BrowseComp is a benchmark dataset used to measure how well large language models handle web-browsing and computer-use tasks. Model creators cite BrowseComp scores alongside SWE-bench and other evaluations to demonstrate real-world agentic performance.