# Methodology for usage estimates

> Methodology behind the Delegate usage estimates: what counts as a task, how task complexity is measured in tool calls, and how many tasks each license tier supports per month.

Your monthly allowance is expressed in tasks, not in cost. This page describes what counts as a task, how task complexity is measured, and how many tasks of each complexity one month's allowance supports. For how the allowance itself works, refer to [Usage and limits](usage-and-limits.md).

How much of your allowance a task consumes depends on two things: how much work the task requires, and which model runs it.

## What counts as a task

A task is one request you send to Delegate, together with all the work Delegate performs to complete it.

Task complexity is measured in tool calls. A tool call is a single invocation of a tool by Delegate — reading a file, querying an application, running a script, or calling a connector each count as one tool call. A task that Delegate answers from the conversation and its existing context alone, with no tool calls, is a chat task.

Tasks fall into five complexity types by tool call count. The ranges do not overlap, so every task is exactly one type.

| Task type | Tool calls | Typical work |
| --- | --- | --- |
| **Chat** | 0 | Answering a question from the conversation and its existing context, with no external lookup |
| **Light** | 1–3 | Reading a single file or record, or drafting a response from one source |
| **Standard** | 4–8 | Gathering information from a few sources and producing a result from it |
| **Advanced** | 9–24 | Multi-step work across several applications, with intermediate checks |
| **Deep** | 25+ | Long-running, multi-application work with extensive iteration |

## Monthly task estimates

The following tables show how many tasks of each type one month's allowance supports, by license tier, for the three models the sample was measured on.

These figures come from internal testing. The task types reflect the tool call distribution observed across completed Delegate tasks, and the counts are derived from the median consumption of those tasks over a 30-day sample. They describe typical observed behavior, not a guaranteed entitlement.

The counts in each row are **additive, not alternatives**. A Basic user on Kimi 2.5 can complete 46 Chat tasks *and* 46 Light tasks *and* 28 Standard tasks *and* 23 Advanced tasks *and* 8 Deep tasks within the same month — 151 tasks in total.

### Kimi 2.5

| License tier | Chat | Light | Standard | Advanced | Deep |
| --- | --- | --- | --- | --- | --- |
| **Basic** | 46 | 46 | 28 | 23 | 8 |
| **Plus, App Tester** | 343 | 343 | 214 | 170 | 59 |
| **Pro, App Test Developer** | 914 | 914 | 570 | 455 | 156 |

### Claude Sonnet 4.6

| License tier | Chat | Light | Standard | Advanced | Deep |
| --- | --- | --- | --- | --- | --- |
| **Basic** | 6 | 7 | 4 | 3 | 1 |
| **Plus, App Tester** | 49 | 49 | 30 | 24 | 8 |
| **Pro, App Test Developer** | 130 | 130 | 81 | 65 | 22 |

### Claude Opus 4.6

| License tier | Chat | Light | Standard | Advanced | Deep |
| --- | --- | --- | --- | --- | --- |
| **Basic** | 4 | 4 | 2 | 2 | 1 |
| **Plus, App Tester** | 27 | 27 | 17 | 14 | 5 |
| **Pro, App Test Developer** | 73 | 73 | 45 | 36 | 12 |

:::note
- These are estimates, not fixed limits. Actual capacity varies with the length and complexity of each task and how much context each request carries.
- The tables use Unified Pricing license names. For the equivalent Flex licenses, refer to [Usage pool consumption for Autopilot, Delegate, and Cartographer](https://docs.uipath.com/automation-cloud/automation-cloud/latest/admin-guide/usage-pool-consumption-for-autopilot-delegate-and-cartographer#flex).
:::

## Estimating for other models

The three models above are the ones the sample was measured on. Because the lineup changes over time, the model in use may not be among them.

For a later version of a measured model, such as a newer Claude Sonnet or Claude Opus, that model's own tables remain the closest guide, allowing for some shift in consumption between generations.

For a model from a different family, the model picker next to the chatbox shows a capability and cost indicator for each available model: the measured model whose indicator sits closest has the nearest consumption profile, and its tables give the approximate figures.

Model availability itself is set by your organization. For the current lineup and those controls, refer to [Choosing the LLM model](choosing-the-llm-model.md).