NVIDIA Releases Open Nemotron 3.5 Super VL and Bolt Models for Agents, From Servers to Laptops
NVIDIA released Nemotron 3.5 Super VL, an open 120-billion-parameter multimodal model for long-running agents, plus Bolt 74B-A9B, a 2-bit model that fits in 24GB for local agents. A smaller Bolt E4B follows Oct. 23.
By 1EyeAI Desk ·
NVIDIA released Nemotron 3.5 Super VL on Oct. 15, 2026: an open mixture-of-experts model with 120 billion total and 12 billion active parameters that takes text, images, documents and video. BF16 and NVFP4 weights are on Hugging Face, with day-zero vLLM support and availability on Baseten, OpenRouter, ModelScope and NVIDIA NIM.
NVIDIA claims Super VL matches its largest reasoning model on agentic tasks at about one-fifth the parameter count, and runs up to 5x faster than the next-best open model in its class on the Artificial Analysis leaderboard. Those are company claims.
Alongside it came Nemotron 3.5 Bolt 74B-A9B 24GB, compressed from Super VL with native 2-bit quantization so it fits in 24GB and runs locally on RTX and DGX Spark systems through llama.cpp, Ollama, LM Studio, Unsloth or vLLM. NVIDIA says it ranks first among open models in its class on PinchBench and second on BrowseComp (company claims). A smaller Bolt E4B 4GB for voice assistants and game characters is due Oct. 23.
Partner list (company-selected): NVIDIA names organizations post-training or integrating Super VL, including Scale AI, Atlassian, CrowdStrike, Perplexity, Salesforce and Ramp. These are NVIDIA’s examples, not independent tests.
What this means for you: If you build agents that read PDFs, screenshots or charts, Super VL is a free, self-hostable model to benchmark against closed APIs. Bolt 74B is worth a try on a 24GB GPU if you want agents that keep your files off the cloud. Test on your own tasks before trusting the speed claims.