# BrowseComp

> BrowseComp is a benchmark of hard-to-find but easy-to-verify questions that require an agent to search the web persistently and creatively, often across many pages, to locate an entangled piece of information.

- Identifier: PTL-0207
- Category: Benchmarks
- Canonical URL: https://protologue.com/t/browsecomp/
- Introduced: 2025

## Description

Answers are short so grading is simple, while finding them can take a person hours. It measures the persistence and search strategy of browsing and deep-research agents rather than knowledge the model already has.

## Broader terms

- [Benchmark](https://protologue.com/t/benchmark/)

## Related terms

- [Web Search Tool](https://protologue.com/t/web-search-tool/)
- [Agentic RAG](https://protologue.com/t/agentic-rag/)
- [GAIA](https://protologue.com/t/gaia/)

## Sources

- Wei et al. (2025). BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents. https://arxiv.org/abs/2504.12516

## Cite this entry

Protologue. (2026). BrowseComp. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0207). https://protologue.com/t/browsecomp/

License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)