Pricing
You pay for tokens at the price listed on each model, from prepaid credits. A 5.50% platform fee is added when you buy credits. There are no subscriptions, seats or minimum monthly spend.
Failed requests cost nothing. If a request errors before any output, you pay $0. If a stream breaks mid-way, you pay only for the prompt and the output you received.
Per-model token prices
Every model shows its input, output and cache-read price per 1M tokens. That is the price you pay for that model's tokens.
From $0.0625 to $3.75 per 1M input tokens across 32 models.
5.50% fee on credit purchases
The fee is added on top of the credits you buy and shown before checkout. Every dollar of credit you buy is a dollar you can spend on tokens.
Buy $5.00–$10,000.00 at a time. Your first purchase can be up to $100.00, and purchases are limited to $500.00 per 24 hours, fee included. Contact support to raise the limit.
Holds, then settlement
Before a request runs, its maximum possible cost is held from your balance. When it finishes, you are charged the actual cost and the rest is released.
Your balance can never go negative.
What a credit purchase costs
Fee = credit amount × 5.50%, rounded up to the next cent. Read live from our platform settings.
| Credits | Platform fee | You pay |
|---|---|---|
| $10.00 | $0.55 | $10.55 |
| $25.00 | $1.38 | $26.38 |
| $100.00 | $5.50 | $105.50 |
What a request costs
Cost = (uncached prompt tokens × input price + cached tokens × cache-read price + completion tokens × output price) ÷ 1,000,000, rounded up to the next micro-dollar ($0.000001). Each API response reports the cost in usage.cost.
What is and is not included
- Every listed model on one API key
- Exact cost on every response and in your activity log
- Per-key credit limits and expiry
- CSV export of your request history
- No subscriptions or monthly minimums
- No bring-your-own provider keys: all requests run on our provider accounts
- One price per model, whichever upstream serves the request
- No charge for failed requests or for our retries between routes
- No hidden markup at settlement: you pay the listed model price for the tokens used
Questions
- Do credits expire?
- No. Credits do not expire while your account is open. Unused credits from a purchase in the last 30 days are refundable on request (the platform fee is not). The sign-up credit is not refundable.
- Can I get a refund?
- Contact support. Refunds of unused credits are handled case by case and recorded in your ledger.
- Why hold more than the request costs?
- The hold covers the largest possible output (your
max_tokens). Lowermax_tokensto hold less. - Do I pay for failed requests?
- No. A request that fails before any output (every route down, a timeout, the model at capacity, an invalid request) costs $0, and so do the retries we make between routes. If a stream breaks after output started, you pay for the prompt and the output you received, never more. Details
- What if my balance runs out mid-request?
- A request only starts if its hold fits in your available balance; otherwise the API returns 402
insufficient_creditsand charges nothing. A stream holds its first 8,192 output tokens and grows the hold as it goes; if your balance can't cover more, it stops with that error and you pay only for the output already sent.