On Oct. 15, 2026, NVIDIA announced three open Nemotron 3.5 models: the 120-billion-parameter Super VL for multimodal AI agents and two smaller Bolt models designed for local use.
Super VL is a mixture-of-experts model that activates 12 billion parameters per token and can process text, images, documents and video while reasoning and using tools. NVIDIA says it is intended for specialized, always-on enterprise agents that handle multistep tasks, such as reviewing documents and checking information across sources. The company also announced Nemotron 3.5 Bolt 74B-A9B, aimed at local agent workloads on NVIDIA RTX systems and DGX Spark, and Bolt E4B, a compact model for conversational agents.
As of Oct. 16, Bolt E4B is scheduled to become available on Oct. 23, while Super VL and Bolt 74B-A9B are available now. NVIDIA says it is releasing model weights, datasets and training recipes, allowing developers to adapt the models to their own data and workflows. The company claims Super VL delivers up to five times the throughput of comparable open models for agent tasks; that is NVIDIA’s performance claim, not a guarantee for every setup.
The launch gives developers options from enterprise-scale multimodal systems to local agents, though the smaller models are designed around NVIDIA hardware.