Sage : Benchmarking and Improving Retrieval for Deep Research Agents
Tiansheng HuYilun ZhaoCanyu ZhangArman Cohan, Chen Zhao [4pt] NYU Shanghai Yale University Center for Data Science, New York University[4pt] https://github.com/HughieHu/Sage · 2026
Read Sage : Benchmarking and Improving Retrieval for Deep Research Agents by Tiansheng HuYilun ZhaoCanyu ZhangArman Cohan, Chen Zhao [4pt] NYU Shanghai Yale University Center for Data Science, New York University[4pt] https://github.com/HughieHu/Sage (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-2602-05975.txt.
Description
Deep research agents can chase hard questions, but they lean on search tools that must find the right papers first. Tiansheng Hu and four co-authors build SAGE, a test for that search step: twelve hundred questions across four science fields, over a corpus of two hundred thousand papers. They run six deep research agents on it and find all six stumble on retrieval that needs real reasoning.
A surprise follows. The old keyword method BM25 beats modern language-model retrievers by about thirty percent, since agents tend to write keyword-heavy sub-questions. To close the gap, the team adds language-model notes, metadata, and keywords to documents at test time. Scores rise by eight percent on short answers and two percent on open answers. SAGE gives the field a hard yardstick for search inside research agents.
Catalog record
| CREATOR | Tiansheng HuYilun ZhaoCanyu ZhangArman Cohan, Chen Zhao [4pt] NYU Shanghai Yale University Center for Data Science, New York University[4pt] https://github.com/HughieHu/Sage |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-05975