NVIDIA Releases Nemotron 3.5 Super VL, a 120B Open Multimodal Agent Model
NVIDIA released Nemotron 3.5 Super VL, a 120-billion-parameter open multimodal agent model activating 12 billion parameters.
NVIDIA released Nemotron 3.5 Super VL, an open 120B mixture-of-experts vision-language agent model that activates 12B parameters per token and accepts text, images, documents, and short video. It offers up to a 1 million-token context, OpenMDW-1.1 weights, and NVFP4 and BF16 checkpoints runnable on hardware such as one H200, two H100s, DGX Spark, or one GB200. NVIDIA-reported NVFP4 scores include 89.90 on GPQA-Diamond, 71.07 on SWE-bench Verified, and 77.5 on BrowseComp, while tau3-bench banking scored 17.78. The model was distilled from Nemotron 3 Ultra after training on about 26 trillion tokens; NVIDIA also released Bolt edge models plus datasets and NeMo Gym recipes.