On April 29, 2026, NVIDIA introduced Nemotron 3 Nano Omni—a cutting-edge, open-source, full-modality model designed with a core focus on "seamless native multimodal comprehension paired with highly efficient inference." This innovation aims to elevate both the efficiency and performance of AI agents tailored for enterprise use. By merging visual, audio, and language processing into a unified framework, the model leverages a 30B-A3B Mixture of Experts architecture. This approach not only enhances inference speed, achieving a remarkable 9-fold increase in throughput, but also dramatically cuts down on computational resource usage and overall inference expenses. Nemotron 3 Nano Omni is well-suited for applications involving the intelligent analysis of intricate documents, as well as comprehensive video and audio interpretation. At present, the model has already been embraced by numerous AI and software enterprises, with industry leaders like Dell and Oracle actively engaged in its evaluation process.