EigentSearch-Q+: Structured Reasoning Tools for AI Research Agents

A new deep research agent, EigentSearch-Q+, has been unveiled by researchers, enhancing web search through the incorporation of structured reasoning tools. This system builds on the 'think' tool framework from Anthropic and existing information-retrieval research, introducing features like query planning, monitoring search progress, and extracting evidence from extensive web snapshots. Q+ is part of Eigent's browser sub-agent, which is an open-source multi-agent system for computer tasks. When tested against four benchmarks—SimpleQA-Verified, FRAMES, WebWalkerQA, and XBench DeepSearch—the browser agent's benchmark-size-weighted average accuracy saw improvements of 3.0, 3.8, and 0.6 percentage points, respectively. This advancement addresses a fundamental AI capability: conducting deep research that necessitates reasoning over web evidence for open-ended inquiries. Many existing agents depend on implicit, unstructured search methods, leading to repetitive exploration and fragile evidence collection. The Q+ tools aim to make web searches more intentional, refining the agent's query development and evidence analysis. The research paper can be found on arXiv with the identifier 2604.07927.