Pricing

# Pricing

The price is planned. Nothing is charged until the API opens.

Not open yet. Opens with early access.

## The price

| Item | Price |
| --- | --- |
| Input tokens | $0.10 per 1M |
| Output tokens | Not charged |
| Free credit on a new account | $5 |

The model returns probabilities, not text, so there’s no output to charge for. A call through the HTTP API, the MCP server, a connected app or the console’s playground is the same call at the same price.

## In practice

1,000 input tokens cost $0.0001, so the free credit covers 50 million input tokens. How many a call uses depends on what you send: a restricted-merchant check is small, a receipt match with many candidates is larger. A fraction of a cent is rounded up, so a counted token is never free.

## What isn’t charged

- An answer a rule settles without the model: it uses no tokens.
- A call that fails: a 400, including one refused for a card, Social Security or account number, or a 502.
- A call refused before it runs: a bad key, no credit or the rate limit.
- A replayed `Idempotency-Key`.

## See what you spend

- A metered `POST /v1/understand` returns `x-understanding-input-tokens` and `x-understanding-cost-usd`.
- Batch items and MCP tool calls return no cost headers: read `GET /v1/usage`, or the `get_usage` tool.
- The console shows requests, tokens, spend and credit left. [Console and keys](/api/docs/console)

## When credit runs out

Calls get a 402: “This account is out of credit. Email api@theunderstanding.company to add more.” There’s no card form yet.

## Limits

60 requests a minute per key. [Limits](/api/docs/limits) lists the rest.

Source: /api/docs/pricing