The Super VL, with 120 billion parameters, is aimed at enterprise users, whilst the Bolt 74B-A9B and E4B are designed to run locally on RTX Spark and DGX Spark machines.
On 19 October, Nvidia unveiled a new generation of open-weight models, Nemotron 3.5, designed for agents. The largest, Nemotron 3.5 Super VL, has 120 billion parameters and is aimed at business applications: document analysis, accounting, scientific experimentation, processing healthcare reimbursement claims and security analysis. Nvidia claims that once fine-tuned for a specific agentic task, it performs it faster than competing models from Anthropic and OpenAI.
Agents that remain on the machine
The Bolt models target the other end of the spectrum: the Nemotron 3.5 Bolt 74B-A9B, which fits into 24 GB, is designed to run locally on PCs equipped with RTX Spark chips and on DGX Spark workstations, so that everyday tasks can be entrusted to an agent without sending data to the cloud. The smallest model, Bolt E4B, is designed for conversational agents.
Super VL and Bolt 74B-A9B are available immediately, whilst Bolt E4B is expected later in October. Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce are among the first users of Super VL, which organisations can retrain using their own data and host on their own infrastructure.
Read the full article on the original source
Go to the source →