NVIDIA Releases Nemotron-3-Embed Models, Topping the RTEB Retrieval Leaderboard

NVIDIA has released Nemotron-3-Embed, a new family of open-weight text embedding models, with its flagship 8-billion-parameter version claiming the top overall spot on the multilingual RTEB retrieval leaderboard. The launch targets the retrieval layer that increasingly underpins retrieval-augmented generation (RAG) and multi-step agent workflows, where the quality of embeddings directly shapes how much relevant context a model can pull in before generating an answer.