The Nemotron-3 embedding model now ranks first on the Retrieval Task Evaluation Benchmark. It outperforms previous leaders by optimizing how agents retrieve relevant documents for complex queries. This update reduces noise in RAG pipelines. Developers can now deploy the model via Hugging Face to improve agentic retrieval accuracy.