NVIDIA Launches Nemotron3 Nano Omni Model: Unified Processing of Video, Audio, Images, and Text to Boost Multimodal Inferencing Efficiency

April 29: NVIDIA officially launched Nemotron 3 Nano Omni, the latest addition to its Nemotron 3 series. The model integrates unified multimodal reasoning into a single, efficient open-source system. NVIDIA explains that agentic systems typically need a single perception-to-action loop spanning screens, documents, audio, video, and text—but they still rely on fragmented model chains: separate visual, audio, and text technology stacks. This boosts reasoning hops and orchestration complexity, driving up costs while eroding cross-modal contextual consistency. Nemotron 3 Nano Omni is designed to replace this fragmented visual-language-audio (VLA) stack, serving as a multimodal perception and context sub-agent within agentic systems. On accuracy, the model topped the document intelligence leaderboard and led in video and audio understanding rankings. Against the open industry benchmark MediaPerf for video understanding, it delivered the highest throughput across all tasks and the lowest inference cost for video-level annotation tasks. In performance testing (fixed per-user interaction threshold): - Video inference: Nemotron 3 Nano Omni delivered higher total system throughput, with effective capacity up to ~9.2x that of other open-source omni models; - Multi-document inference: It reached effective capacity up to ~7.4x that of peers. NVIDIA notes the model aims to replace traditional multi-model concatenation architectures, cut inference complexity and costs, and accelerate multimodal AI adoption across finance, healthcare, research, media, and other sectors.