docs

Pricing

Modelled per-token prices for the Sieve decision models. Output tokens are free. Final prices follow benchmarking.

These prices are modelled, not final

The figures below come from a hosting-cost model (GPU cost at 50 percent fleet load, FP8, 500-token requests). They will be replaced by measured prices after the fan-out benchmark and before any price is published as final. Do not budget against them without checking.

Decision models

Decision models are priced per 1M input tokens. Output tokens are free.

ModelModelled price per 1M input tokens
sieve-4b$0.025
sieve-9b$0.042 (the same as Jev)
sieve-auto (4B first, 31B on low confidence)$0.050
sieve-31b$0.160
BatchNot available: there is no batch API yet, so no batch discount applies

Notes:

  • usage.input_tokens in each response is what you are billed for: the state counted once, plus each answered question. Image tokens count as input.
  • Asking many questions about one state is cheap because the state is processed once.
  • sieve-auto is billed at the Auto price for the request's input tokens, whether or not any question escalates to 31B.
  • Custom models from Sieve Foundry add a per-decision fee set in your contract.
  • Cost is rounded up once per request: cost = ceil(input tokens x price per 1M / 1,000,000), in millionths of your currency unit.
  • sieve-0.8b, sieve-2b and sieve-27b have no modelled price yet.
  • A price increase is announced at least 30 days before it applies. A price decrease can apply straight away.

Currencies

Your organisation's currency is set by its billing country when you create it, and cannot be changed later. Prices are planned in USD and in INR. The INR list will be fixed and revised quarterly, so it does not move with exchange rates; it is not published yet. INR prices are shown plus 18 percent GST. See Invoices and tax.

Limits and tiers

Rate limits are set per organisation through its tier. Default numbers are not published yet: the console's Limits page shows the limits that apply to your organisation, and you can request an increase there. See Limits.

Enterprise

For large or regulated use there are annual or quarterly prepaid commitments with a custom price book, custom rate limits, Vault licences and Foundry builds. These are by contract. Contact Sieve.

Next steps

Last updated