# Rate limits (/rate-limits)

Each key has a per-minute budget per request class. Each account also has a shared budget, so adding keys doesn't
multiply what an account can send. A request must fit both budgets.

| Class | Requests | Default per key | Default per account |
| --- | --- | --- | --- |
| `read` | `GET`, `HEAD` | 120 / min | 600 / min |
| `write` | other methods, such as `POST /v1/costs/estimate` | 30 / min | 120 / min |

Your agreement may set different limits. The headers are the source of truth for your key.

Budgets are token buckets. A full minute's budget can be spent in a burst, and it refills continuously.

## Headers [#headers]

Every authenticated response carries:

| Header | Meaning |
| --- | --- |
| `X-RateLimit-Limit` | Your budget per minute for this request class. |
| `X-RateLimit-Remaining` | Requests left in the bucket right now. |
| `X-RateLimit-Reset` | Seconds until the bucket is full again. |
| `X-RateLimit-Class` | `read` or `write`. |

## When you hit the limit [#when-you-hit-the-limit]

You get `429` with code `rate_limited` and a `Retry-After` header in seconds. Wait that long and retry.

```python
import time
import requests

def get(session, url, **kw):
    for _ in range(5):
        r = session.get(url, **kw)
        if r.status_code != 429:
            return r
        time.sleep(int(r.headers.get("Retry-After", "1")))
    return r
```

## Spending less [#spending-less]

- Use [conditional requests](/caching). A `304` still counts, but it carries no body.
- Use [bulk exports](/bulk-exports) for whole days of history instead of paging through feeds.
- Most data changes at the provider's polling interval (15 to 60 minutes). Polling the same route more often than
  once a minute gets you the same data.