K-BrowseComp: Korean Web-Browsing Agent Benchmark

K-BrowseComp is a Korean version of the BrowseComp benchmark for evaluating web-browsing agents. The dataset contains 300 handcrafted questions grounded in Korean contexts that require retrieving information across multiple Korean websites. It was created by native Korean speakers from the prometheus-eval organization and was last updated on June 1, 2026.

Use Cases

• Benchmarking Korean-language web-browsing agents based on the 300-question verified subset.
• Evaluating cross-website information retrieval capabilities in a Korean context.
• Training or fine-tuning agents to handle tasks grounded in Korean culture and websites.

Strengths

• Contains a verified subset of 300 questions entirely handcrafted by native Korean speakers.
• Every item underwent thorough manual revision and validation, as stated in the description.

Limitations

• Row count, column definitions, and file formats are unknown, limiting suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

prometheus-eval

Collection Method

Items are handcrafted by native Korean speakers and manually revised.

Freshness

Last updated 2026-06-01 08:24:26; freshness should be verified.

Geography

South Korea (inferred from Korean context and websites)