# NVIDIA posts quantized Nemotron 3.5 Super VL model

> The 120-billion-parameter vision-language model accepts text, images and video. NVIDIA says its NVFP4 version uses less memory than the BF16 model.

Oossa · 2026-10-15 · https://oossa.com/en/nvidia-posts-quantized-nemotron-3-5-super-vl-model

NVIDIA has posted NVIDIA Nemotron 3.5 Super VL NVFP4 on Hugging Face. The vision-language model can take text, images and video as input, and its weights are available under the OpenMDW 1.1 license for commercial or non-commercial use.

The model has 120 billion total parameters, with 12 billion active at a time, and supports up to 1 million tokens of context. NVIDIA lists deployments on hardware including one H200 or two H100 GPUs; it says the NVFP4 version scored 89.90 on GPQA-Diamond, compared with 89.52 for its BF16 version.

## The facts

- NVIDIA lists the model’s release date as October 15, 2026.
- The model has 120B total parameters and 12B active parameters.

## Why it matters

Developers with supported NVIDIA GPUs can use the posted weights to build text-and-vision agents, but the source does not give a price or independent benchmark results.

## Sources & references

1. [New model on Hugging Face: nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4 (image-text-to-text)](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4) – NVIDIA, 2026-10-12
2. [NVIDIA Nemotron 3.5 Super VL 120B A12B NVFP4](https://fireworks.ai/models/fireworks/nvidia-nemotron-3p5-super-vl-120b-a12b-nvfp4) – Fireworks AI
3. [New model on Hugging Face: nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 (image-text-to-text)](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16) – NVIDIA, 2026-10-12
4. [New model on Hugging Face: nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Base-BF16 (image-text-to-text)](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Base-BF16) – NVIDIA, 2026-10-12

Last updated: 2026-10-15