# Nemotron 3 Ultra

Canonical page: https://agentsky.dev/models/nvidia/nemotron-3-ultra-550b-a55b
JSON record: https://gateway.agentsky.dev/v1/models/nvidia/nemotron-3-ultra-550b-a55b

Nemotron 3 Ultra is a text-input model by Nvidia, available through the AgentSky API as nvidia/nemotron-3-ultra-550b-a55b at $0.6963 per million input tokens and $2.7852 per million output tokens, with a 203K context window.

NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE).

## Facts

- **Model id:** `nvidia/nemotron-3-ultra-550b-a55b`
- **Maker:** Nvidia
- **Input modalities:** Text
- **Context:** 203K
- **Max output tokens:** 182,520
- **Aliases:** `nvidia/nemotron-3-ultra-550b-a55b`, `nemotron-3-ultra-550b-a55b`

## Pricing

USD per million tokens on the default plan. Signed-in pages show your own plan's price; every response carries the cost of that call in `usage.cost`.

| Meter | Per 1M tokens |
|---|---|
| Input | $0.6963 |
| Output | $2.7852 |
| Cache read | $0.13926 |
| Cache write | $0.6963 |

## Use it

Anthropic wire base URL: `https://gateway.agentsky.dev`. OpenAI wire base URL: `https://gateway.agentsky.dev/v1`. Authenticate with an AgentSky API key (`ast_…`) and pass the model id.

### OpenAI SDK

```js
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://gateway.agentsky.dev/v1",
  apiKey: process.env.AGENTSKY_API_KEY, // ast_…
});

const completion = await client.chat.completions.create({
  model: "nvidia/nemotron-3-ultra-550b-a55b",
  messages: [{ role: "user", content: "Summarise this ticket." }],
});
console.log(completion.usage.cost); // USD, in the response
```

### Codex

```bash
export OPENAI_BASE_URL="https://gateway.agentsky.dev/v1"
export OPENAI_API_KEY="$AGENTSKY_API_KEY"

codex --model nvidia/nemotron-3-ultra-550b-a55b
```

### OpenCode

```js
// opencode.json — then switch models any time with /models
{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "agentsky": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "AgentSky",
      "options": {
        "baseURL": "https://gateway.agentsky.dev/v1",
        "apiKey": "{env:AGENTSKY_API_KEY}"
      },
      "models": { "nvidia/nemotron-3-ultra-550b-a55b": {} }
    }
  }
}
```

## Questions

### What is Nemotron 3 Ultra?

Nemotron 3 Ultra is a model from Nvidia. NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE).

### How much does Nemotron 3 Ultra cost?

$0.6963 / $2.7852 per million tokens at the prices shown, for input and output respectively. The exact price of every call comes back in the response as usage.cost, so you never have to compute it.

### What is the context length of Nemotron 3 Ultra?

203K tokens, and it writes up to 183K tokens in one reply. These are the limits listed for this model in the current AgentSky catalog.

### Which clients can call Nemotron 3 Ultra?

Anything that speaks the OpenAI-compatible request format. Point the client's base URL at the gateway, use an AgentSky key, and pass nvidia/nemotron-3-ultra-550b-a55b as the model.

### How do I name Nemotron 3 Ultra in a request?

The gateway accepts nvidia/nemotron-3-ultra-550b-a55b, nemotron-3-ultra-550b-a55b. Any of them resolves to the same model, so an existing integration usually needs no change to its model string.

### Can Nemotron 3 Ultra read images?

No image input is listed for this model. Its supported input types are text. For screenshots or photographs, choose a model that lists image input.

### What output can Nemotron 3 Ultra produce?

The listed output types are text. This model returns text. Producing image, video or audio files requires a suitable tool or another model; it is not included in text generation.

### How should I plan a long prompt for Nemotron 3 Ultra?

Plan your instructions, conversation history and source material around the 203K-token context limit, leaving room for the response. The listed response ceiling is 183K tokens. A larger window lets you supply more relevant material; it does not make unrelated content useful.

### What would 10,000 input tokens and 2,000 output tokens cost with Nemotron 3 Ultra?

At the displayed input and output rates, the model-token subtotal is $0.0125334. This example assumes uncached input and excludes additional pricing categories, tool calls and agent runtime. Actual usage determines the final charge.

### How is cached input priced for Nemotron 3 Ultra?

The catalog lists $0.13926 per million cache-read tokens. Cache writes are listed at $0.6963 per million tokens. Repeating a prompt does not by itself guarantee a cache hit; only usage reported as cached receives that rate.

### Does the Nemotron 3 Ultra token price cover an entire agent task?

No. The model price covers the model usage described above. A task can also incur tool and runtime charges, and several model calls may be needed to complete it. Compare total task usage when evaluating cost.

### How do I use Nemotron 3 Ultra on AgentSky?

This model is listed for API use. Use the model ID nvidia/nemotron-3-ultra-550b-a55b with a supported gateway operation. The agent composer is not offered here because the current agent model menu has no compatible match.