Introducing NVIDIA Nemotron 3 Nano Omni: Long-Context Multimodal Intelligence for Documents, Audio and Video Agents

Benchmark highlights Building on Nemotron Nano V2 VL, Nemotron 3 Nano Omni delivers substantial visual gains and adds entirely new audio and video+audio capabilities - while also leading another open-weights omni model, Qwen3-Omni, in many domains.

Efficiency highlights Compared to other open omni models with the same interactivity, Nemotron 3 Nano Omni delivers 7.4x higher system efficiency for multi-document use cases and 9.2x higher system efficiency for video use cases Figure 1. Total system throughput for multi-document and video use cases sustained by each model at a fixed per‑user interactivity threshold (tokens/sec/user)

At a high level, Nemotron 3 Nano Omni is aimed at five classes of workloads:

Figure 2. Model architecture of NVIDIA Nemotron 3 Nano Omni 30B-A3B

The SFT stages are trained on NVIDIA H100 , scaling from 32 to 128 nodes depending on the stage. The stack uses Megatron-LM , Transformer Engine , and Megatron Energon , with tensor parallelism, expert parallelism, sequence parallelism, context parallelism for the long context stages, online sequence packing, and selective activation recomputation.

Post-SFT reinforcement learning uses NeMo-RL and NeMo Gym with a Megatron backend. The RL infrastructure used a Ray-based distributed setup across B200 and H100 clusters , plus multimodal deduplication, so repeated rollouts do not multiply image, video, and audio memory.

The Nemotron 3 Nano Omni model is specifically trained for agentic computer use, enabling it to assist with tasks in graphical user interface (GUI) environments. Its capabilities include interpreting screenshots, monitoring the state of the user interface, grounding its reasoning in on-screen visuals, and helping with action selection or workflow automation.

As one example, we generated approximately 11.4M synthetic QA pairs (~45B tokens) from a large corpus of real-world PDFs using NeMo Data Designer . This dataset is used to strengthen long-context document reasoning during post-training and delivers a 2.19× improvement in overall accuracy on MMLongBench-Doc.

Many enterprise and developer workflows depend on mixed audio and visual evidence: screen recordings with narration, training videos, meetings with slides, tutorials, product demos, customer support captures, and long-form video archives. Nemotron 3 Nano Omni is built to reason over those inputs jointly.

Nemotron 3 Nano Omni is trained on an enhanced dataset that emphasizes high-quality reasoning across multiple modalities. We significantly expand task coverage and introduce synthetic data for complex reasoning scenarios where public datasets are limited. To support this, we build task-specific, multi-stage pipelines for scalable synthetic data generation.

This is not only about OCR. The model is positioned for long, messy, high-value documents where understanding depends on layout, tables, figures, formulas, section structure, and cross-page references. Think contracts, technical papers, reports, manuals, multi-page forms, or compliance packets. The model can handle 100+ page documents.

Nemotron 3 Nano Omni includes strong speech understanding capabilities that enable high-quality transcription across diverse audio conditions. It handles long-form audio with varying speakers, accents, and background noise. These capabilities can be integrated into broader workflows, allowing spoken content to be transcribed, analyzed, and combined with other modalities for tasks like summarization, question answering, and cross-modal reasoning.

The model is designed for more than perception. It excels at reasoning-intensive tasks that require synthesizing information across long context windows, multiple modalities, and structured or semi-structured evidence. It can carry out multi-step reasoning, perform calculations, and connect signals from text, images, tables, and other inputs to arrive at coherent, well-supported answers.