---
title: "Nvidia Introduces Nemotron 3.5 AI Model Family with Deep Customization Support"
description: "Nvidia unveiled the Nemotron 3.5 AI model family, headlined by the 120-billion-parameter Super VL supporting post-training and visual analysis for enterprise agents."
date: 2026-10-17T02:26:31.989Z
lang: en
url: https://xab.info/en/posts/nvidia-introduces-nemotron-3-5-ai-model-family
tags: [nvidia, nemotron, artificial-intelligence, llm, enterprise-ai, machine-learning]
author: "Alexander Buran"
publisher: "XAB.info"
---

# Nvidia Introduces Nemotron 3.5 AI Model Family with Deep Customization Support

![Data visualization and neural network architecture interface for Nvidia Nemotron models](https://xab.info/media/2026/10/17/nvidia-predstavila-semeystvo-ii-modeley-nemotron-3-5/nvidia-predstavila-semeystvo-ii-modeley-nemotron-3-5-1.webp)

## Key Points

- Three new Nemotron 3.5 models announced for enterprise workloads and local agent deployment.
- Flagship Nemotron 3.5 Super VL features 120 billion parameters and multimodal visual analysis.
- Post-training Super VL on proprietary customer data yields accuracy improvements from 9% to 44%.
- Bolt and E4B models are optimized for local execution on consumer-grade hardware.

Nvidia has expanded its artificial intelligence portfolio by introducing three new models in the Nemotron 3.5 family, tailored for complex agentic tasks and enterprise workflow integration. The flagship of the new lineup is the open Nemotron 3.5 Super VL model with 120 billion parameters, combining advanced language capabilities with visual understanding of documents, charts, and web pages.

### Key Features of the Flagship Nemotron 3.5 Super VL

The new Super VL model is specifically designed to handle multimodal content, including PDF files, tables, images, and web interfaces. According to Nvidia, enterprise customers can independently perform post-training on their own data, tools, and security compliance policies. This approach allows organizations to deploy the model either on their internal infrastructure or within secure cloud environments with strict data privacy requirements.

During comparative benchmarking, specialized customized versions of Nemotron 3.5 Super VL demonstrated significant accuracy gains after fine-tuning. For instance, in health insurance claims processing tasks, accuracy increased from 28% to 72%, outperforming competing solutions. In cybersecurity threat resistance, the metric improved from 35% to 76%, also exceeding results of comparable specialized systems on the market.

### Local Agents and Lightweight Lineup Additions

Alongside the flagship solution, the company announced two compact models for local execution. The Nemotron 3.5 Bolt 74B-A9B 24GB model features 74 billion total parameters, with approximately 9 billion active per token. The neural network is optimized to fit within 24 GB of VRAM, enabling local execution on consumer or workstation hardware for code generation and multi-step research.

The third addition is the lightweight Nemotron E4B 4GB model, designed to power conversational agents and local utility tools, with a full commercial launch scheduled for October 23, 2026. Developers note that these new architectures complement earlier releases in the product line, including the high-throughput Nemotron 3.5 Lightning model designed for repetitive background workflows.

### Significance for the Enterprise AI Market

The emergence of customizable Nemotron 3.5 models marks a new milestone in the evolution of autonomous AI agents capable of seamlessly interacting with complex enterprise ecosystems. The ability to tailor models to narrow, specialized domains reduces barriers for neural network adoption across finance, healthcare, and cybersecurity. Over time, this is expected to help organizations optimize computing expenditure and increase the autonomy of digital assistants when processing large volumes of structured and visual data.

## Fact-Check & Sources

- [ИИ-модель Nvidia Nemotron 3.5 Super VL с пост-обучением превзошла Claude Opus 5 и GPT-6.1 Sol в специализированных задачах window-new](https://3dnews.ru/1150010/iimodel-nvidia-nemotron-35-super-vl-s-postobucheniem-prevzoshla-claude-opus-5-i-gpt61-sol-v-spetsializirovannih-zadachah) - Primary technical specifications and details regarding the Nemotron models are verified via 3dnews.ru.
- [Модель Nemotron 3.5 Lightning от NVIDIA ускорит выполнение агентами повторяющихся задач](https://overclockers.ru/blog/vokrugsveta/show/261568/NVIDIA-zapustila-model-Nemotron-3-5-Lightning-dlya-uskoreniya-vypolneniya-agentami-povtoryajuschihsya-zadach) - Context regarding the Nemotron 3.5 Lightning model for routine tasks is supported by overclockers.ru.
- [NVIDIA выпустила модель Nemotron 3.5 Lightning для ускорения рутинных ИИ-задач до 4 раз](https://overclockers.ru/blog/This_is_the_way/show/261578/NVIDIA-vypustila-Nemotron-3-5-Lightning-dlya-uskoreniya-rutinnyh-II-zadach-do-4-raz) - Licensing details (OpenMDW-1.1) and throughput benchmarks are backed by secondary reporting from overclockers.ru.
- [Nvidia представила Nemotron 3 Ultra: ИИ-модель для автономных агентов стала быстрее в пять раз и дешевле на 30%](https://www.ixbt.com/news/2026/06/04/nvidia-nemotron-3-ultra-30.html) - Information regarding broader previous Ultra models in the lineup is confirmed via ixbt.com.

## FAQ

### What is the Nemotron 3.5 Super VL model?

It is an open, large language model with 120 billion parameters featuring enhanced visual and textual analysis capabilities for documents, charts, and web pages.

### What is the purpose of the Nemotron 3.5 Bolt model?

Bolt is optimized for local agent execution, fits within 24 GB of memory, and is used for software code generation and multi-step research tasks.

### What accuracy gains does Super VL post-training provide?

Depending on the specialized domain, accuracy improvements after fine-tuning on enterprise data range from 9% to 44%.