NVIDIA released Nemotron 3.5 Super VL. It is a 120-billion-parameter open model for specialized agents. It delivers frontier accuracy at one-fifth the size of comparable models. It is the fastest open agentic model in its class. Super VL is available now on Hugging Face, Open Router and ModelScope. Bolt 74B-A9B 24GB is also available now. Bolt E4B 4GB will launch on October 23. Super VL runs on DGX Spark, DGX Station, RTX PRO and RTX systems. Bolt 74B-A9B runs on RTX Spark laptops, GeForce RTX PCs and DGX systems. Bolt E4B runs on GeForce RTX, Jetson and RTX/DGX Spark. Organizations can post-train Super VL on their own data. It supports optical character recognition, tables, charts and structured documents. It combines text and visual understanding in one model.
What changed
Nemotron family had Lightning and Ultra models only.
What it unlocks
Running specialized agents on NVIDIA hardware from data center to edge devices.
- 120B parameters
- 26T tokens trained
- 3.8M post-training samples
- 5x faster than next best in class
What you need to act on it
- NVIDIA infrastructure
Sources