NVIDIA Says Nemotron Reranking Cuts AI Agent Token Use

NVIDIA says adding Nemotron 3.5 Rerank to its retrieval pipeline reduced the tokens used by an AI agent per query while improving retrieval accuracy in the tested configurations. The company’s evaluation compares retrieval setups with the same agent, tools, tasks, corpus and execution policy, while varying the search system. NVIDIA’s announcement says reranking lowered agent token costs by about 14% with Nemotron 3.5 Embed 8B and 13% with Nemotron 3.5 Embed 1B.