NVIDIA Nemotron 3 Ultra
Nemotron 3 Ultra is the largest model in NVIDIA's Nemotron 3 family, released in June 2026 under the OpenMDW 1.1 license. It is a hybrid Mamba-2/Transformer latent mixture-of-experts with 550B total and 55B active parameters and multi-token prediction. Reasoning can be switched on or off and capped with a token budget.
Specifications
- Context window
- 1Mabout 750,000 words, or 1,500 pages
- Max output
- Not published
- Input price
- $0.50per 1M tokens
- Output price
- $2.20per 1M tokens
- Cached input
- $0.10per 1M tokens
- Weights
- OpenOpenMDW-1.1
- Parameters
- 550B total, 55B activeMoE
- Input types
- Text
- Output types
- Text
- Reasoning mode
- Yes
- Tool calling
- Yes
- Released
- Jun 4, 2026Current
DeepInfra serverless price (DeepInfra serves a 256K context). Fireworks lists Nemotron 3 Ultra (Preview) at $0.60 / $0.12 cached / $2.40. Official context is up to 1M; the default serving config is 256K.
What it costs in practice
| Workload | Monthly cost at list price |
|---|---|
| Support chatbot30,000 conversations a month, 2,000 input and 400 output tokens each | $56 |
| Document summaries2,000 documents a month, 12,000 input and 800 output tokens each | $16 |
| Coding agent1,000 tasks a month, 60,000 input and 6,000 output tokens each | $43 |
Standard rates, no caching or batch discounts. Run your own numbers.
Strengths
- Permissive OpenMDW 1.1 license
- Only 9 of 83 layers use attention, so the KV cache stays small at long context
- Reasoning budget control for cost management
Trade-offs to weigh
- Minimum of 8x B200/H200 or 16x H100 to serve
- Text-only input
Where it fits
NVIDIA positions NVIDIA Nemotron 3 Ultra for:
Where you can use it
- Hugging Face
- Ollama
- DeepInfra
- Fireworks
- build.nvidia.com
Run it yourself
Released under OpenMDW-1.1. Approximate memory needed just to load the weights:
| Precision | Memory for weights |
|---|---|
| 16-bit (BF16/FP16) | 1,100 GB |
| 8-bit (FP8/INT8) | 550 GB |
| 4-bit (Q4_K_M, about 4.85 bits) | 333 GB |
This is a mixture-of-experts model: all 550B total parameters must sit in memory, but only about 55B are used per token, so it runs faster than a dense model of the same size.
Add memory for the KV cache (it grows with context length and concurrent users) and 1-2 GB of runtime overhead. Size it precisely in the VRAM calculator.
Alternatives to consider
| Model | Tier | Context | Input / output per 1M | |
|---|---|---|---|---|
| Mistral Large 3Mistral AI | Advanced | 256K | $0.50 / $1.50 | Compare |
| DeepSeek-V4.1-FlashDeepSeek | Advanced | 1M | $0.30 / $1.20 | Compare |
| MiniMax-M3MiniMax | Advanced | 1M | $0.30 / $1.20 | Compare |
| Gemini 3.8 FlashGoogle | Advanced | 1M | $0.75 / $3.75 | Compare |
Frequently asked questions
How much does NVIDIA Nemotron 3 Ultra cost?
NVIDIA Nemotron 3 Ultra lists at $0.50 per million input tokens and $2.20 per million output tokens, with cached input at $0.10 (checked October 10, 2026). DeepInfra serverless price (DeepInfra serves a 256K context). Fireworks lists Nemotron 3 Ultra (Preview) at $0.60 / $0.12 cached / $2.40. Official context is up to 1M; the default serving config is 256K. Confirm current rates on the provider's pricing page before you commit.
What is the context window of NVIDIA Nemotron 3 Ultra?
NVIDIA Nemotron 3 Ultra accepts up to 1,000,000 tokens of context (about 750,000 words, or 1,500 pages). Long prompts cost more and can be slower, so most apps send only what the task needs.
Is NVIDIA Nemotron 3 Ultra open source?
NVIDIA Nemotron 3 Ultra is an open-weight model released under the OpenMDW-1.1 license. You can download the weights from Hugging Face (nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16) and run or fine-tune them yourself, within the license terms. Open weights isn't the same as fully open source: training data and code may not be published.
Can I run NVIDIA Nemotron 3 Ultra locally?
Yes, if you have enough memory. The weights alone need about 333 GB at 4-bit quantization and about 1,100 GB at 16-bit, plus room for the KV cache. Use the VRAM calculator on LLModel.com to check a specific GPU or Mac.
What is NVIDIA Nemotron 3 Ultra best for?
NVIDIA positions NVIDIA Nemotron 3 Ultra for agents, reasoning, coding, enterprise. It accepts text as input, has a reasoning (thinking) mode and supports tool calling. Test it on a sample of your own tasks before you choose it for production.
Sources
- huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
- deepinfra.com/pricing
- huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/raw/main/config.json
- docs.fireworks.ai/serverless/pricing
- ollama.com/library
Facts checked against these pages on October 10, 2026. Spot an error? Tell us.
Still choosing a model?
Send us your use case and constraints. We'll come back with a shortlist and an evaluation plan you can run on your own data.
- Free and vendor-neutral: no model provider pays for placement.
- Your details go only to partners you agree to.