PRICING
Pay per token. Nothing else.
No subscription, no seat count, no minimum. You are charged for the tokens a call actually used — at the same rates below whichever provider ends up serving it.
Rates per 1M tokens
Generated from the live catalog — the same numbers the API bills against.
| Model | Provider | Input | Output | 1k in → 500 out |
|---|---|---|---|---|
| openai/gpt-5.4-minichat | OpenAI | $0.05 | $0.4 | $0.00025 |
per 1M tokens
How a call is billed
- 1
Estimated before the call
Your prompt plus max_tokens gives an upper bound on the cost. If your available balance cannot cover it, the request is rejected with 402 and never reaches a provider — a runaway script cannot push the account negative.
- 2
Held, not charged
That estimate is held against your balance while the call runs. Available balance is credits minus everything currently held. Holds live in the database with an expiry, so a crash releases them instead of stranding your credit.
- 3
Settled on actual usage
When the response comes back you are charged for the tokens actually used, and the rest of the hold is returned. Since max_tokens is an upper bound, the settled amount is usually lower than the hold.
- 4
Failed attempts cost nothing
If a provider errors, the request is retried on the next channel carrying that model. Only the attempt that produced a response is billed. Failed attempts are logged at zero cost so you can still see what happened.
Pricing questions
The rates above are what you pay. There is no separate platform fee and no monthly minimum.
Ready to send the first call?
Create a key, change one base URL, and keep the OpenAI client you already have.