Prepaid vault, per-second billing, no surprises.
Your Vault is a prepaid balance measured in cents. You load credits into your Vault, and GPU runs draw from it. No credit card charges after each run — just a clean debit from your balance. If your Vault hits zero, new runs are rejected until you top up.
You pay only for the seconds your code is actively executing on a GPU. Cold start time (container pull, initialization) is not billed. Billing begins when your Python code starts and ends when it exits.
cost = execution_seconds x per_second_rate
Every run follows a three-step billing flow to prevent overspend:
1. Reserve
Before execution, we reserve an estimated amount from your Vault based on the GPU tier and a maximum duration (default: 5 minutes). This hold ensures funds are available.
2. Settle
After your code finishes, we calculate the actual cost based on real execution time and debit the exact amount from your Vault.
3. Release
The difference between the reserved amount and the settled amount is released back to your available balance immediately.
Per-second rates by GPU tier:
4090 — $0.000139/sec ($0.50/hr)
A100 — $0.000583/sec ($2.10/hr)
H100 — $0.001056/sec ($3.80/hr)
Example: a 45-second run on an A100 costs 45 x $0.000583 = $0.026 (about 3 cents).
Your dashboard shows a complete ledger of all charges: each run with its duration, GPU tier, cost, and timestamp. Filter by agent, date range, or GPU type. Export as CSV for accounting.
Billing questions? Reach out at [email protected]