Web research and search
Using the web to get things done: answering hard research questions, and finishing tasks on live websites such as booking or shopping. Good means the answer is right or the task is finished.
Sources last checked 2026-10-23.
We can't show scores for this kind of work yet. These testers publish results: FutureSearch (one test), University of Waterloo (Tevatron) (one test) and OSU NLP Group (one test). See who tests it
At a glance
Vendor claims have not been collected for any category yet.
ⓘ More about this category
- Reliability not yet measured
- 3 sources, linked results only
- Sources last checked 2026-10-28
- Ring 4: needs rings 1 to 3 first
- DRB = Deep Research Bench, BCP = BrowseComp-Plus, M2W = Online-Mind2Web
Tested by
Deep Research Bench
FutureSearch, checked 2026-10-28How well an agent answers hard real-world research questions by searching a frozen offline copy of the web, checked against answers the tester curated.
Last updated 2026-06-10Covers 16 of the models we trackFutureSearch uses this benchmark to measure its own forecasting and research agents.Scores not shown: no license for reuse stated
See their resultsMore about this test: Deep Research Bench
Reliability not measured by this source.
Technical details for Deep Research Bench
Score: How well an agent answers hard real-world research questions by searching a frozen offline copy of the web, checked against answers the tester curated.
License: Score reuse not stated. Checked 2026-10-28.
BrowseComp-Plus
University of Waterloo (Tevatron), checked 2026-10-28Whether a research agent finds the right short answer to hard-to-locate questions while searching a fixed collection of web pages, graded by a language model against the known answer.
Last updated 2026-10-16Covers 5 of the models we trackThe test is built on top of BrowseComp (openai.com/index/browsecomp).Scores not shown: no license for reuse stated
See their resultsMore about this test: BrowseComp-Plus
Reliability not measured by this source.
Technical details for BrowseComp-Plus
Accuracy: Whether a research agent finds the right short answer to hard-to-locate questions while searching a fixed collection of web pages, graded by a language model against the known answer.
License: Score reuse not stated. Checked 2026-10-28.
Online-Mind2Web
OSU NLP Group, checked 2026-10-28Whether a web agent finishes everyday tasks on real live websites, such as booking, shopping or looking things up, as judged by human reviewers.
Last updated 2026-08-04Covers 1 of the models we trackScores not shown: no license for reuse stated
See their resultsMore about this test: Online-Mind2Web
Reliability not measured by this source.
Technical details for Online-Mind2Web
Average SR: Whether a web agent finishes everyday tasks on real live websites, such as booking, shopping or looking things up, as judged by human reviewers.
License: Score reuse not stated. Checked 2026-10-28.
What this doesn't tell you
- Tests on live websites can give different results later, because the sites change.
- None of these tests used your questions or your sources.
- Scores from different sources are not comparable, even when they share a unit.
- None of these tests were run on your own documents, code or data.
- No source for this task allows us to show its figures yet. Follow the links for their results.
Data version 2026-10-28+91ac3fc9d071 · Terms of use