FS-Researcher: Test-Time Scaling for Long-Horizon Research Tasks with File-System-Based Agents
Chiwei Zhu, Benfeng Xu, Mingxuan Du, Shaohan Wang, Xiaorui Wang, Zhendong Mao, Yongdong Zhang · 2026
Read FS-Researcher: Test-Time Scaling for Long-Horizon Research Tasks with File-System-Based Agents by Chiwei Zhu, Benfeng Xu, Mingxuan Du, Shaohan Wang, Xiaorui Wang, Zhendong Mao, Yongdong Zhang (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-2602-01566.txt.
Description
Chiwei Zhu and co-authors present FS-Researcher, a dual-agent file-system framework for deep research beyond the context window. Long deep-research trajectories exceed model context limits, compressing budgets for evidence and reports, blocking test-time scaling. A context builder agent gathers web sources into structured notes in a hierarchical knowledge base outgrowing context length. A report writer drafts the final report section by section from that knowledge base. On two open-ended benchmarks, it shows leading report quality across backbones, with quality rising with context-builder computation.
Catalog record
| CREATOR | Chiwei Zhu, Benfeng Xu, Mingxuan Du, Shaohan Wang, Xiaorui Wang, Zhendong Mao, Yongdong Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45303 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.01566