← Back to resources

Benchmark

DeepResearch Bench

Profile in progress

USTC / Metastone

Evaluate deep-research agents across 100 expert-authored PhD-level tasks in 22 fields.

Open official page ↗

Resource profile

Access
Access status unknown
Access note
Primary project page; no availability claim is inferred.
Inputs
Generated research reports and agent retrieval/citation outputs.
Outputs
RACE report-quality scores and FACT retrieval/citation measures.
License
License not yet verified
Limits
Published table is a historical evaluation; it is not a current universal model ranking.
Evaluation conditions
50 Chinese and 50 English tasks; report RACE and FACT separately with agent/tool configuration.
Code release
Not verified
Data release
Not verified
Site execution check
Not verified

Associated timeline

No verified associated update is currently displayed.