What will the API cost me?

Pick a job, change the numbers to match yours, and see a month's API bill on every model we have prices for, with its score beside it.

An estimate, not a quote. Prices are as each maker listed them on the date below the table. We did not run these models. Real bills depend on retries, tool calls and long prompts. Nothing you type is stored or sent.

Start from a job

Example sizes we chose to show typical jobs, not measurements. Change any number below.

Size of the work

Tokens in per taskEverything sent: instructions, documents, the chat so far.

Tokens out per taskThe replies you get back.

Thinking tokens per taskHidden reasoning, billed as output. 0 if thinking is off.

Tasks a monthChats, documents, sessions or rows.

1,000 tokens is about 750 English words.

Discounts and price steps

Share of input that repeats (cached), %The same instructions or document sent again each time. Most makers charge far less for it.

Batch: results can wait up to a dayHalf price on models with a batch price.Long promptsTick if one prompt goes past the model's short-context limit (OpenAI models, or over 100,000 tokens on Haiku 5.5).

Monthly cost per model, cheapest first

#	Model	Score	A month	Per task

How we worked this out

fresh input  = tokens in × (1 − cached share)
cached input = tokens in × cached share
input cost   = (fresh × in price + cached × in price × cache-read rate)
               × batch rate ÷ 1,000,000
output cost  = (tokens out + thinking tokens) × out price × batch rate ÷ 1,000,000
per task     = input cost + output cost
a month      = per task × tasks a month

Worked example

Batch rate is 0.5 when batch is ticked and the model has a batch price, otherwise 1. It applies to cached input too.
Long prompts multiply the list price by the model's step: OpenAI models ×2 in and ×1.5 out, Haiku 5.5 ×5, Opus 5.5, Sonnet 5.5 and Fable 5.1 ×1 (no extra charge). A model with no long-prompt price in our file drops out of the ranking.
Thinking is billed at the output price. Each model counts tokens its own way, so the same text can cost a little more on one model than another.
Not included: cache-write premiums, Gemini's hourly cache storage fee, retries, tool-call overhead, tax, free credits and volume deals.
Where a model has no cache or batch price in our file, we show the price you would pay without that discount and say so in its row.
The same maths as the Learn cost tools. Amounts of $10 or more are rounded to the dollar.