Nvidia has expanded Nemotron 3.5 with three open models intended to cover enterprise servers, local workstations and small edge devices. The 120-billion-parameter Super VL and Bolt 74B-A9B 24GB are available now; the smaller Bolt E4B 4GB is scheduled for October 23.
Super VL combines text and visual processing for webpages, tables, charts and PDFs. Nvidia says organizations can post-train it on their own policies and workflows and keep deployment on private infrastructure. The company reports training on about 26 trillion tokens followed by roughly 3.8 million post-training samples, and is releasing weights, datasets, recipes and reinforcement-learning environments.
Bolt 74B-A9B uses native two-bit quantization to fit in 24 GB of memory alongside other applications. It supports local agent tasks on RTX and DGX Spark systems through tools including llama.cpp, Ollama, LM Studio, Unsloth and vLLM. Bolt E4B targets conversational and physical agents on GeForce RTX and Jetson devices.
Nvidia claims Super VL is five times faster than the next model in its class and that Dynamo serving can support eight times more users with 85% lower latency. Those are vendor measurements tied to specified configurations and benchmarks; buyers will need to test their own documents, tools and hardware.