Docs menu

Docs / Concepts

Pricing

Four public rates per model, charged per token, rounded up per request, with a minimum per successful request.

Last updated


Four rates per model

Each chat model has four public rates in USD per 1 million tokens:

RateApplies to
InputInput tokens that were not read from or written to the cache
Cache readInput tokens the model read from its prompt cache
Cache writeInput tokens written to the prompt cache
OutputOutput tokens, including reasoning tokens

Cache writes are billed at the input rate, for both the 5-minute and the 1-hour lifetime. Models that report the lifetime still show cache_write_1h in GET /v1/models, with the same value as the input rate.

The live rates of every model, with the official price and the actual discount, are on the pricing page and in GET /v1/models.

How a request is charged

fresh input = input total − cache read − cache write − cache write 1h
cost        = fresh input × input rate
            + cache read × cache read rate
            + cache write × cache write rate
            + cache write 1h × cache write 1h rate
            + output × output rate              (rates per 1M tokens)
charged     = max(cost rounded up to $0.000000001, minimum per request)
  • Costs are counted in units of $0.000000001 and rounded up once per request.
  • A minimum of $0.00002 applies to each successful request. A model can have its own minimum; it is shown on the model's page.
  • Cache rates apply only when the model reports cache tokens for your request. Cent AI never estimates or assumes cache usage.
  • A request that fails before a model accepts it costs $0.
  • Each request uses the prices in effect when Cent AI received it.

You can recompute every charge yourself from the rates and the token counts in the response usage.

Worked examples

The rates below are examples to show the arithmetic, not current prices.

Small request under the minimum. Input rate $0.0075 and output rate $0.030 per 1M tokens. 500 input tokens and 50 output tokens cost 500 × 0.0075 + 50 × 0.030 = 5.25 millionths of a dollar, $0.00000525. That is below the minimum, so the request is charged $0.00002.

Large cache read. Input $0.10, cache read $0.01 and output $0.50 per 1M tokens. The response reports prompt_tokens 50,000 with cached_tokens 40,000, and completion_tokens 2,000. Fresh input is 10,000 tokens.

TokensRate per 1MCost
10,000 input$0.10$0.001000
40,000 cache read$0.01$0.000400
2,000 output$0.50$0.001000
Charged$0.002400

Price changes

Prices can go down or up. A new price applies from the moment it is published in the changelog, which lists the model, the old and new rates and when the change took effect. A request already received keeps the price it started with.

Balance and packs

Charges come from your balances in a fixed order: promotional credit, then budget packs, then pay-as-you-go balance. Packs use the same four rates. See Balance.