BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents
This benchmark addresses the need for a standardized test of browsing persistence and creativity in AI agents, though it is incremental as it focuses on a specific capability rather than full user query handling.
The authors introduced BrowseComp, a benchmark with 1,266 questions designed to test web browsing agents' ability to persistently navigate and find hard-to-find information, with short, verifiable answers to simplify evaluation.
We present BrowseComp, a simple yet challenging benchmark for measuring the ability for agents to browse the web. BrowseComp comprises 1,266 questions that require persistently navigating the internet in search of hard-to-find, entangled information. Despite the difficulty of the questions, BrowseComp is simple and easy-to-use, as predicted answers are short and easily verifiable against reference answers. BrowseComp for browsing agents can be seen as analogous to how programming competitions are an incomplete but useful benchmark for coding agents. While BrowseComp sidesteps challenges of a true user query distribution, like generating long answers or resolving ambiguity, it measures the important core capability of exercising persistence and creativity in finding information. BrowseComp can be found at https://github.com/openai/simple-evals.