NVIDIA releases Nemotron 3 Embed models ranking #1 on RTEB

NVIDIA has released Nemotron 3 Embed, a collection of open and commercially available embedding models designed to improve retrieval quality for production-scale RAG, agentic retrieval, code retrieval, and agent memory. The collection features an 8B model that ranks #1 overall on the RTEB leaderboard, alongside efficient 1B variants optimized for deployment.

• Nemotron-3-Embed-8B-BF16 tops the RTEB leaderboard with a 78.5% score and achieves state-of-the-art retrieval across accuracy-efficiency curves.
• The 1B BF16 variant reduces error rates by 27% over its predecessor while maintaining high retrieval quality for lower-cost deployment.
• Nemotron-3-Embed-1B-NVFP4 utilizes Blackwell-optimized NVFP4 acceleration to deliver up to 2x higher throughput than BF16 with minimal accuracy loss.
• Models support a 32k context window and are available via Hugging Face, NVIDIA NIM microservices, and vLLM.

Stronger retrieval returns relevant evidence earlier, helping agentic systems avoid repeated searches and unnecessary reasoning turns, thereby reducing downstream token costs.