WebExplorer : Explore and Evolve for Training Long-Horizon Web Agents
Junteng Liu 1* Yunji Li 2* Chi Zhang 2 Jingyang Li 2 Aili Chen 2 Ke Ji 2 Weiyu Cheng 2 Zijia Wu 2 Chengyu Du 2 Qidi Xu 2 Jiayuan Song 2 Zhengmao Zhu 2 Wenhu Chen 3 Pengyu Zhao 2 Junxian He 1 1 The Hong Kong University of Science and Technology 2 MiniMax 3 University of Waterloo [email protected] jun · 2025
Read WebExplorer : Explore and Evolve for Training Long-Horizon Web Agents by Junteng Liu 1* Yunji Li 2* Chi Zhang 2 Jingyang Li 2 Aili Chen 2 Ke Ji 2 Weiyu Cheng 2 Zijia Wu 2 Chengyu Du 2 Qidi Xu 2 Jiayuan Song 2 Zhengmao Zhu 2 Wenhu Chen 3 Pengyu Zhao 2 Junxian He 1 1 The Hong Kong University of Science and Technology 2 MiniMax 3 University of Waterloo [email protected] jun (2025) 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-2509-06501.txt.
Description
Training web agents that can hunt through many pages for one answer is the subject of a 2025 artificial intelligence paper led by Junteng Liu, with fourteen co-authors. Language models increasingly act as agents that browse the web, but hard, long searches expose weak spots. Good training examples for such searches are scarce.
The authors build data by letting a model explore and then evolving questions from long to short forms, creating query-answer pairs that demand many steps. Training on this data, first with supervised learning and then reinforcement learning, yields an eight-billion-parameter agent. It handles very long contexts and up to a hundred tool calls in one task. On several search benchmarks it beats much larger rivals, and it also generalizes well to a broad knowledge test.
Catalog record
| CREATOR | Junteng Liu 1* Yunji Li 2* Chi Zhang 2 Jingyang Li 2 Aili Chen 2 Ke Ji 2 Weiyu Cheng 2 Zijia Wu 2 Chengyu Du 2 Qidi Xu 2 Jiayuan Song 2 Zhengmao Zhu 2 Wenhu Chen 3 Pengyu Zhao 2 Junxian He 1 1 The Hong Kong University of Science and Technology 2 MiniMax 3 University of Waterloo [email protected] jun |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-06501