# Rate Limits
Source: https://docs.mserve.ai/docs/resources/rate-limits
Summary: Per-key limits, 600 requests per minute and 20 concurrent by default. OpenAI header names. 429 with Retry-After.
Availability: available
Last reviewed: 2026-08-25

## Defaults

| Limit               | Default                        | Scope   |
| ------------------- | ------------------------------ | ------- |
| Requests per minute | 600, sliding one-minute window | Per key |
| Concurrent requests | 20                             | Per key |

Raise them through support, or set a lower `rpm_limit` on a key you hand to a less trusted service.

## Headers

Every response carries the OpenAI header names.

```http
x-ratelimit-limit-requests: 600
x-ratelimit-remaining-requests: 584
x-ratelimit-reset-requests: 12s
```

## When limited

```json title="429"
{ "error": { "message": "Rate limit exceeded. Retry after 12 seconds.", "type": "rate_limit_error", "param": null, "code": "rate_limit_exceeded" } }
```

The response carries `Retry-After` in seconds. Wait that long, then retry. Do not retry faster. The OpenAI SDKs do this on their own.

> **Capacity is not a rate limit**
> A `503` with `model_booting` or `no_capacity` means the model is scaling up, not that your key is throttled. See [Capacity errors](/docs/inference/realtime#capacity-errors).