Intel / article

Benchmarking Hybrid Deep Research Across Database Querying and Web Search

Source: arxiv.org
Author: Ruofan Wu, Peiran Xu, Xiaolong Li, Fan Shu, Soyoung Yoon, Yite Wang, Xiaodong Yu, Boyi Liu, Feng Yan, Debiao Li, Yuxiong He, Zhewei Yao
Date: Published Sep 10, 2026

Context

HybridDeepResearch tests whether a research can carry information correctly between web search and database queries. Its 380 tasks require both sources to produce a complete answer. Some require database results to guide a search, others require web findings to guide a query, and others combine independently gathered results. The benchmark paper reports that the directional tasks were substantially harder than combining parallel results.

This exposes a weakness that separate search and database tests can miss: an agent must preserve the question's constraints when one tool's output determines its next action. Being competent with each tool individually does not establish that the agent can connect their evidence correctly.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.

Key quotes

“even state-of-the-art models like GLM-5.2, Claude-Sonnet-4.6 and GPT-5 achieve only about 50-54% Pass@8 on the hard subset”

“directional reasoning is substantially more difficult than parallel intersection”

Read the source arxiv.org