# Metrics and tokens

> Model usage and the amount of work for every run.

For every task (and every research and audit as well) you can see **metrics**: how much was done, what it cost and how long it took — no “black box”.

## Tiles in the task card

Under the stage strip there are four tiles. Each has a number and a line explaining what it is made of:

- **Changes** — how many files were touched and by how many lines, `+added −deleted`. The tile opens a per-file list.
- **Cost** — charged for completed subtasks, in roubles on your plan.
- **Tokens** — how many the model used, split into input and output.
- **Time** — how long it took from the request to review (or to the merge), and for a run in progress — how long it has been working.

_Illustration: Task metrics — under the stage strip_

## How tokens are counted

The model is billed on two quantities, and both are shown in full:

- **input** — everything the model read: the request itself and **cache hits** (the cache is cheaper, but these are still input tokens);
- **output** — everything the model wrote: the answer and the **reasoning**.

So “input + output” is exactly the total in the “Tokens” tile. The breakdown — how much of the input came from the cache and how much of the output was reasoning — is shown in the tooltip on hover.

Large values are abbreviated: `31.3M`, `609K`, `12K` — so the number reads at a glance.

<Callout type="info">
The same two quantities appear wherever there is a token counter: in the header of an “Ask” conversation, in a research and in an audit section. The same work is counted the same way everywhere.
</Callout>