Pricing and margin
NetonAPI prices have a single anchor: the official price. Every price is written as "a share of the official price", so customers can compare easily and operators can do the math.
Where a user's price comes from
When a user calls a model, the price is looked up in this order and the first match wins:
| Layer | When it applies | How it is written |
|---|---|---|
| Negotiated price | The operator negotiated this model for this user | N% of official, e.g. 85% |
| Group price | The user's API key belongs to a billing group (enterprise, dedicated line…) | A group ratio, or per-model prices inside the group |
| Site price | None of the above | Official × site markup (a model can have its own markup) |
In the console, users see each model's price, what share of the official price it is, and whether it is a negotiated price. In the admin console, the Negotiated Pricing page shows which layer each of that user's prices comes from.
An example
Say Claude Sonnet's official price is $3 input / $15 output per million tokens and the site markup is 1.0 (selling at official price):
- A regular user pays $3 / $15.
- The enterprise group has ratio 0.95: $2.85 / $14.25.
- Key account Acme negotiated 85%: $2.55 / $12.75.
One Acme request with 10,000 input and 2,000 output tokens is charged 10,000 × 2.55 / 1,000,000 + 2,000 × 12.75 / 1,000,000 = $0.051 — exactly what the unit prices in their console add up to.
Why you can't sell at a loss
Every upstream channel records its own cost, e.g. "OpenAI direct at 82% of official" or "partner relay at 55%". Before any price is saved — site, group or negotiated — NetonAPI checks the margin on every channel that would actually serve it:
- Below cost: the price can't be saved, and you are told which group, channel and price component loses money.
- Below your minimum margin: also refused.
In the example above, if the channel carrying Claude costs 90% of official, the 85% negotiated price can't be saved — use a cheaper channel or give a smaller discount.
The model pricing list shows each model's worst margin across all group × channel combinations, with losses in red; the overview breaks down revenue, cost and margin by group and channel.
When the official price changes
NetonAPI periodically pulls official prices from a public source (the same one new-api uses by default; models.dev, LiteLLM or OpenRouter also work), lists every increase and decrease, and applies them after an operator confirms. Because every price is a share of the official price, site, group and negotiated prices all follow automatically — the "85%" you promised a customer stays true.
Details
- Cache: cache reads and writes are priced separately; when an upstream has no cache price, cached tokens are billed at the input price, never given away.
- Per call and long context: image models can be priced per call; long-context models can be tiered by input length.
- Ledger: each request reserves an upper bound first and settles on actual usage, with the price snapshot frozen per request — price changes never affect in-flight requests.
- Price history: every change records who, when, and from what to what.