The development
What happened
NVIDIA’s 15 October release spans visual-language agents and local workloads. Nemotron 3.5 Super VL and Bolt 74B-A9B 24GB are available, while the smaller Bolt E4B 4GB has a separate release date.
Key facts
Super VL has 120 billion parameters and targets webpages, PDFs, tables and charts alongside agent workflows.
Open weights, datasets, training recipes and customization resources are offered; distribution includes Hugging Face, OpenRouter, ModelScope and NVIDIA NIM.
Bolt 74B-A9B uses native 2-bit quantization to target a 24GB memory footprint, with llama.cpp, Ollama, LM Studio, Unsloth and vLLM support.
Bolt E4B 4GB is scheduled for 23 October and remains an upcoming model.
The context
Why it matters
TideReach assessment: The range gives teams different choices for document-heavy server agents and local execution. Practical memory overhead, task accuracy and operating cost still need workload-specific testing; NVIDIA’s comparative speed claims were not independently reproduced here.