How NVIDIA AI-Q Reached \#1 on DeepResearch Bench I and II

NVIDIA's AI-Q deep research agent achieved first place on the DeepResearch Bench I and II by using an open, multi-agent architecture. This system features an orchestrator, planner, and researcher built on the NVIDIA NeMo Agent Toolkit to perform iterative research loops. A key factor in its success is a fine-tuned NVIDIA Nemotron 3 Super model trained on over 67,000 search-and-synthesis trajectories filtered by a principle-based judge model. The architecture enables evidence-grounded planning and dispatches specialist agents to gather and synthesize information, producing well-cited reports.