Nvidia releases Nemotron 3.5 open AI models for agents

Nvidia ↘ 1.43% on Thursday, Oct. 15, added three open-weight models to its Nemotron family. The flagship is Nemotron 3.5 Super VL, a 120-billion-parameter multimodal model for "specialized always-on agents." It comes with two edge models, Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB, Nvidia said. Together, the models are meant to run AI agents on everything from data center systems to desktop computers, Yahoo Finance reported. Super VL and Bolt 74B-A9B are available now, and Bolt E4B is due Oct. 23.developer.nvidia.com+1

Super VL targets enterprise agents

Super VL uses a mixture-of-experts design that activates 12 billion of its 120 billion parameters per token. It handles text, images, documents and video in a single model, according to Nvidia's developer blog. Nvidia says the model delivers up to five times higher throughput on agentic tasks than other open models its size when run on a single GPU. It also says Super VL scored 80.8% on the CharXiv chart-reasoning benchmark.developer.nvidia.com

The model was distilled from Nemotron 3 Ultra, and businesses can post-train it on their own data. Nvidia pointed to a few early users. Scale AI expects inference costs for one vendor-credit workload to drop 40-fold, and Ramp expects up to 10 times lower latency for reading receipts. Super VL is offered on Hugging Face, OpenRouter, ModelScope and as an Nvidia NIM microservice.blogs.nvidia.com+1

Bolt models move agents to local hardware

Bolt 74B-A9B has 74 billion parameters and uses 9 billion of them per token. Two-bit quantization lets it fit in 24GB of memory. Like other Nemotron models, its weights, training data and recipes are fully open, Nvidia said. It runs through llama.cpp, Ollama, LM Studio, Unsloth and vLLM. Nvidia built it for its RTX Spark laptops, which go on sale Friday, Oct. 16. Bolt E4B is designed for physical and conversational agents, such as voice assistants and interactive game characters, on GeForce RTX, Jetson and DGX Spark hardware.blogs.nvidia.com+1

Nvidia's open-model strategy

Open-weight models have become part of Nvidia's growth strategy. Cheaper models can bring more organizations into AI, which in turn could lift long-term chip demand, Yahoo Finance noted. Chinese developers such as Moonshot AI and Alibaba ↘ 1.82% currently lead the conversation on open models with their Kimi and Qwen families. CEO Jensen Huang has urged U.S. companies to build and adopt more open models so the industry doesn't become locked into a few closed systems. ZDNET called Super VL "another step forward for US open-weight AI."zdnet.com+1