MMLForge Search models, datasets, tasks… Start free
Home › Datasets › Tevatron/browsecomp-plus
B

Tevatron/browsecomp-plus

Question Answering · Tevatron· 57.6K
mit 11 GB task_categories:question-answeringlicense:mitsize_categories:n<1Kformat:parquetmodality:text

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

Open in MLForge Sign up free Desktop app
# download instantly
mlforge datasets pull Tevatron/browsecomp-plus

Dataset details

Task
Question Answering
License
mit
Size
11 GB
Rows / images
830
Classes
6
Creator
Tevatron
Downloads
57.6K
Source
huggingface_datasets
Updated
2025-12-20

About Tevatron/browsecomp-plus

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 human-verified evidence documents sufficient to answer the queries, and mined hard negatives to keep the task challenging.