Nemotron 3 Nano Omni
NVIDIA's Nemotron 3 Nano Omni: a 30B MoE multimodal reasoning model with a 256K context. Unifies text, image, audio, and video with open weights.
Model Overview
Capabilities, design details, and architectural traits
Nemotron 3 Nano Omni - NVIDIA omnimodal reasoning model
Nemotron 3 Nano Omni is an NVIDIA open multimodal reasoning model that processes text, images, audio, and video together in a single inference pass. The model is presented as a 30B total-parameter Mixture-of-Experts (MoE) family member that activates a smaller subset of parameters per forward pass and is released with weights and training recipes.
| Trait | Model-specific fact |
|---|---|
| Architecture family | Hybrid NemotronH MoE combining Mamba-transformer backbone with specialized encoders for vision and audio (RADIO vision tower and Parakeet sound encoder) |
| Parameter footprint | 30B total parameters with approximately 3B active parameters (MoE activation) |
| Multimodal scope | Natively accepts and reasons over text, images, audio and video within a shared multimodal context |
| Context length | Supports very long context (documented up to 256K tokens) |
| Deployment & openness | Released with open weights, training recipes and data (where redistributable); supported for conversion and standard fine-tuning (SFT, LoRA) |
| Production focus | Designed for agentic and production multimodal workflows, with engineered inference paths and partner-hosted endpoints |
Defining technical idea
Nemotron 3 Nano Omni combines a hybrid MoE language backbone with modality-specific encoders so perception and reasoning run inside a single, unified model rather than chaining separate perception and reasoning models. This integration is the model's central differentiator.
Notable operational notes
The release emphasizes open assets (weights and recipes) and multiple deployment paths including cloud partners and on-prem containers, and documents quantization and inference workflows for production use.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | NVIDIAHost Nemotron 3 Nano Omni | ||
|---|---|---|---|
| General Info | |||
| Provider | NVIDIA | Anthropic | Anthropic |
| Release Date | April 29, 2026 | June 9, 2026 | May 28, 2026 |
| Knowledge Cutoff | — | — | — |
| Context & Limits | |||
| Context Window | 256K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.07 Best Input Pricing | $10 | $5 |
| Output Pricing | $0.30 Best Output Pricing | $50 | $25 |
| Modalities | |||
| Inputs | textaudioimagevideo | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 14.9 | 59.9 Best Intelligence Index | 55.7 |
| Coding Index | — | 76.5 Best Coding Index | 74.3 |
| Agentic Index | — | 52.8 Best Agentic Index | 47.2 |
GPQA Benchmark
Graduate-level reasoning and expert Q&A evaluation.
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
IFBench Instruction
Evaluation of strict instruction following.
Benchmark scores are independent evaluations sourced from Artificial Analysis. Intelligence, Coding, and Agentic indices reflect composite performance ratings (0–100).