Billing and credits
Smart Inference is pay-as-you-go. No free tier, no subscription, no seat licenses. You buy credits, we deduct them per request. If you don’t like what you see, you stop.
- Minimum top-up $10, paid through Stripe Checkout
- Default monthly cap $100, adjustable down to $10 or up to whatever you set
- Auto-recharge (optional) fires when your balance drops below half your chosen recharge amount
- Receipts per request in the dashboard, with cost, chosen model, routing score, and latency
Account states
inactive- new account, no payment method yet. Cannot call the API.active- payment method on file. Full API access.suspended- paused by support (fraud review, abuse, or on request).payment_failed- auto-recharge failed. Requests are blocked until the card is fixed.
Credits
You start with zero. Buying credits opens a Stripe Checkout session for the amount you pick (minimum $10 / 1,000 cents). Once payment clears, your balance is updated. Every call to /v1/chat/completions is checked against your balance before it routes; if you’re short, the request returns 402 insufficient_credits with the current balance in the error message.
Deductions happen after the response. You are billed for prompt tokens and completion tokens at the rates shown in /v1/pricing.
Every credit movement is recorded in a ledger - you can export it from the dashboard.
Auto-recharge
Enable it from Billing in the dashboard. Pick an amount (for example, $25). Auto-recharge fires when your balance drops below half of that amount. If the card fails, the account moves to payment_failed and requests block; update the card to get back to active.
Auto-recharge is off by default. You keep explicit control.
Monthly spend cap
Every account has a monthly ceiling. The default is $100; the floor is $10; the ceiling is whatever you set. The cap resets on the first day of each calendar month. Once you hit it, requests are refused with 402 until you either raise the cap or the next month starts.
The cap is independent of your credit balance. You can have $500 in credits and a $50 cap - in that case $50 is the monthly spend limit until you raise it.
Partial-response billing for streams
If a streaming response is cut off mid-flight — provider issue, client disconnect, whatever — you are billed only for the tokens that actually arrived. You are never billed for bytes you didn’t receive.
Refunds
Credits are generally non-refundable, but we’ll refund a top-up in full if the account has never made a successful API call and you ask within 30 days. For production mishaps — a runaway loop, an outage that burned credits you didn’t use - write to support with the request IDs.
GPU deployment billing
GPU spot rentals are billed from the same credit balance. Credit holds, minimum sessions, per-GPU pricing, and idle timeouts are covered in the dedicated GPU billing docs:
What you see per request
Each chat completion response carries X-SI-* headers. The one you care about for billing is X-SI-Cost: the estimated cost in USD for that request. This matches the rates shown in /v1/pricing and is the amount deducted from your balance.