NVIDIA releases Nemotron-3-Embed 1B and 8B models

NVIDIA has released the Nemotron-3-Embed 1B and 8B text embedding models, optimized for retrieval and semantic similarity tasks. These models are designed to serve as foundational components in text-based Retrieval-Augmented Generation (RAG) systems.

The models generate dense vector embeddings from multilingual text inputs, enabling efficient similarity matching for questions or passages. They support evaluation across 34 languages, including English, Chinese, Spanish, Hindi, and Arabic. As of July 2026, the Nemotron-3-Embed-8B-BF16 variant achieves state-of-the-art performance on the multilingual RTEB leaderboard.

Both models are ready for commercial use.