Nemotron 3 Ultra
NVIDIAReleased Jun 4, 2026
NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it...
- Context window
- 1M tokens
- Max output
- 16K tokens
- Input
- text
- Output
- text
- Tokenizer
- Other
- Released
- Jun 4, 2026
- Reasoning
- optional
Pricing
Per 1M tokens. The provider price and our flat 3% fee are separate columns — what you pay is their sum.
| Per 1M tokens | Provider | + 3% fee | You pay |
|---|---|---|---|
| Input | $0.500 | $0.015 | $0.515 |
| Output | $2.20 | $0.066 | $2.27 |
| Cache read | $0.100 | $0.0030 | $0.103 |
Benchmarks
Artificial Analysis
- Intelligence index
- 37.8
- Coding index
- 49.3
- Agentic index
- 27.4
Design Arena
| Category | Elo | Win rate | Rank |
|---|---|---|---|
| 3Dmodels | 1213 | 43.6% | #40 |
| ASCII artmodels | 1093 | 35.8% | #44 |
| Codemodels | 1174 | 37% | #60 |
| Data vizmodels | 1167 | 38.7% | #63 |
| Game devmodels | 1195 | 39.3% | #53 |
| SVGmodels | 1140 | 38.6% | #45 |
| UI componentsmodels | 1180 | 38.7% | #54 |
| Websitesmodels | 1140 | 32.6% | #76 |
Head-to-head preference voting. How we filter and rank
Supported parameters
- frequency_penalty
- include_reasoning
- logit_bias
- max_tokens
- min_p
- presence_penalty
- reasoning
- repetition_penalty
- response_format
- seed
- stop
- structured_outputs
- temperature
- tool_choice
- tools
- top_k
- top_p
Call it
OpenAI-compatible: point your SDK at api.openkey.ai/v1 and use model nvidia/nemotron-3-ultra-550b-a55b.
curl https://api.openkey.ai/v1/chat/completions \
-H "Authorization: Bearer $OPENKEY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "nvidia/nemotron-3-ultra-550b-a55b",
"messages": [{"role": "user", "content": "Hello"}]
}'import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.openkey.ai/v1",
api_key=os.environ["OPENKEY_API_KEY"],
)
completion = client.chat.completions.create(
model="nvidia/nemotron-3-ultra-550b-a55b",
messages=[{"role": "user", "content": "Hello"}],
)
print(completion.choices[0].message.content)import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.openkey.ai/v1",
apiKey: process.env.OPENKEY_API_KEY,
});
const completion = await client.chat.completions.create({
model: "nvidia/nemotron-3-ultra-550b-a55b",
messages: [{ role: "user", content: "Hello" }],
});
console.log(completion.choices[0].message.content);Questions
- How much does Nemotron 3 Ultra cost via API?
- Through OpenKey, Nemotron 3 Ultra costs $0.515 per 1M input tokens and $2.27 per 1M output tokens. That is the provider price ($0.500 / $2.20) plus a flat 3% fee — nothing else.
- What is Nemotron 3 Ultra's context window?
- Nemotron 3 Ultra accepts up to 1M tokens of context and returns up to 16K tokens per request.
- Is Nemotron 3 Ultra OpenAI-compatible?
- Yes. Send requests to OpenKey's /v1/chat/completions endpoint with model "nvidia/nemotron-3-ultra-550b-a55b" using any OpenAI SDK — only the base URL and API key change.
- What inputs does Nemotron 3 Ultra support?
- Nemotron 3 Ultra accepts text input and produces text output.
More from NVIDIA
All NVIDIA models →Nemotron 3 Ultra (free)
NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it...
ctx 1Min Freeout Free
Nemotron 3.5 Content Safety (free)
NVIDIA Nemotron 3.5 Content Safety is a compact 4B-parameter multimodal guardrail model from NVIDIA, fine-tuned from Google Gemma-3-4B. It moderates both inputs to and responses from LLMs and VLMs, accepting...
ctx 128Kin Freeout Free
Nemotron 3 Nano Omni (free)
NVIDIA Nemotron™ 3 Nano Omni is a 30B-A3B open multimodal model designed to function as a perception and context sub-agent in enterprise agent systems. It accepts text, image, video, and...
ctx 256Kin Freeout Free
Nemotron 3 Super
NVIDIA Nemotron 3 Super is a 120B-parameter open hybrid MoE model, activating just 12B parameters for maximum compute efficiency and accuracy in complex multi-agent applications. Built on a hybrid Mamba-Transformer...
ctx 1Min $0.088out $0.412