Nvidia has launched Nemotron 3.5 Super VL, an open-weight 120-billion-parameter multimodal model optimized for autonomous enterprise AI agents. Built with a Mixture-of-Experts architecture activating 12 billion parameters per token, Super VL matches the reasoning accuracy of larger models while delivering up to five times higher throughput. Nvidia also introduced the compact Bolt 74B and Bolt E4B models for local PC and edge hardware. Enterprise partners including Scale AI and CrowdStrike are post-training Super VL on proprietary domain datasets.
Release of Nemotron 3.5 Super VL
- ▪Nvidia released Nemotron 3.5 Super VL, an open-weight multimodal 120-billion-parameter model designed for continuous enterprise AI agents on October 15, 2026
- ▪Nvidia released base, post-trained, and compressed weights for Nemotron 3.5 Super VL under the OpenMDW license, alongside training recipes and 25 trillion tokens of pretraining data
- ▪Nvidia developed Nemotron 3.5 Super VL in collaboration with the Nemotron Coalition, which includes Mistral AI, Thinking Machines Lab, Perplexity, and Black Forest Labs
Architecture of Nemotron 3.5 Super VL
- ▪Nvidia compresses Nemotron 3.5 Super VL using 4-bit NVFP4 precision, allowing the model to run entirely in 4-bit precision on Blackwell GPUs
- ▪Nemotron 3.5 Super VL uses a Mixture-of-Experts architecture where only 12 billion of its 120 billion total parameters are active per token
- ▪Nemotron 3.5 Super VL parses PDFs, tables, charts, and videos without requiring a separate vision model, achieving an 80.8 percent score on the CharXiv chart benchmark
Nemotron 3.5 Super VL benchmark performance
- ▪Nvidia states that Nemotron 3.5 Super VL matches the accuracy of the larger Nemotron 3 Ultra model on agent instruction and tool-calling tasks at one-fifth the size
- ▪In Artificial Analysis speed benchmarks, InclusionAI's Ling 3.0 Flash and DeepSeek's V4.1 Flash placed second and third behind Nemotron 3.5 Super VL
- ▪Nemotron 3.5 Super VL generates up to five times more tokens per second on agent tasks compared to other open models in its size class, according to Artificial Analysis tests
Enterprise post-training and domain adoption
- ▪Post-training Nemotron 3.5 Super VL on specialized domain data improved model accuracy by 9 to 44 percent across document, cybersecurity, finance, science, and medical tasks
- ▪Organizations including Scale AI, EY, CrowdStrike, Ramp, and AI21 are post-training Nemotron 3.5 Super VL for specialized enterprise workflows such as audit, credit analysis, and cybersecurity
Deployment licensing and API pricing
- ▪Nvidia charges $4,500 per GPU annually for the Nvidia AI Enterprise software license required to run production deployments of Nemotron 3.5 Super VL
- ▪Nvidia offers developers a rate-limited hosted API at no charge, while partner companies set their own prices for hosted Nemotron APIs
Nemotron 3.5 Bolt model variants
- ▪Nvidia announced Nemotron 3.5 Bolt E4B 4GB, a compact model scheduled for release on October 23, 2026, aimed at voice assistants and local devices requiring 4GB of memory
- ▪Nvidia released Nemotron 3.5 Bolt 74B-A9B 24GB, a hybrid Mixture-of-Experts model using 2-bit quantization that activates 9 billion parameters to run local agent tasks on 24GB machines
- ▪Nemotron 3.5 Bolt 74B-A9B 24GB is optimized for Nvidia RTX Spark laptops and DGX Spark workstations, while supporting software frameworks including llama.cpp, Ollama, LM Studio, and vLLM
Story comments
Loading comments…