DevelopingOpen source
Nvidia releases open Nemotron 3.5 embedding, reranker and parsing models
Nvidia released Nemotron 3.5 Embed (8B and 1B), Rerank-8B and a 631M-parameter Parse model with open weights on Hugging Face. Nvidia says Embed-8B scores 61.17 nDCG@10 on ViDoRe v3.1, and the reranker cut agent token cost about 14%. Benchmarks are Nvidia's own.
- HEAT
- OUTLETS
- 0
- FIRST SEEN
- LAST SIGNAL
Primary source
Hugging Facehuggingface.co/blog/nvidia/nemotron-retriever-world-class-agentic-retrievalCoverage · 1 article, oldest first
Friday, October 23, 2026
Hugging FacePrimary
Nvidia releases open Nemotron 3.5 embedding, reranker and parsing models