Skip to main content
Keyed TokenPath requests cost $1 per million tokens, pay as you go. Tokens are counted across the document, question, and answer you send. Output is free, because TokenPath does not generate any. New accounts start with 10 million free tokens, and no card is needed. For volume pricing or an annual commitment, email [email protected]. Small keyless attribution requests are free. They do not reserve credits or appear in account usage. Supplying a valid key selects billed access, even for small requests.

​
Buying credits

Credits are prepaid. Buy them on the Credits page at platform.tokenpath.ai in any whole-dollar amount from 1to1 to 1,000. Every dollar buys 1,000,000 tokens. Credits appear as soon as payment clears.

​
How a keyed request is charged

1

Hold

Before running, TokenPath estimates the request at 4 characters per token and holds that many tokens from your available balance. If your balance cannot cover the estimate, the request fails with 402 insufficient_credits and nothing runs.
2

Run

Attribution runs while the hold is in place.
3

Settle

On success, you pay for the exact number of tokens the model processed, counted with its own tokenizer. Any unused part of the hold goes back to your balance. If the real count is higher than the estimate, the difference comes from your available balance.
4

Release

If the request fails, or you disconnect before it finishes, the whole hold goes back to your balance. Failed requests are never charged.
The estimate is:
For a 1,000-character document, a 50-character question, and a 200-character answer, the hold is ceil(1250 / 4) = 313 tokens, or $0.000313. The settled amount can be higher or lower than the hold, for two reasons. The model reads a small fixed prompt around your text, which matters most for tiny requests. The quickstart request, when sent with a key, holds 69 tokens and settles at 96. Also, real text does not split at exactly 4 characters per token. Repetitive or token-sparse documents often settle below the hold. As a rule of thumb, a million characters of English prose is about 250,000 tokens, or $0.25. Long documents that are split into windows cost a little more, because the question and answer are read once per window.

​
Check your balance

available_tokens is what you can spend now. reserved_tokens is held by requests that are still running. Each held token is either charged or returned when its request finishes.

​
Track spend

GET /v1/me/usage lists your billing events. Each state change is its own event, so one successful request produces a reserved event and then a settled event:
Costs are in microdollars, or millionths of a dollar. Divide by 1,000,000 to get USD, so 96 is 0.000096.At0.000096. At 1 per million tokens, microdollars equal tokens.