# Rate Limits

> How Kurrens applies rate limits and how to handle 429 responses.

Kurrens applies per-key limits on requests per minute and tokens per minute. Limits depend on your account and the model.
Contact [sales](/contact) for higher limits or a [dedicated endpoint](/docs/dedicated/overview).

## We fail fast instead of queueing

When a model is at capacity in its primary region, the request moves to the next [region](/docs/getting-started/regions)
instead of waiting in a queue. If no region has capacity — or you're over your own limit — Kurrens returns
`429 Too Many Requests` **immediately**. That keeps latency predictable and lets your client retry or fall back right away.

## Handling 429

- Retry with exponential backoff and jitter, starting around one second.
- Respect the `Retry-After` header when present.
- Spread large jobs over time instead of sending bursts.

```python
from openai import OpenAI, RateLimitError

client = OpenAI(base_url="https://api.kurrens.ai/v1")

def complete(messages, attempts=5):
    for attempt in range(attempts):
        try:
            return client.chat.completions.create(model="deepseek-ai/DeepSeek-V4.1-Flash", messages=messages)
        except RateLimitError:
            time.sleep(min(30, 2 ** attempt) + random.random())
    raise RuntimeError("rate limited")
```

The OpenAI SDKs also retry `429` responses automatically (`max_retries`).