Nemotron 3 Nano Omni 30B A3B: benchmark results
Nemotron 3 Nano Omni 30B A3B supports video, audio, image and text input with text output, 256K context, and approximately 3B active parameters per generated token. The publisher describes a 31B hybrid MoE backbone; encoder processing adds modality-dependent work. Provider: NVIDIA. Released 2026-04-28. Access: Open.
Aggregate Rating 1576 ± 18, rank #514 of 1640 rated models, from 87 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OCRBench v2 | 65.8 | Average (self-reported) | 92.3 |
| OCRBench-V2 (en) | 65.8 | Score (self-reported) | 88 |
| MMLongBench-Doc - Accuracy | 57.6 | Accuracy (%) | 77.8 |
| AA IFBench | 63.2 | Accuracy (%) | 74.4 |
| AA Omniscience - Software Engineering (SWE) - Kotlin | 20 | Accuracy (%) | 63 |
| AA Omniscience - Software Engineering (SWE) - Rust | 50 | Accuracy (%) | 59 |
| OCRBench-V2 (zh) | 59 | Score (self-reported) | 56 |
| ODU-Bench (Audio-Only) - Demand Localization | 39.3 | Temporal IoU (%; overlap of the predicted and reference dema | 53.8 |
| Roboflow Vision Evals - Visual Understanding | 64.18 | Pass Rate (%) | 52 |
| AA Omniscience - Software Engineering (SWE) - Swift | 32 | Accuracy (%) | 51.2 |
| Tau2-Bench Telecom | 45.32 | Success Rate (%) | 50.6 |
| AA Omniscience - Software Engineering (SWE) - Java | 15 | Accuracy (%) | 49.8 |
Interactive version: theaggregate.ai/model?slug=nemotron-3-nano-omni-30b-a3b · How It Works · Data refreshed daily, snapshot 2026-10-13.