Skip to main content
Your workspace has one prepaid wallet. Everything you spend on Omnia (shared inference, dedicated GPU time, and fine-tuning) is deducted from it. Balances are tracked internally in integer micro-cents (1,000,000 micros = $1.00), so every debit and credit is exact.

​
Topping up

In the dashboard, go to Usage and add funds with a card. Funds are available immediately, and there is no minimum commitment; top up only what you need.
New accounts can claim a one-time $1 welcome credit from the Get-started checklist by connecting a card; nothing is charged. It’s enough for the setup check and your first requests.
Payments are processed by Stripe; Omnia never stores your card number.
Doing one normal top-up is also how you save a card for auto-reload. Once a card is on file, Omnia can top up your wallet automatically when it runs low.

​
Your balance

The dashboard distinguishes total balance from what’s actually spendable right now: New requests are gated on available balance, not total balance, because holds reserve funds that in-flight requests may still consume.

​
Holds

Before a per-token request runs, Omnia places a short-lived hold on your wallet, sized to the request’s worst-case cost:
  • the prompt token estimate at the input rate, plus
  • the maximum possible output at the output rate,
priced at the sell price (including markup). The maximum output is bounded by the model’s context window (the provider rejects a larger max_tokens before any billing happens), so a hold can never balloon beyond what the request could actually cost. When the request settles, the hold is replaced by the exact charge for the tokens actually used, and any excess reservation is released. Holds are released automatically; you never manage them by hand. This mechanism is what lets Omnia safely gate a near-empty wallet without ever overcharging.
Settlement is idempotent, keyed on the request id, so a request is never double-charged, even if a settle is retried.

​
Streaming and cancellation

For streaming responses, Omnia bills the tokens actually produced, including any partial output on a stream that is cancelled or disconnects. You’re charged for exactly what was generated: never for nothing, and never for tokens that weren’t produced.

​
Running low

Your available balance can dip to at most −$1.00, a small buffer so a request already in flight won’t fail over a rounding cent. A wallet at or below that buffer refuses new requests with 402 insufficient_balance rather than letting you overspend. To avoid interruptions:

Keep a buffer

Maintain a balance comfortably above your typical burst spend.

Enable auto-reload

Automatically top up from a saved card when you drop below a threshold.

​
Dedicated endpoints and your wallet

Dedicated endpoints draw on the same wallet but are billed per GPU-hour rather than per token:
  • Deploying requires enough prepaid balance to cover a minimum runway (default one hour). If your balance falls short, deployment is refused and the message states the exact shortfall.
  • A running endpoint whose wallet becomes exhausted is auto-stopped to prevent further charges.
See Dedicated endpoints for the full lifecycle.

​
Refunds

If a payment is refunded, the corresponding credit is reversed from your wallet automatically; no manual adjustment needed.