Overview
- Nvidia’s Nemotron 3 Nano Omni launched Tuesday with open weights and is live on Hugging Face, OpenRouter, build.nvidia.com and partner platforms.
- The 30B-A3B hybrid mixture‑of‑experts folds vision, audio and language into one model, removing separate perception steps to preserve context and cut latency.
- Nvidia says the model reaches up to 9x higher throughput at similar interactivity, posts leading scores on document, video and audio tests, and lowers inference cost.
- The model targets agentic tasks such as computer-use agents, document parsing and audio‑video analysis, with H Company reporting real‑time HD screen understanding.
- Early adopters include Aible, ASI, Eka Care, Foxconn, H Company, Palantir and Pyler, while analysts question broad hyperscaler uptake and expect most use on Nvidia’s stack.