Nvidia releases Nemotron 3.5 Super VL for text and visual agent tasks

Nvidia released Nemotron 3.5 Super VL, an open model that combines reasoning and tool use with text, image, document and video understanding. Developers can adapt its weights and deploy it locally, in data centers or in supported cloud environments. The model has 120B parameters but activates 12B per token, using only part of its capacity for each generation step. The release includes base, post-trained and quantized checkpoints, plus training recipes, data and reinforcement-learning environments. Nvidia says it delivers up to 5x higher throughput on a single GPU for agentic tasks than other open models in its size category. The company also reports expected customer gains, including Ramp's forecast of up to 50% higher throughput for receipt processing.