Nvidia adds Super VL and Bolt to Nemotron 3.5 open model range Nvidia’s Nemotron 3.5 models are intended to support a variety of agent deployments. · Verdict · El editorial/Shutterstock.com. Anwesha Pattanaik Fri, October 16, 2026 at 6:39 AM EDT 3 min read NVDA -1.84% Trade NVIDIA on Coinbase Trading disclosure Trading disclosure The above button links to Coinbase. Yahoo Finance is not a broker-dealer or investment adviser and does not offer securities or cryptocurrencies for sale or facilitate trading. Coinbase pays us for certain activity generated through this link. Prices displayed are informational.
Nvidia has announced the release of three new models in its Nemotron family, including Nemotron 3.5 Super VL, a 120 billion-parameter open model.
The company states that this new model is designed to enable higher efficiency and accuracy for always-on AI agents performing specialised tasks, with two additional 'Bolt' models aimed at edge deployment scenarios.
The Nemotron 3.5 Super VL and Nemotron 3.5 Bolt 74B-A9B 24GB are available immediately, while the Nemotron 3.5 Bolt E4B 4GB is expected to launch on 23 October.
The Nemotron 3.5 models are intended to support a variety of agent deployments, ranging from enterprise servers to local hardware and edge devices.
Nvidia said that with the new releases, organisations using its infrastructure can select models optimised for their needs without compromising between capability and deployability.
The Nemotron range now comprises the new Super VL and Bolt models alongside earlier releases named Ultra and Lightning.
According to Nvidia, Nemotron 3.5 Super VL achieves accuracy on multimodal agentic tasks, such as instruction following, workflow automation, and visual understanding, on par with its larger reasoning model, but at roughly one-fifth of the parameter count.
The company claims that Super VL is five times faster than comparable open models in completing complex workflows, and that organisations can customise the model with their own data and deploy it on their own systems.
On the edge deployment front, Nemotron 3.5 Bolt 74B-A9B 24GB has been optimised to run on Nvidia RTX and DGX Spark hardware, using a 2-bit quantisation method to fit into a 24GB memory footprint.
The company says this balancing of intelligence and speed enables a wider range of local tasks by users without relying on cloud services.
The Bolt E4B 4GB model has been designed for conversational and physical agents running on GeForce RTX, Jetson, and other Nvidia systems.
Nvidia reported that dozens of organisations are customising and post-training the Super VL model for specific business use cases, including financial document analysis, medical claims management, and scientific discovery.
Companies such as Scale AI, EY, AfterQuery, AI21, Mercor, ModelScope, Ramp, and others are named as users adapting the model for their sectors.
The Nemotron 3.5 Super VL's design allows it to process structured data, images, web pages, and PDFs, integrating both text and visual information for agent tasks.
The model is released with open weights, datasets, and training recipes, supporting customisation through Nvidia NeMo and other developer tools.
Nemotron 3.5 Super VL was trained on approximately 26 trillion tokens, followed by post-training using around 3.8 million samples.
The model is available through Hugging Face, the ModelScope community, Open Router, and as an Nvidia NIM microservice, and can be deployed across Nvidia DGX Spark, DGX Station, RTX, RTX PRO, and Jetson systems.
Earlier this month, Nvidia began preliminary talks to either expand its investment in Reflection AI or consider acquiring the US-based open-source AI startup.
"Nvidia adds Super VL and Bolt to Nemotron 3.5 open model range" was originally created and published by Verdict, a GlobalData owned brand.
