Mansa API requests debit your Mansa Credits balance. There is no separate Mansa invoice line item for each call; usage is metered in the units below and subtracted from the balance you fund in the console.
Rates
These apply to the input and output tokens the service reports for a request.
Token billing
Input tokens include your message, history, and caller system instructions. A summary of older thread context placed in system also counts as input. Output tokens include the assistant reply returned for the turn.
Longer history arrays and thread summaries increase input cost. Store the full conversation in your database and send only the recent window the model needs — see Chat.
Because billing is per token, not per request, you only pay for what a turn actually uses. A short reply costs less than a long one, even when you set a high max_tokens ceiling.
Example: chat turn cost
Suppose one /v1/chat turn uses 2,000 input tokens and 800 output tokens:
max_tokens and cost
max_tokens is an output budget, not a fixed charge. The default is 4096. You can raise it up to 8192 when you need longer answers.
Higher max_tokens allows longer replies but does not bypass rate limits. Throughput is still guarded by requests per minute and concurrent stream limits. What changes is the upper bound on a single reply and the maximum possible cost of that reply if the model uses the full budget.
Rough upper bound on output token cost alone:
Add input tokens for the full turn cost.
If the model hits the output budget, meta.finishReason is "length". Retry with a higher max_tokens or a shorter prompt if you need more text.
Funding and visibility
Top up Mansa Credits from the Billing page. The minimum checkout amount is $5. Usage appears in Usage. When balance is insufficient, requests fail with 402 and context insufficient_credit — see Errors.
Prices may change; the effective catalog version is tied to your account at request time. Contact Mansa for sustained production volume or enterprise terms.