ACL 2025 Conference paper

EscapeBench: Towards Advancing Creative Intelligence of Language Model Agents

Room-escape environments for studying creative problem-solving, implicit goals, and unexpected uses of familiar tools.

Cheng Qian, Peixuan Han, Qinyu Luo, Bingxiang He, Xiusi Chen, Yuji Zhang, Hongyi Du, Jiarui Yao, Xiaocheng Yang, Denghui Zhang, Yunzhu Li, Heng Ji

EscapeAgent design with foresight, action, and reflection modules. — Fig. 4 — Original EscapeAgent design.Original source ↗

01 / The question

How do agents solve unfamiliar problems when neither the goal nor the right tool use is obvious?

02 / The approach

EscapeBench

EscapeBench places agents in interactive room-escape games. EscapeAgent adds foresight for exploring possible tool uses and reflection for revisiting unresolved tasks, making creative adaptation part of the agent loop.

Read the full paper

03 / Citation

Cite this work

BibTeX

@inproceedings{qian-etal-2025-escapebench,
    title = "{E}scape{B}ench: Towards Advancing Creative Intelligence of Language Model Agents",
    author = "Qian, Cheng  and
      Han, Peixuan  and
      Luo, Qinyu  and
      He, Bingxiang  and
      Chen, Xiusi  and
      Zhang, Yuji  and
      Du, Hongyi  and
      Yao, Jiarui  and
      Yang, Xiaocheng  and
      Zhang, Denghui  and
      Li, Yunzhu  and
      Ji, Heng",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.acl-long.39/",
    doi = "10.18653/v1/2025.acl-long.39",
    pages = "798--820",
    ISBN = "979-8-89176-251-0"
}

Select the citation to copy it manually.

04 / Versions & records

Publication record

Earlier or alternate titles

EscapeBench: Pushing Language Models to Think Outside the Box

2 Google Scholar records

READ HERE · 23 PAGES

Full paper

Document

Loading document…

Scroll through every page. Select text to copy, or use search to find a passage.

More research

View all

How do agents solve unfamiliar problems when neither the goal nor the right tool use is obvious?

EscapeBench

Cite this work

Publication record

Full paper

More research

World Model as Tool↗

One-Shot On-Policy Distillation↗

World Model as Tool

One-Shot On-Policy Distillation