> ## Documentation Index
> Fetch the complete documentation index at: https://docs.os3compute.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Normal and fast

> The fast id: the same model on a separate, less crowded lane, billed by the lane that served each answer.

`deepseek/deepseek-v4.1-flash-1017-fast` is the same model with the same settings on a separate, less crowded lane of GPUs. It is not a different or faster engine: on idle servers the two ids answer at about the same speed. It helps when the normal lane is busy.

Every plan offered now includes it.

* The fast lane runs at most 2 requests at once on each GPU server, and one organisation at most 2 of them.
* When the fast lane has no room, the normal lane serves the request if it has room. If neither has room the request is refused (429 or 503), as for the normal id.
* Each answer is billed by the lane that served it. Its `x-os3-lane` header says which: `fast`, or `bulk` for the normal lane, billed at the normal id's price.
* A conversation that starts on the normal lane stays there while it is active. After about 45 seconds without a request, its next request is placed anew.
* A prompt longer than about 262,144 tokens is always served by the normal lane, so both ids take the full context.

## Which lane answered

Read the `x-os3-lane` header of each answer:

| `x-os3-lane` | Served by | Billed at |
| - | - | - |
| `fast` | The fast lane | The fast id's price |
| `bulk` | The normal lane | The normal id's price |

See [Pricing](https://docs.os3compute.com/pricing.md) for both prices and [Quickstart](https://docs.os3compute.com/quickstart.md#the-fast-id) for code that reads the header.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.