Pricing & billing
Rial pricing, the pre-flight hold, settlement, and plan allowance vs wallet credit.
Currency and rate
Prices are in Rial. Each model's base rate is in USD and converted at the daily FX rate; whichever rate is active when a request is priced is the rate recorded on that request.
Hold, then settle
Before any paid request goes out, an amount matching its estimate is held from your balance. The estimate is based on the prompt length and max_tokens — and for embeddings, on the input itself, since there is no completion. When the response completes, the hold is replaced by the real charge for what was actually consumed.
- Estimate → hold from wallet
- Dispatch upstream, receive the response
- Compute the real amount → release the hold, charge for real
max_tokens at the model ceiling holds a large amount per request even when the answer is short. A realistic number means smaller holds and far fewer insufficient_balance rejections.Plan allowance or wallet
| Mode | What it spends | Model access |
|---|---|---|
| Plan (bundled) | The plan's token allowance | The plan's models |
| Credit | Rial balance in the wallet | Every active model |
This choice is per key, not per account — so you can hold plan keys and credit keys at the same time.
Prompt caching
Tokens read from cache are billed at a separate, much lower rate, and cache writes cost slightly more than normal input. Each model's rates are on its own page.
The other rates
Every token lands in exactly one class and is billed at that class's rate; where the model quotes no separate rate for a class, the base input or output rate applies. The classes:
- Reasoning (thinking) tokens — a rate instead of the output rate, never on top of it.
- Image input tokens and generated image tokens — image models price generated image tokens far above text.
- Audio in and out, and cached audio.
- Cache writes with the one-hour TTL (ttl: 1h) — dearer than the five-minute write.
Some models have conditional prices: a higher rate for prompts past a threshold (say 200K tokens), or a different rate inside a UTC clock window. The condition is judged on the whole prompt, cache included, and the request's start time. Every rate, the conditional ones included, is on each model's page in Rial.
Invoices and tax
An invoice is issued for every wallet top-up. VAT is inclusive: the invoice total equals exactly what you paid, and the taxable base is derived from it.
Invoices live in the dashboard.