STT-Arena: A More Realistic Environment for Tool-Using with Spatio-Temporal Dynamics
Tingfeng Hui, Hao Xu, Pengyu Zhu, Hongsheng Xin, Kun Zhan, Sen Su, Chunxiao Liu, Ning Miao · 2026
Read STT-Arena: A More Realistic Environment for Tool-Using with Spatio-Temporal Dynamics by Tingfeng Hui, Hao Xu, Pengyu Zhu, Hongsheng Xin, Kun Zhan, Sen Su, Chunxiao Liu, Ning Miao (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-18548.txt.
Introduction
Tingfeng Hui and seven co-authors present STT-Arena, a 2026 benchmark of 227 interactive tasks for tool-using models under spatio-temporal dynamics. Injected triggers abruptly invalidate ongoing plans, forcing models to detect state shifts and replan. Even the best proprietary models score under 40 percent overall. Failure analysis finds three recurring error modes, and an iterative trajectory refinement method plus online RL yields STT-Agent-4B, which beats frontier models on the benchmark.
Catalog record
| CREATOR | Tingfeng Hui, Hao Xu, Pengyu Zhu, Hongsheng Xin, Kun Zhan, Sen Su, Chunxiao Liu, Ning Miao |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 134448 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.18548