NVIDIA Launches Open-Source Multimodal Model Nemotron 3 Nano Omni

NVIDIA announced the release of its open-source multimodal model, Nemotron 3 Nano Omni, on the X platform today. According to Odaily, the model features a 30B-A3B mixture of experts architecture (MoE) and supports a 256K context, enabling unified processing of video, audio, image, and text inputs. Compared to similar open-source models with interactive capabilities, Nemotron 3 Nano Omni offers a ninefold increase in throughput, significantly reducing inference costs and enhancing scalability. The model is now available on Hugging Face, OpenRouter, and NVIDIA NIM, and has been adopted by companies such as Aible, Applied Scientific Intelligence, and H Company.