Capability in benchmark · Version 1

Web Retrieval

Web Retrieval means the product can find live-web material that supports an answer when it exists, and can tell when it does not.

Web Search for AI →Read the global definition →

4 scenarios · 7 current published Results · 5 tools with published Results

How the tools performed

Explore current published Results across this capability’s scenarios. Open a scenario to compare its tools, or a Result to inspect the evidence.

Each cell reports its own published test set. There is no combined capability score or claim that different Results used identical tests.

Scenarios in this capability

Tools with published Results appear first, alphabetically. Remaining participants stay visible below.
Participating tool	The request uses different wording than the source →	A page on the web answers the question →	Something more famous has the same name →	Search only these websites →
Brave Search API	The request uses different wording than the source → 1 Pass 1/1 assessed1/1 gradableView Result →	A page on the web answers the question →No published result	Something more famous has the same name →No published result	Search only these websites →No published result
Exa	The request uses different wording than the source →No published result	A page on the web answers the question →No published result	Something more famous has the same name → 1 Pass 1/1 assessed1/1 gradableView Result →	Search only these websites →No published result
Linkup	The request uses different wording than the source →No published result	A page on the web answers the question → 1 Pass 1/1 assessed1/1 gradableView Result →	Something more famous has the same name → 1 Pass 1/1 assessed1/1 gradableView Result →	Search only these websites →No published result
SerpAPI	The request uses different wording than the source →No published result	A page on the web answers the question →No published result	Something more famous has the same name → 1 Pass 1/1 assessed1/1 gradableView Result →	Search only these websites → 1 Pass 1/1 assessed1/1 gradableView Result →
Serper	The request uses different wording than the source →No published result	A page on the web answers the question →No published result	Something more famous has the same name → 1 Pass 1/1 assessed1/1 gradableView Result →	Search only these websites →No published result
Firecrawl	No published result across these scenarios
GPT-5.6 terra	No published result across these scenarios
OpenAI web search	No published result across these scenarios
Perplexity Sonar	No published result across these scenarios
Tavily	No published result across these scenarios
Valyu	No published result across these scenarios
You.com	No published result across these scenarios

Reading the comparison

Publication availability and test coverage describe different things.

Published evidence

Publication is not testing progress

“No published result” says only that there is no current public Result. It does not indicate whether a tool has been tested, passed, failed or is applicable.

Test coverage

Read each Result’s test set

Assessed includes Pass, Fail and Not gradable. Gradable includes Pass and Fail. Each fraction uses the pinned tests in that same published Result.

Go deeper

Choose the view you need

A scenario compares tools on one situation. A tool page follows one tool across this benchmark. A Result explains the outcome and shows the evidence.

Benchmark methodology & resources →