NVIDIA · Model
Nemotron 3 Ultra pricing, context window and specs
Nemotron 3 Ultra costs $0.60 per million input tokens and $2.40 per million output tokens, and reads up to 202,800 tokens per request. By blended price it is the most expensive of NVIDIA’s 4 current models here.
- Input / 1M
- $0.60
- Output / 1M
- $2.40
- Cached / 1M
- $0.12
- Context
- 202,800
- Max output
- Not published
- Released
- 2026-06-04
Estimate your cost with Nemotron 3 UltraCompare all NVIDIA modelsCount tokens for Nemotron 3 Ultra
Pricing
Nemotron 3 Ultra API pricing
| Price per 1M tokens | Input | Output |
|---|---|---|
| Standard | $0.60 | $2.40 |
| Cached input (read) | $0.12 | – |
Output costs 4.0× the input price, so long replies drive the bill more than long prompts. If your requests share a long fixed prefix, such as a system prompt or tool definitions, caching cuts that part to 20% of the normal input price.
History
Nemotron 3 Ultra price history
Price increase, +14% blended
- Input
- from $0.50 to $0.60
- +20%
- Output
- from $2.20 to $2.40
- +9.1%
- Cached input
- from $0.10 to $0.12
- +20%
Nemotron 3 Ultra has open weights, so this is the price of OpenRouter’s top-ranked host, which can move when hosts change rather than because NVIDIA changed its price. Every change across all models is in the AI pricing changes log.
Examples
What Nemotron 3 Ultra costs in practice
| Workload | Tokens in / out | Requests/day | Per request | Per month |
|---|---|---|---|---|
| Support chatbot50% cached | 1,500 / 400 | 1,000 | $0.0015 | $45.63 |
| RAG app20% cached | 6,000 / 500 | 500 | $0.00422 | $64.24 |
| Coding agent80% cached | 40,000 / 2,000 | 200 | $0.0134 | $81.76 |
| Document summariser | 8,000 / 600 | 300 | $0.00624 | $56.94 |
Change any number in the cost calculator.
Context
Context window
Nemotron 3 Ultra accepts up to 202,800 tokens per request, roughly 152,100 English words or 304 pages. That budget is shared between your prompt and the reply, and no separate cap on the reply is published.
Filling the whole window with one prompt costs about $0.12 at list price, so for repeated long-document work, retrieval (sending only the relevant parts) or prompt caching is usually much cheaper.
Features
Capabilities
- Image input
- No
- Tool / function calling
- Yes
- Reasoning mode
- Yes
- Prompt caching
- Yes
- Structured output
- Yes
- Open weights
- Yes
- Accepts
- text
- Knowledge cutoff
- Not published
As declared by the provider’s API. It shows a feature exists, not how well it works.
API
Nemotron 3 Ultra model ID
- OpenRouter
nvidia/nemotron-3-ultra-550b-a55b- Open weights
- nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
Use the OpenRouter ID when calling it through OpenRouter; NVIDIA’s own API may name it differently. See the model ID reference for the providers we document.
NVIDIA
Where Nemotron 3 Ultra sits in NVIDIA’s line-up
Ranked by blended price (three parts input to one part output, $1.05 per 1M for Nemotron 3 Ultra), it is the most expensive of NVIDIA’s 4 current models here. One step down, Nemotron 3 Super costs $0.1725 blended (84% less). Unlike Nemotron 3 Ultra, it reads up to 262,144 tokens (202,800 here). See every NVIDIA model and price
| Model | Input / 1M | Output / 1M | Context | Released |
|---|---|---|---|---|
| Nemotron 3.5 Lightning | $0.07 | $0.20 | 262,144 | 2026-08-11 |
| Nemotron 3 Nano 30B A3B | $0.06 | $0.24 | 262,144 | 2025-12-14 |
| Nemotron 3 Super | $0.08 | $0.45 | 262,144 | 2026-03-11 |
| Nemotron 3 Ultra (this page) | $0.60 | $2.40 | 202,800 | 2026-06-04 |
NVIDIA’s current models with a page here, cheapest first. Older and superseded models are listed on the provider page and in the comparison table.
Save
Cheaper alternatives
- Nemotron 3 Nano 30B A3BNVIDIA90% cheaper
- Gemma 4 26B A4BGoogle90% cheaper
- Qwen3.5-9BQwen89% cheaper
- Seed 1.6 FlashByteDance Seed88% cheaper
- GPT-6 LunaOpenAI81% cheaper
Current models from major providers, one per provider, that keep tool calling, a reasoning mode, a usable context window and output limit. Compared by blended price.
Compare
Similarly priced models
- Qwen3.6 27BQwen$0.32 / $3.20
- Mistral Large 4Mistral$0.68 / $2.09
- Seed 2.1 TurboByteDance Seed$0.50 / $2.50
- Grok Build 0.1xAI$1 / $2
- Gemini 3.5 Flash LiteGoogle$0.30 / $2.50
Current models from other major providers, closest blended price.
FAQ
Nemotron 3 Ultra questions
How much does Nemotron 3 Ultra cost?
Nemotron 3 Ultra costs $0.60 per million input tokens and $2.40 per million output tokens, with cached input at $0.12. A typical chatbot reply (1,500 tokens in, 400 out) costs about $0.00186. Use the LLM cost calculator for your own workload.
What is Nemotron 3 Ultra’s context window?
202,800 tokens, about 152,100 English words or 304 printed pages, shared between your prompt and the reply. No separate limit on the reply is published, so the window is the limit.
Does Nemotron 3 Ultra support prompt caching and batch requests?
Yes, cached input is billed at $0.12 per million tokens (20% of the normal input price). No batch price is published for it.
What are cheaper alternatives to Nemotron 3 Ultra?
With the same essentials (tool calling), the cheapest options are Nemotron 3 Nano 30B A3B (90% cheaper), Gemma 4 26B A4B (90% cheaper), Qwen3.5-9B (89% cheaper). Cheaper doesn’t mean equivalent, so test them on your own prompts.
What is the API model ID for Nemotron 3 Ultra?
On OpenRouter it is nvidia/nemotron-3-ultra-550b-a55b. NVIDIA’s own API may use a different name, so check its model list before you deploy. Pass the ID as the model value in each request.
Is Nemotron 3 Ultra open source?
Yes, Nemotron 3 Ultra has open weights, so you can run it yourself or choose between several hosting providers. The prices here are a typical hosted price.