# Credit and limits

## How you are charged

You pay for tokens: the tokens you send, and the tokens the model writes. Each model has its own price per million tokens, in Australian dollars. Prices per million tokens are on the [pricing page](https://www.continuumlabs.ai/pricing).

Every response tells you exactly what it used, in `usage`. The [console](/console) adds it up by day and by model.

```json
"usage": {"prompt_tokens": 412, "completion_tokens": 186, "total_tokens": 598}
```

As a rough guide, 1,000 tokens is about 750 words.

## Credit

Usage is taken from your account's credit as you go. New accounts start without credit. Tell us what you're building and we'll add trial credit.

Before each request reaches the model, we set aside the most it could cost: your prompt plus the longest reply allowed by `max_tokens`. When the request finishes, you pay only for what it used and the rest goes straight back. If your credit doesn't cover that maximum, the request returns `402 insufficient_quota`. Set a lower `max_tokens` to make it fit, or add credit.

Because every request sets its cost aside first, requests sent at the same moment can never spend more than your account holds.

To add credit, or to arrange monthly invoicing for your organisation, [talk to us](/contact?topic=committed).

## Limits

| Limit | Value |
|---|---|
| Request size | 4 MB |
| Longest request | 5 minutes |
| Reply length | 4,096 tokens unless you set `max_tokens`, up to 32,768 |
| Keys per account | 10 |
| Context | Depends on the model. See [models](/docs/models). |

There is no fixed requests-per-minute limit today. If a model is busy you get `429 rate_limited`. Retry with backoff, as described in [errors](/docs/errors). If you expect heavy traffic, tell us first and we will make sure there is capacity.

## Saving money

- Use a smaller model for simple, high-volume work.
- Set `max_tokens` so replies stay as long as you need and no longer.
- Send only the parts of a document the task needs.