DRACO
DRACO is a benchmark of 100 deep research tasks from Perplexity and Harvard that scores reports against expert rubrics.
What it means
DRACO stands for Deep Research Accuracy, Completeness, and Objectivity. Researchers from Perplexity and Harvard introduced it in a 2026 paper. Its tasks come from anonymized deep research requests, filtered and extended, and they span 10 domains with sources from 40 countries.
Each task has an expert-written rubric with about 40 criteria across four areas: factual accuracy, breadth and depth, presentation, and citation quality. A judge model checks each criterion and returns met or unmet. A tool's score is the normalized share of criteria it earns. Hyperresearch's October 2026 DRACO run used 10 tasks and one judge, and it scored 77.7. That is a smaller test than the full 100-task set. The rubric rewards accuracy as well as completeness, so a high score needs more than a long report.
Continue this reading path
Cite this entry
https://bestdeepresearchai.com/glossary/draco/