What does a reseller need from an API supplier?

Three things: a price that leaves room for your own margin, metering you can pass through to your customers, and the ability to change upstream without changing what your customers see. The endpoint and the key format should stay yours; the supplier behind them should be swappable.

How does volume pricing work?

The discount rises with cumulative top-up volume, so the tier you reach is a function of how much you have bought, not of how much you promise. Every tier is priced as a percentage off the model vendor's official list price, published per model.

Can I issue separate keys per client or project?

Yes. Keys are created in your account, so you can keep one per project, per environment or per client, and disable a key the moment it leaks. Usage is itemised, which is what makes pass-through billing possible without guesswork.

What happens if an upstream degrades?

The model line is deliberately multi-supplier: the same model can be served by more than one upstream, and we swap the one behind the endpoint when quality or price slips. Because your key and the API shape do not change, the swap does not reach your customers.

How do I start without committing volume?

Top up $10, create a key, and run your own prompts through the models you intend to resell. Measure latency and cost per million tokens at your real concurrency, then move to a volume tier when the numbers from your own workload justify it.

AI API for resellers and agencies: volume pricing without a platform build

If you resell models — to clients, to your own SaaS users, or as a line next to hardware — the work is not calling the API. It is pricing it, metering it, and being able to change supplier without your customers noticing.

Short answer: buy wholesale at a discount against each model's official price, meter with itemised per-call usage, and issue your own keys per project or client from inside your account. The endpoint stays the same while the upstream behind it can be swapped when price or quality moves.

What you actually need from a supplier

Requirement	Why it decides whether the deal works
Discount against official price	Your margin is the gap between what you charge and what the model costs. A vague "cheap rate" cannot be priced into a quote.
Itemised usage per call	You cannot invoice a client for something you cannot see. Tokens in, tokens out, amount billed — per call.
Multiple keys	One key per project, environment or client, revocable the moment it leaks.
Swappable upstream	Quality and price both move. The swap should not require a migration on your side.
Balance you can read	Prepaid balance plus a statement and spend curve beats an invoice that arrives three weeks later.

Margin maths you can do in one line

For each model you resell, note the official list price per million tokens, apply your tier's discount, and compare that with what you charge. Two things are worth measuring properly:

Input and output separately. Most models price output several times higher than input, so a workload with long prompts and short answers looks completely different from the reverse.
What you pay for idle capacity. A prepaid balance is used at your own pace; a subscription is paid whether you use it or not.

Then sanity check the whole line: if your clients' traffic concentrates on three models, your effective cost is the cost of those three — not the headline discount on the catalogue.

Keep the supplier behind an endpoint of your own

Resellers lose customers when a swap becomes visible: a changed URL, a new key format, a different error shape. Design so that your clients talk to something you own — your own gateway, your own key namespace — and the model provider sits behind it. That is what makes a two-supplier arrangement useful instead of theoretical: when one upstream degrades, the traffic moves and the client does not.

Adding a model line next to hardware

Resellers of servers, GPUs and memory already have the customer: someone building an application. An API line is billed per token, needs no stock and no shipping, and can be priced and quoted the same way a part number is. The practical entry point is small — $10 to establish the account and test, then volume tiers once the workload is real.

Get a volume quote

Tell us the models and expected monthly volume and we will quote the tier. Or start at $10, create a key, and price it against your own traffic first.

Top up and get an API key · Model list and discounts · Ask for a volume quote