Billing basics
What a request costs, how the prepaid balance pays for it, and where to see your spend.
Sieve is paid for from a prepaid balance in your organisation's currency. Each request is priced by the input tokens it uses; output tokens are free.
What you pay for
- Input tokens only.
usage.input_tokensin every response is the billed amount: the state counted once, plus each answered question. Asking ten questions about one state in one request is much cheaper than ten requests. - Per model. Each model has a price per 1M input tokens.
sieve-autois billed at the Auto price whether or not a question escalates. Custom Foundry models add a per-decision fee set in your contract. - Rounded once per request:
cost = ceil(input_tokens × price per 1M / 1,000,000), in millionths ("micros") of your currency unit.
Work out a request's cost
Take usage.input_tokens from the response and the model's price per 1M input tokens from Pricing, in micros (for example $0.042 is 42,000 micros):
import math
def cost_micros(input_tokens: int, price_per_million_micros: int) -> int:
return math.ceil(input_tokens * price_per_million_micros / 1_000_000)
# 145 input tokens on a model priced at 0.042 per 1M tokens (42,000 micros):
print(cost_micros(145, 42_000)) # 7 micros = 0.000007 of your currency unitHow the balance works
- Top up. An Admin buys credit on the console's Billing page (minimum 500 INR or 5 USD before GST). Indian customers complete Billing → Tax first. INR wallets can also use auto-recharge.
- Each request reserves its cost before it runs and settles when it finishes, so usage cannot take the balance below zero.
- Out of credit: the API answers 429
insufficient_quotawith noretry-after. Retrying will not help; add credit, and the new balance is seen within a few seconds. - Promotional and granted credit is used first, paid credit last. Each block shows its expiry on Billing.
The prepaid balance caps total spend. An Admin can also set monthly budgets, per key or for all keys together, so a runaway client stops at a known amount (429 key_budget_exceeded or project_budget_exceeded); see Monthly budgets. An Admin can set a low-balance alert on Billing to get an email before the balance runs out.
See your spend
- Home in the console shows requests, tokens and organisation spend for any UTC date range, by model or by API key, with a CSV export. See Usage and exports.
- Billing shows the balance, credit blocks, top-ups and invoices. See Invoices and tax.
Next steps
Last updated