# NVIDIA releases multilingual multimodal reranker on Hugging Face

> The Nemotron-3.5-Rerank-8B model, able to rank text and images in 34 languages, is now available for commercial use.

Oossa · 2026-10-15 · https://oossa.com/en/nvidia-releases-multilingual-multimodal-reranker-on-hugging-face

NVIDIA uploaded a new reranking model called Nemotron-3.5-Rerank-8B-BF16 to Hugging Face. It scores how relevant a document—text, image, or both—is to a query. The model works in 34 languages for text and 18 for image queries. It runs on NVIDIA GPUs and is released under the OpenMDW‑1.1 license for both commercial and non‑commercial projects.

## The facts

- Release date: 2026-10-20 ("Release Date: Hugging Face 10/20/2026")
- Model size: 8.38 billion parameters

## Why it matters

Developers can improve search results in multilingual, image‑rich applications without building their own reranker.

## Sources & references

1. [New model on Hugging Face: nvidia/Nemotron-3.5-Rerank-8B-BF16 (text-ranking)](https://huggingface.co/nvidia/Nemotron-3.5-Rerank-8B-BF16) – NVIDIA, 2026-10-15

Last updated: 2026-10-15