Nvidia's Nemotron 3.5 Super VL is an open model built to run enterprise AI agents nonstop
decoder ·
Nvidia has released Nemotron 3.5 Super VL, an open multimodal model with 120 billion parameters designed for continuous enterprise AI agent tasks. Built using a Mixture-of-Experts architecture, the model activates only 12 billion parameters per token to optimize efficiency while processing text, images, and documents. Nvidia also introduced two smaller Bolt variants: a 74B-A9B version for local machines with 24 GB of memory and a compact 4GB model aimed at voice assistants. The company claims that Super VL matches the accuracy of its larger Ultra predecessor but operates with one-fifth the parameters and significantly higher throughput on Nvidia hardware. To enhance performance, Nvidia utilizes 4-bit compression and caching techniques that reduce latency and allow the model to serve more concurrent users. Early adopters in finance and cybersecurity are already fine-tuning the model for specific use cases such as fraud detection and vulnerability analysis. While the weights and training datasets are openly available, the models remain heavily optimized for Nvidia’s proprietary chips and systems.
decoder