A Tale of Two Graphs: Separating Knowledge Exploration from Outline Structure for Open-Ended Deep Research
This new AI architecture solves a major flaw in how LLMs do deep research.
Deep Dive
Researchers have unveiled 'DualGraph,' a new AI agent architecture that separates knowledge exploration from report structure to tackle long-horizon research. It uses two co-evolving graphs—a Knowledge Graph for semantic memory and an Outline Graph for structure—to generate targeted searches. The system consistently outperforms state-of-the-art baselines, achieving a 53.08 RACE score on the DeepResearch Bench using GPT-5, and improves report depth, breadth, and factual grounding.
Why It Matters
It enables AI agents to conduct more comprehensive, accurate, and efficient deep research, moving beyond simple Q&A to structured analysis.