# Rate limits (/docs/concepts/rate-limits)

What happens when you send too many requests, and how to back off.



The gateway limits request rates to keep the service fair and stable, and upstream providers apply limits of their own. When you exceed a limit you get `429 Too Many Requests`.

## Handling 429 [#handling-429]

Retry with exponential backoff and jitter: wait about one second, then two, four and so on, adding a random delay each time, and give up after a few attempts.

The official OpenAI and Anthropic SDKs already retry `429` and `5xx` responses with backoff. You can raise the number of attempts:

```python
from openai import OpenAI

client = OpenAI(base_url="https://api.cheapaimodel.ai/v1", max_retries=5)
```

## Avoiding limits [#avoiding-limits]

* Run requests with bounded concurrency instead of firing them all at once.
* Use streaming for long generations so a single slow request does not hold a connection open without output.
* If your workload needs sustained high throughput, email [support@cheapaimodel.ai](mailto:support@cheapaimodel.ai).