NVIDIA Unveils Nemotron-3-Embed

NVIDIA has unveiled the Nemotron-3-Embed models (1B and 8B parameters), optimized for retrieval and semantic similarity tasks. These models are designed as a core component of text-based RAG (Retrieval-Augmented Generation) systems.

Key features include:

• Multilingual support: Provides robust cross-lingual retrieval capability across 34 languages, including Korean, English, Chinese, and Japanese.
• Performance: As of July 2026, achieved SOTA (State-of-the-art) performance on the multilingual RTEB leaderboard.
• Use cases: Ready to be deployed out of the box for semantic search, information retrieval, and agentic RAG workflows through dense vector generation.
• Commercial use: The models are prepared to be available for commercial use.