Nemotron 3 Ultra
NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE).
Model ID
nvidia/nemotron-3-ultra- Context window
- 1M
- tokens
- Max output
- 66K
- tokens
- Input
- $0.6
- per 1M tokens
- Output
- $2.40
- per 1M tokens
Cache read $0.12
Quickstart
Call Nemotron 3 Ultra in one request.
Point the SDK you already use at xKiro and pass the model ID. The same key works for every model in the catalog.
SDK guidesimport osfrom openai import OpenAI client = OpenAI( api_key=os.environ["XKIRO_API_KEY"], base_url="https://api.xkiro.com/v1",) res = client.chat.completions.create( model="", messages=[ {"role": "user", "content": "Explain HTTP caching in two sentences."}, ],) print(res.choices[0].message.content)✓ 200 OK
"model": "nvidia/nemotron-3-ultra"
HTTP caching stores copies of responses so later requests can be served without contacting the origin server. Headers such as Cache-Control and ETag decide how long a copy stays fresh and how it is revalidated.
14 input · 42 output tokens · $0.00011 at this model’s rates
Create an API key in the dashboard and export it as XKIRO_API_KEY.
Pricing
What Nemotron 3 Ultra costs.
Billed per million tokens at these rates, from your plan’s budget or your wallet.
$0.6
Input
per 1M tokens
$2.40
Output
per 1M tokens
$0.12
Cache read
per 1M tokens
On a plan
From $0.02 / $0.08 on a plan
Plan: up to −96%
Usage on a plan comes out of your allowance, so Nemotron 3 Ultra costs from $0.02 / $0.08 per 1M tokens when you use the whole allowance — up to 96% less than pay-as-you-go.
Compare plans →Billed fairly
- A request that fails before any output is generated is never billed
- An identical blocking request retried within two minutes replays the first result — no double charge
- If a stream breaks off or you cancel it, you pay only up to that point
Specifications
Limits and capabilities.
- Context window
- 1,000,000 tokens
- Max output
- 65,536 tokens
- Input
- Text
- Output
- Text
- Provider
- NVIDIA
- Model ID
- nvidia/nemotron-3-ultra
- Good for
- Coding · Long context
Capabilities
- Tool useSupported
- VisionNot supported
- Video inputNot supported
- ReasoningSupported
Reasoning effort
- off
- on
Set it with reasoning_effort in the OpenAI format, or thinking in the Anthropic format.
Supported parameters
- max_tokens
- temperature
- top_p
- stop
- frequency_penalty
- presence_penalty
- seed
- stream
- tools
- tool_choice
- response_format
- structured_outputs
- reasoning
- include_reasoning
Compare
Similar models.
Closest in price and context window — same provider first, then alternatives from other providers.
Browse all 69 models| Model | Input / 1M | Output / 1M | ||
|---|---|---|---|---|
| 1M | $0.6 | $2.40 | $0.12 | |
| 1M | $0.15 | $0.5 | $0.02 | |
| 256K | $0.05 | $0.2 | $0.03 | |
| 1M | $0.4875 | $2.44 | $0.0488 | |
| 1M | $0.4875 | $2.44 | $0.0488 | |
| 1M | $0.4875 | $2.44 | $0.0488 | |
| 1M | $0.4875 | $2.44 | $0.0488 |
Start building with Nemotron 3 Ultra.
Create a key, point your SDK at xKiro and call nvidia/nemotron-3-ultra.
No credit card · No auto-renewal · 26 free models, free forever