Nemotron 3.5 Super VL

NVIDIA's new open multimodal model, built to power specialized agents that reason over images and text.

Why it matters

An open multimodal model from NVIDIA for building specialized agents that need to read documents, charts and video as well as text. It can be fine-tuned on your own data with NeMo.

Highlights

• Mixture-of-experts with 120B total parameters and 12B active per token. Text, image, document and video understanding with reasoning and tool use in one model.
• NVIDIA reports up to 5x higher agentic throughput than open models in its size category on a single GPU, and 80.8% on the CharXiv chart benchmark.
• Ships base, post-trained and NVFP4-quantized checkpoints with training recipes, data and RL environments, plus open datasets on Hugging Face.
• Inference features include LiLiCorr speculative decoding (up to 2.3x faster generation on DGX Spark), multi-token prediction, and context reuse across agent turns.
• Deployable via NVIDIA NIM and Dynamo; hosted access is listed on OpenRouter.

Taken from the project's own README, docs or launch page. Claims about speed or accuracy are the maker's unless stated otherwise.

Worth knowing: Performance figures and customer results (Scale AI, Ramp, AI21) are NVIDIA's own claims.

Quick facts

Links

• Official site: https://developer.nvidia.com/blog/nvidia-nemotron-3-5-super-vl-powers-specialized-agents-with-multimodal-reasoning
• Repository: https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16

Related ranking

See how established tools compare in Best AI Agents (General & Computer-Use). New releases aren't scored until we review them against the rubric.

What we couldn't verify

• quick_facts.pricing (hosted)
• quick_facts.license (read from model metadata, license text not reviewed)

Checked 2026-10-17 from the model's own pages. Prices and star counts change, so confirm on the official site. Listing here is not a ranking or an endorsement.