One final price per model.
Every model in the catalog carries its final price — per token for chat, per published unit for media models. That is what a request is charged, and the exact charge comes back on the response. No subscription, no separate platform fee.
Every model, priced.
The price shown is the price a request is charged — the cheapest published endpoint for each model. Open a model for every endpoint, region and data policy.
What a month costs.
Pick a model type and a model, set your traffic, and read the estimate from its final prices.
Your own model, two prices.
Training is charged once per job, per 1M training tokens. The endpoint that serves your model is billed per GPU-hour while it runs, and that is usually the larger cost. Both are final prices.
A final price per 1M training tokens
- Priced by base model and method: LoRA or full fine-tune, each for SFT or DPO.
- Estimate = training tokens × epochs + validation tokens × evaluations, at that price. It is an estimate until the job finishes.
- Before training starts, a hold of the estimate plus 20% is placed on your wallet. Not enough funds: 402, and nothing starts.
- You are charged for the tokens actually trained, with a per-job minimum charge; the rest of the hold is released.
- A cancelled job is charged for the training completed before it stopped. A job that fails on invalid input or a provider error is not charged.
Billed per GPU-hour while running
- Your model runs on a dedicated endpoint. It is billed per GPU-hour, at the hardware's final hourly price per replica, while it is starting or running.
- Prorated per minute and posted every 5 minutes. A stopped endpoint costs nothing and holds nothing.
- Auto-stop: after 15 idle minutes by default, at the maximum hours or the budget you set, or when the wallet runs out.
- Starting an endpoint needs at least 1 hour of its hourly price available in your wallet.
- Requests to your own endpoint carry no per-token charge: the hourly price covers them.
Same prices, four ways to pay.
Plans differ in who pays when and which support tier comes with them — never in the price of a model or the features you get.
A hold is never a charge.
Every amount carries its state, typed differently in the API and drawn differently in the console.
Pre-flight range shown before a request runs. Never charged.
A hold on the wallet for the estimate ceiling, released on completion.
The cost state a response carries while the provider's usage is still being confirmed — labeled as such, never as final.
Rated once provider-reported usage is in. Lives on the request record and the statement; a response never promises a number as final.
Plan comparison.
| Feature | Prepaid | Postpaid | BYOK | Support |
|---|---|---|---|---|
| What you pay | final model price | final model price | BYOK endpoint price | tier price on request |
| Separate platform fee | none | none | none | none |
| Wallet / billing | prepaid wallet | monthly statement | per billing mode | any billing mode |
| Projects, environments, roles | ✓ | ✓ | ✓ | every tier |
| Budgets | warn + hard stop | warn + hard stop | warn + hard stop | every tier |
| SSO / SCIM / approvals / audit | ✓ | ✓ | ✓ | every tier |
| Export files kept | 7 days | 7 days | 7 days | 7 days |
| Response targets | per support tier | per support tier | per support tier | targets, not SLA |
Prices are published per endpoint in the catalog with an effective time; a request is charged at the price that was live when it ran, and that charge is on its request record. Support response numbers are targets, not an SLA; no support tier price is published. Service credits are usage credit, never cash.