Nemotron-3
About
The Nemotron-3 8B family of foundation models is tailored for enterprise-level AI solutions, enabling the creation of generative AI applications. This collection features types including base, chat, and Q&A models, each optimized for distinct tasks. The Nemotron-3-8B-Base model is designed for customization, supporting fine-tuning and continuous pretraining to fit specific domain needs. Chat variants—Nemotron-3-8B-Chat-SFT, Chat-RLHF, and Chat-SteerLM—are built for chatbot deployment with varying customization levels. The Nemotron-3-8B-QA model excels in question-and-answer contexts. All models integrate with the NVIDIA NeMo framework for seamless customization and deployment, supporting 53 human and 37 programming languages, and are trained on datasets with trillions of tokens 145.
Specifications(9 models)
| Model | Released | Context | Parameters | Vision | Multimodal | Fn Calling | Tool Use | Structured Outputs |
|---|---|---|---|---|---|---|---|---|
| Nemotron 3 Nano | 2026-03 | 256K | 3.97B | No | No | Yes | Yes | No |
| Nemotron 3 VoiceChat | 2026-03 | — | 12B | Yes | Yes | No | No | No |
| Nemotron 3 Super-120B-A12B | 2026-03 | 1M | 120B | No | No | No | No | Yes |
| Nemotron 3 8B | 2026-03 | 4K | 8B | No | No | No | No | No |
| Nemotron 3 Super | 2026-03 | 1M | 120.6B (12.7B active) | No | No | No | No | No |
| NVIDIA Nemotron 3 Super 120B | 2026-03 | 262K | 120B total, 12B active | No | No | No | No | Yes |
| Llama 3.3 Nemotron Super 49B v1 | 2025-06 | 128K | 49B | No | No | No | No | No |
| Llama 3.1 Nemotron 70B Reward | 2024-10 | 4K | 70B | No | No | No | No | No |
| Nemotron 3 Ultra | 2024-09 | 128K | — | No | No | No | No | No |
Available From(4 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| Nemotron 3 Super-120B-A12B | OpenRouter | $0.09 | $0.45 | Serverless |
| NVIDIA Nemotron 3 Super 120B | DeepInfra | $0.1 | $0.5 | Serverless |
| Nemotron 3 8B | Microsoft Foundry | $0.37 | $1.1 | Provisioned |
Frequently Asked Questions
- What is Nemotron-3?
- The Nemotron-3 8B family of foundation models is tailored for enterprise-level AI solutions, enabling the creation of generative AI applications. This collection features types including base, chat, and Q&A models, each optimized for distinct tasks. The Nemotron-3-8B-Base model is designed for customization, supporting fine-tuning and continuous pretraining to fit specific domain needs. Chat variants—Nemotron-3-8B-Chat-SFT, Chat-RLHF, and Chat-SteerLM—are built for chatbot deployment with varying customization levels. The Nemotron-3-8B-QA model excels in question-and-answer contexts. All models integrate with the NVIDIA NeMo framework for seamless customization and deployment, supporting 53 human and 37 programming languages, and are trained on datasets with trillions of tokens 145.
- How many models are in the Nemotron-3 family?
- The Nemotron-3 family contains 9 models.
- What is the latest Nemotron-3 model?
- The latest model is Nemotron 3 Nano, released in 2026-03.
- How much does Nemotron-3 cost?
- Nemotron-3 models range from $0.09/1M to $0.37/1M input tokens depending on the model and provider.
- Is Nemotron-3 open source?
- 2 of 9 Nemotron-3 models are open source.
Related Model Families
Models(9)
Nemotron 3 Nano
Nemotron 3 VoiceChat
Nemotron 3 Super-120B-A12B
Nemotron 3 8B
Nemotron 3 Super
NVIDIA Nemotron 3 Super 120B
Llama 3.3 Nemotron Super 49B v1
Llama 3.1 Nemotron 70B Reward
Nemotron 3 Ultra