Nemotron 3.5 Super VL
NVIDIA released Nemotron 3.5 Super VL, an open 120B parameter multimodal mixture-of-experts model with 12B active parameters, built for long-running enterprise agents.
It takes text, images, documents, GUI screenshots and video and returns text, using a hybrid Mamba-Transformer design with multi-token prediction and a 1M token context. NVIDIA says it matches its largest reasoning model on multimodal agentic tasks at about one-fifth the parameter count, and is 5x faster than the next best open model in its class. Two edge models, Nemotron 3.5 Bolt 74B-A9B 24GB and Bolt E4B 4GB, were announced alongside it.
- Date
- Thursday 15 October 2026
- Lab
- NVIDIA
- Kind
- open-weights
- Access
- open weights
- Price
- $0.10 / $0.40 per 1M tokens on OpenRouter, plus a free endpoint
Figures
| Measure | Value | Measured by |
|---|---|---|
| Total parameters | 120B 12B active, per OpenRouter listing | company |
| Context window | 1M tokens | company |
| Speed vs next best open model in class | 5x faster NVIDIA claim, benchmark details not read | company |
Built with the Nemotron Coalition. Bolt 74B-A9B is available now and Bolt E4B is due Oct 23.
Sources
- blogs.nvidia.com/blog/nemotron-3-5-super-vl/
- openrouter.ai/nvidia/nemotron-3.5-super-vl
- techmeme.com/261015/p34#a261015p34
This record was checked and corrected against its sources on 15 October 2026. How we check