NVIDIA today announced the launch of Nemotron 3 Nano Omni, an open multimodal model that unifies vision, audio, and language processing into a single system. The model delivers up to 9x higher throughput than current open omni-models, dramatically reducing latency and cost for enterprise AI agents.
According to NVIDIA, the 30B-A3B hybrid mixture-of-experts architecture with Conv3D and EVS enables agents to handle video, audio, images, text, and documents simultaneously—outputting only text. This eliminates the need for separate models that slow down inference and fragment context.
“To build useful agents, you can’t wait seconds for a model to interpret a screen,” said Gautier Cloix, CEO of H Company, an early adopter. “By building on Nemotron 3 Nano Omni, our agents can rapidly interpret full HD screen recordings — something that wasn’t practical before. This isn’t just a speed boost: It’s a fundamental shift in how our agents perceive and interact with digital environments in real time.”
Breaking Efficiency Frontier
Nemotron 3 Nano Omni now tops six leaderboards for complex document intelligence, video, and audio understanding. The model is available starting April 28, 2026 via Hugging Face, OpenRouter, build.nvidia.com, and 25+ partner platforms.
Early adopters include Aible, Applied Scientific Intelligence (ASI), Eka Care, Foxconn, H Company, Palantir, and Pyler. Companies evaluating the model include Dell Technologies, Docusign, Infosys, K-Dense, Lila, Oracle, and Zefr.
Background: The Multimodal Bottleneck
Traditional AI agent systems rely on separate models for vision, speech, and language. When data passes from one model to another, context fragments and latency multiplies. For example, a customer-support agent processing a screen recording while analyzing uploaded call audio and checking data logs would require multiple inference passes—each adding cost and potential inaccuracies.
Nemotron 3 Nano Omni solves this by combining vision and audio encoders within a single architecture (30B-A3B hybrid MoE with 256K context). This allows agents to perceive and reason across modalities in one pass, reducing repeated inference and keeping context intact.
What This Means for Enterprise AI
For enterprises and developers, Nemotron 3 Nano Omni provides a production-ready path to build faster, more accurate multimodal agents. The 9x throughput gain translates directly into lower cost and better scalability without sacrificing responsiveness.
The model functions as the “eyes and ears” in a system of agents, working alongside larger models like Nemotron 3 Super and Ultra or proprietary models. This flexibility enables organizations to maintain control over deployment while achieving best-in-class accuracy at low cost.
Key Specifications at a Glance
- Input modalities: Text, images, audio, video, documents, charts, graphical interfaces
- Output modality: Text only
- Architecture: 30B-A3B hybrid MoE with Conv3D, EVS
- Context window: 256K tokens
- Efficiency: Up to 9x higher throughput than other open omni-models
- Availability: April 28, 2026 via Hugging Face, OpenRouter, build.nvidia.com, and 25+ partners
“This isn’t just a speed boost,” Cloix reiterated. “It’s a fundamental shift in how our agents perceive and interact with digital environments in real time.”