Tevatron/

browsecomp-plus-corpus

Project Page | Paper | Code BrowseComp-Plus is a new benchmark for Deep-Research system, isolating the effect of the retriever and the LLM agent to enable fair, transparent comparisons of Deep-Research agents. The benchmark sources challenging, reasoning-intensive queries from OpenAI's BrowseComp. However, instead of searching the live web, BrowseComp-Plus evaluates against a fixed, curated corpus of ~100K web documents from the web. The corpus includes both… See the full description on the dataset page: https://huggingface.co/datasets/Tevatron/browsecomp-plus-corpus.