On April 28 (local time), NVIDIA unveiled its groundbreaking open multimodal model, the 'Nemotron 3 Nano Omni'. This innovative model seamlessly integrates a multitude of functionalities, empowering agents with sophisticated multimodal reasoning prowess. It offers enterprises and developers a streamlined production pathway for crafting multimodal AI agents. Leveraging a cutting-edge 30B-A3B mixture of experts architecture, the model incorporates visual and audio encoders to bolster large-scale reasoning efficiency. Boasting exceptional multimodal perception accuracy, the AI system's throughput soars to nine times that of comparable open omnidirectional models, delivering both cost-effectiveness and high scalability. Presently, several companies have already embraced this model, which is also capable of collaborating with a diverse array of models to bolster sub-agents within agent workflows. Furthermore, the Nemotron 3 series models have amassed over 50 million downloads in the past year alone.