# Nemotron 3 Ultra (NVIDIA) — API pricing and providers

> Large open-weight reasoning model (preview). Model id `nvidia/nemotron-3-ultra`. Text model by NVIDIA.

- Page: https://infro.io/models/nvidia/nemotron-3-ultra
- Vendor: NVIDIA (https://infro.io/models/nvidia)
- Modality: Text
- Endpoint through INFRO: `POST /v1/chat/completions` at https://api.infro.io/v1
- List rate: $0.60 per 1M input tokens and $2.40 per 1M output tokens, with cached input at $0.12 per 1M tokens
- Rates checked: 2026-10-05

Rates are NVIDIA's published list price: what the customer's own provider account bills, not an INFRO charge. INFRO is a BYOK control plane; it charges a software subscription and adds 0% to model spend.

## Providers that serve it

| Provider | Model id sent |
| --- | --- |
| Fireworks AI | `accounts/fireworks/models/nemotron-3-ultra-nvfp4` |
| OpenRouter | `nvidia/nemotron-3-ultra` |

## Questions

### How much does the Nemotron 3 Ultra API cost?

NVIDIA's published list rate for Nemotron 3 Ultra is $0.60 per 1M input tokens and $2.40 per 1M output tokens, with cached input at $0.12 per 1M tokens (checked 2026-10-05). The provider account you connect bills you at the rate on that account, so a negotiated rate replaces the list rate. INFRO adds 0% to model spend.

### Which providers serve Nemotron 3 Ultra?

In INFRO's provider registry, Nemotron 3 Ultra can be served by Fireworks AI and OpenRouter. You connect your own account with any of them, and INFRO routes and fails over between the ones you connect.

### What is the Nemotron 3 Ultra model ID?

Request it from INFRO as "nvidia/nemotron-3-ultra". INFRO sends each provider its own id: Fireworks AI "accounts/fireworks/models/nemotron-3-ultra-nvfp4", OpenRouter "nvidia/nemotron-3-ultra".

### How do I call Nemotron 3 Ultra through an OpenAI-compatible API?

Point an OpenAI SDK, or any HTTP client, at https://api.infro.io/v1 with an INFRO gateway key and send model "nvidia/nemotron-3-ultra" to POST /v1/chat/completions. INFRO runs the request on a provider account you connected and returns the provider's response with an estimated cost.

### Does INFRO mark up Nemotron 3 Ultra usage?

No. NVIDIA or the provider you connect bills model usage to your own account. INFRO charges a software subscription for routing, failover, observability, budgets and policy, and takes 0% of model spend.