Nemotron 3.5 Lightning API Pricing
The hosted-endpoint buyer view: per-token billing and when it wins.
Nemotron 3.5 Lightning API access is billed per token on hosted endpoints like build.nvidia.com and third-party providers. There is no license fee, only usage cost. Always confirm the live rate on the provider's page.
This page is the buyer view for hosted API access, which is different from self-hosting the open weights for free. Here you pay a provider to run the model so you do not have to.
You will learn how API access works, how per-token billing is structured, how to get a key, and when the API beats running your own GPUs.
How API Access Works
API access lets you call Nemotron 3.5 Lightning over the internet without owning a GPU. A provider hosts the model and you send requests to an endpoint.
The main first-party option is build.nvidia.com, where NVIDIA serves the model directly. Third-party providers may also host it.
You send a prompt and receive a completion, using a standard chat format. The provider handles the hardware, scaling, and uptime.
This is the same model as the open weights, just run by someone else. You trade a usage bill for zero setup and no operations.
The API is the fastest way to reach production. You skip GPU procurement, driver setup, and serving code, and start sending real requests the same day.
Weighing the Nemotron 3.5 Lightning hosted API against self-hosting? We model per-token cost versus GPU spend on your real usage.
Book a ConsultationThe Per-Token Pricing Model
Hosted APIs charge per token, which means you pay for the text you send and receive. Input tokens and output tokens are usually priced separately.
A token is a small chunk of text, roughly a few characters. Your bill scales with prompt length and response length.
Because this model targets long-running agents with up to 1 million tokens of context, token counts can climb fast. Long histories cost more per call.
We do not quote a rate here on purpose. Prices change, and the only accurate number is the live one on build.nvidia.com or your chosen provider.
How to Get an API Key
To get a key, sign in at build.nvidia.com, open the Nemotron 3.5 Lightning model page, and create an API key from your account. The key authorizes your requests.
Store the key as an environment variable, never inside your source code. A leaked key can run up a real bill.
Send a small test request first to confirm the key works. Then watch your token usage in the provider dashboard.
Third-party providers follow a similar flow. You create an account, generate a key, and point your client at their endpoint.
Always Verify the Live Rate
Check the current price on the provider's own page before you budget. Model pricing shifts, and stale numbers lead to bad forecasts.
build.nvidia.com shows the first-party terms for the model. Read them alongside the model card.
If you use a third-party host, compare its published rate and limits directly. Providers differ on price, speed, and quotas.
In our work evaluating open-weight models for clients, the recurring failure mode is quoting a launch-day price that changed a week later. Pull the live figure every time.
Rate Limits and Quotas
Hosted APIs apply rate limits and quotas that cap how many requests or tokens you can use in a window. Check your provider's limits before you scale.
Limits protect the shared service and shape your architecture. If you plan a burst of traffic, confirm your tier can handle it.
We do not print any limit number here, because these values change and vary by provider and account tier. The live figure on your dashboard is the one that counts.
If you hit a limit, most providers offer a higher tier or a limit increase on request. Plan that lead time before a launch, not during it.
When the API Beats Self-Hosting
The API wins when your volume is low or bursty, your team is small, or you need to launch this week. You pay only for what you use.
Low or spiky traffic makes owning a GPU wasteful. Idle hardware still costs money, while API calls cost nothing when idle.
Small teams gain the most from skipping operations. The provider handles scaling, patching, and uptime for you.
Self-hosting the free weights wins later, once volume is steady and high. Many teams start on the API and migrate when the math flips.
A simple test helps you decide. Estimate your monthly tokens, price them at the live API rate, and compare that to the cost of a GPU that could serve the same load.
Why Input and Output Are Priced Apart
Most providers price input tokens and output tokens at different rates. Understanding the split helps you forecast a bill correctly.
Input tokens are the prompt and any context you send. On a long-context agent, this can be the larger share of your spend.
Output tokens are the text the model generates. Capping response length keeps this side predictable.
When you compare providers, check both rates, not just one. A cheap output rate can hide an expensive input rate on long prompts.
Keeping API Spend Under Control
Control API spend by trimming context, capping output length, and monitoring token usage daily. Small habits prevent large bills.
Send only the context the task needs. A leaner prompt on a million-token model saves real money over many calls.
Set a maximum output length so runaway responses cannot balloon your bill. This also improves latency.
Watch the dashboard and set alerts. Catching a spike on day one is far cheaper than finding it on the invoice.
Cache or reuse stable context where the provider allows it. Resending the same long preamble on every call is a quiet source of waste.
What you need to run Nemotron 3.5 Lightning yourself
Nemotron 3.5 Lightning needs real memory, but it is within reach of a high-end workstation or a couple of professional GPUs — and many teams simply rent instead of buying. Match the path below to whether you want to own the hardware or pay by the hour.
| Path | What it is | Best for | Get started |
|---|---|---|---|
| Call the hosted API | Use Nemotron 3.5 Lightning as a pay-per-token API — zero hardware | Most teams; getting started | OpenRouter |
| Rent GPUs by the hour | Spin up an H100 / A100 for a few dollars an hour | Flexible self-hosting without buying cards | RunPod |
| Local on unified memory | One Mac with enough unified memory to hold a 4-bit quant | A single quiet on-prem box | Apple Mac Studio (M4 Max, 128GB) |
| Local on a workstation GPU | One 48GB pro card, or two 24GB consumer cards | Power users who want hardware they own | NVIDIA RTX 6000 Ada (48GB) |
To put Nemotron 3.5 Lightning to work once it is live, connect a coding client like Cursor (via OpenRouter) or a local runner such as Ollama.
Frequently Asked Questions
- Hosted API access is billed per token, with input and output usually priced separately. There is no fixed license fee. Always confirm the live rate on build.nvidia.com or your chosen provider.
- The first-party option is build.nvidia.com, where NVIDIA serves the model directly. Third-party providers may also host it with their own per-token rates and limits.
- Sign in at build.nvidia.com, open the Nemotron 3.5 Lightning model page, and create a key from your account. Store it as an environment variable, not in your code.
- It depends on volume. The API wins for low or bursty usage and small teams, while self-hosting the free weights wins once traffic is steady and high enough to justify GPUs.
- Model prices change often, so any number printed here would quickly go stale. The only accurate rate is the live one shown on build.nvidia.com or your provider's page.
Right-Size Your Nemotron API Spend
API or self-host is a cost decision, not just a technical one. We model both against your real usage so you stop guessing.
Book a Consultation