K-BrowseComp : A Web Browsing Agent Benchmark Grounded in Korean Contexts
Nahyun Lee, Dongkeun Yoon, Guijin Son, Geewook Kim, Dayoon Ko, Jeonghun Park, Haneul Yoo, Jaewon Cho, Junghun Park, Changyoon Lee, Kyochul Jang, Jaeyeon Kim, Eunsu Kim, Woojin Cho, Seungone Kim, Chung-Ang University, Seoul National University, OnelineAI, NAVER Cloud AI, Carnegie Mellon University [email protected] [email protected] · 2026
Read K-BrowseComp : A Web Browsing Agent Benchmark Grounded in Korean Contexts by Nahyun Lee, Dongkeun Yoon, Guijin Son, Geewook Kim, Dayoon Ko, Jeonghun Park, Haneul Yoo, Jaewon Cho, Junghun Park, Changyoon Lee, Kyochul Jang, Jaeyeon Kim, Eunsu Kim, Woojin Cho, Seungone Kim, Chung-Ang University, Seoul National University, OnelineAI, NAVER Cloud AI, Carnegie Mellon University [email protected] [email protected] (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-2606-02404.txt.
Description
Lee and co-authors build a benchmark for web browsing agents in Korean settings. Top model tests now stress agentic skills, not just chat skills. Korean agentic benchmarks are scarce. Their set, K-BrowseComp, holds 400 problems. Native Korean speakers built and checked 300 of them by hand.
Top frontier models score just 30 to 45.67 percent on the checked subset. That is a steep drop from the English BrowseComp set. Korean models from a national program score near zero to 10.33 percent. A second split of 100 problems is synthetic. It was built to stress known failure modes. The best model reaches only 26 percent there. The team releases the data and code in public. The gap shows how far browsing agents still have to go in Korean.
Catalog record
| CREATOR | Nahyun Lee, Dongkeun Yoon, Guijin Son, Geewook Kim, Dayoon Ko, Jeonghun Park, Haneul Yoo, Jaewon Cho, Junghun Park, Changyoon Lee, Kyochul Jang, Jaeyeon Kim, Eunsu Kim, Woojin Cho, Seungone Kim, Chung-Ang University, Seoul National University, OnelineAI, NAVER Cloud AI, Carnegie Mellon University [email protected] [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 98 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.02404