hyunji Min, Gyeom Hwangbo, Jonghyun Choi, and Paul Hongsuck Seo. 2025. Di-alnav: Multi-turn dialog navigation with a remote guide. arXiv preprint arXiv:2509.12894. — [arXiv:2509.12894](https://arxiv.org/abs/2509.12894) find
  4. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. ArXiv.
  5. Xu Huang, Weiwen Liu, Xiaolong Chen, Xingmei Wang, Hao Wang, Defu Lian, Yasheng Wang, Ruiming Tang, and Enhong Chen. 2024. Understanding the planning of llm agents: A survey. Preprint, arXiv:2402.02716. find
  6. Physical Intelligence, Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, and 1 others. 2025. π 0.5 : A Vision-Language-Action Model with Open-World Generalization. arXiv preprint arXiv:2504.16054. find
  7. Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Foster, Grace Lam, Pannag Sanketi, Quan Vuong, Thomas Kollar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn. 2024. Openvla: An open-source… find
  8. Minghao Li, Feifan Song, Bowen Yu, Haiyang Yu, Zhoujun Li, Fei Huang, and Yongbin Li. 2023. Api-bank: A benchmark for tool-augmented llms. Preprint, arXiv:2304.08244. — [arXiv:2304.08244](https://arxiv.org/abs/2304.08244) find
  9. Xiao Liu, Hao Yu, Hanchen Zhang, Yifan Xu, Xuanyu Lei, Hanyu Lai, Yu Gu, Hangliang Ding, Kaiwen Men, Kejuan Yang, Shudan Zhang, Xiang Deng, Aohan Zeng, Zhengxiao Du, Chenhui Zhang, Sheng Shen, Tianjun Zhang, Yu Su, Huan Sun, and 3 others. 2024. Agentbench: Evaluating llms as agents. In International…
  10. Shishir G. Patil, Tianjun Zhang, Xin Wang, and Joseph E. Gonzalez. 2023. Gorilla: Large language model connected with massive apis. ArXiv, abs/2305.15334.
  11. Yujia Qin, Shengding Hu, Yankai Lin, Weize Chen, Ning Ding, Ganqu Cui, Zheni Zeng, Yufei Huang, Chaojun Xiao, Chi Han, Yi Ren Fung, Yusheng Su, Huadong Wang, Cheng Qian, Runchu Tian, Kunlun Zhu, Shihao Liang, Xingyu Shen, Bokai Xu, and 22 others. 2024. Tool learning with foundation models.
  12. Yujia Qin, Shi Liang, Yining Ye, Kunlun Zhu, Lan Yan, Ya-Ting Lu, Yankai Lin, Xin Cong, Xiangru Tang, Bill Qian, Sihan Zhao, Runchu Tian, Ruobing Xie, Jie Zhou, M. Gerstein, Dahai Li, Zhiyuan Liu, and Maosong Sun. 2023. Toolllm: Facilitating large language models to master 16000+ real-world apis. In…
  13. Yujia Qin, Yining Ye, Junjie Fang, Haoming Wang, Shihao Liang, Shizuo Tian, Junda Zhang, Jiahao Li, Yunxin Li, Shijue Huang, and 1 others. 2025. Ui-tars: Pioneering automated gui interaction with native agents. arXiv preprint arXiv:2501.12326.
  14. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.
  15. Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, M. Lomeli, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. 2023. Toolformer: Language models can teach themselves to use tools. In Neural Information Processing Systems.
  16. Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. 2023a. Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face. Advances in Neural Information Processing Systems, 36:38154–38180.
  17. Yongliang Shen, Kaitao Song, Xu Tan, Wenqi Zhang, Kan Ren, Siyu Yuan, Weiming Lu, Dongsheng Li, and Y. Zhuang. 2023b. Taskbench: Benchmarking large language models for task automation. In arXiv.org. find
  18. Jeonghoon Shim, Gyuhyeon Seo, Cheongsu Lim, and Yohan Jo. 2025. Tooldial: Multi-turn dialogue generation method for tool-augmented language models. Preprint, arXiv:2503.00564. — [arXiv:2503.00564](https://arxiv.org/abs/2503.00564) find
  19. Yifan Song, Weimin Xiong, Dawei Zhu, Wenhao Wu, Han Qian, Mingbo Song, Hailiang Huang, Cheng Li, Ke Wang, Rong Yao, Ye Tian, and Sujian Li. 2023. Restgpt: Connecting large language models with real-world restful apis. In arXiv.org.
  20. Dídac Surís, Sachit Menon, and Carl Vondrick. 2023. Vipergpt: Visual inference via python execution for reasoning. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 11888– 11898. models with 3000 simulated cases. arXiv preprint arXiv:2306.05301.
  21. Pei Wang, Yanan Wu, Noah Wang, Jiaheng Liu, Xiaoshuai Song, ZY Peng, Ken Deng, Chenchen Zhang, Junran Peng, Ge Zhang, and 1 others. 2024a. Mtu-bench: A multi-granularity tool-use benchmark for large language models. In The Thirteenth International Conference on Learning Representations. find
  22. Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A. Smith, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. Self-instruct: Aligning language models with self-generated instructions. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long… find
  23. Zezhong Wang, Xingshan Zeng, Weiwen Liu, Liangyou Li, Yasheng Wang, Lifeng Shang, Xin Jiang, Qun Liu, and Kam-Fai Wong. 2024b. Toolflow: Boosting llm tool-calling through natural and coherent dialogue synthesis. In arXiv.org. find
  24. Rui Yang, Lin Song, Yanwei Li, Sijie Zhao, Yixiao Ge, Xiu Li, and Ying Shan. 2023. Gpt4tools: Teaching large language model to use tools via self-instruction. In Advances in Neural Information Processing Systems 36 (NeurIPS 2023 Poster). Poster. find
  25. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing reasoning and acting in language models. In International Conference on Learning Representations (ICLR).
  26. Fan Yin, Zifeng Wang, I-Hung Hsu, Jun Yan, Ke Jiang, Yanfei Chen, Jindong Gu, Long T. Le, Kai-Wei Chang, Chen-Yu Lee, Hamid Palangi, and Tomas Pfister. 2025. Magnet: Multi-turn tool-use data synthesis and distillation via graph translation. Preprint, arXiv:2503.07826. find
  27. Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, Yifan Du, Chen Yang, Yushuo Chen, Zhipeng Chen, Jinhao Jiang, Ruiyang Ren, Yifan Li, Xinyu Tang, Zikang Liu, and 3 others. 2025. A survey of large language models. Prepri…
  28. Yuchen Zhuang, Yue Yu, Kuan Wang, Haotian Sun, and Chao Zhang. 2023. Toolqa: A dataset for llm question answering with external tools. In Advances in Neural Information Processing Systems 36 (NeurIPS 2023 Poster). Poster.
  29. Qiaoyu Tang, Ziliang Deng, Hongyu Lin, Xianpei Han, Qiao Liang, Boxi Cao, and Le Sun. 2023. Toolalpaca: Generalized tool learning for language find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/min-2026-goat-training-framework-goal-oriented/