88Z AI, 17 July — NVIDIA has released the open-source Nemotron 3 Embed text embedding model series, optimized for retrieval-augmented generation and agent search. The series offers 8B and 1B parameter sizes, supporting a 32k context window and 34 languages. The flagship 8B model ranks first on the RTEB multilingual retrieval benchmark with a score of 78.5%, while the 1B version retains 95% retrieval accuracy while significantly reducing inference costs through pruning and distillation.
At a glance
- The flagship 8B model leads the RTEB benchmark with a 78.5% score
- The 1B version retains 95% accuracy after optimization while lowering inference costs
The update
NVIDIA has officially open-sourced the Nemotron 3 Embed text embedding model series, designed to enhance performance in retrieval-augmented generation (RAG) and agent search scenarios. The model family includes versions with 8 billion and 1 billion parameters, both supporting a context window of up to 32k and covering 34 major languages, enabling broad cross-language processing capabilities. In terms of performance, the flagship 8B model excels, ranking first on the RTEB multilingual retrieval benchmark with a score of 78.5%. For resource-constrained scenarios, the 1B version is optimized using pruning and distillation techniques. It significantly reduces inference costs while effectively retaining 95% of retrieval accuracy, offering developers an efficient and cost-effective option.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- EmbedSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.