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

# Model Savings

Model savings shows which tasks use models that cost more than they need to, and how much you could save each year by switching each task to the most cost-efficient model that still gives comparable quality.

<img src="https://mintcdn.com/nebulyai/tyxr6Gd4zuVUslkl/images/image-57.png?fit=max&auto=format&n=tyxr6Gd4zuVUslkl&q=85&s=587a0bd0ff51ff9aee7de305197a6c54" alt="Image" width="1507" height="781" data-path="images/image-57.png" />

Most organizations send every request to the same powerful model, regardless of the task. A quick email rewrite does not need the same model as a multi-step financial analysis, but it often gets routed and billed the same way.

Nebuly analyzes usage **one interaction at a time** over a **rolling 30-day window**. This window gives you an up-to-date snapshot of how your organization or teams are currently using AI.

As new interactions are added and older ones fall outside the 30-day window, the analysis updates to reflect your latest usage patterns.

Nebuly then uses those interactions to identify where a different model could handle the same work more efficiently.

1. **Reviews each interaction.** Nebuly looks at what the user asked for and what the model returned.
2. **Estimates its complexity.** Each interaction gets a complexity score based on how hard the request was, for example how much reasoning, context, or specialized knowledge it needed.
3. **Forecasts the best-fit model.** Using that score, Nebuly predicts the cheapest model tier that would have produced the same output quality.
4. **Calculates the savings.** Nebuly reprices each interaction at that model's catalog price, compares it with what you actually paid, and adds up the difference by task over a full year.

Because the estimate is built from your real interactions, the recommendation reflects how your people actually use AI, not a generic benchmark.

### Model tiers

Every model is placed in one of four tiers, from cheapest to most expensive:

<table>
  <colgroup>
    <col width="100" />

    <col width="624" />
  </colgroup>

  <thead>
    <tr>
      <th>Tier</th>
      <th>Typical use</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>**Light**</td>
      <td>Simple, short tasks such as rewording, formatting, or quick lookups</td>
    </tr>

    <tr>
      <td>**Medium**</td>
      <td>Everyday work that needs some reasoning or context</td>
    </tr>

    <tr>
      <td>**Frontier**</td>
      <td>Complex analysis, multi-step reasoning, or long context</td>
    </tr>

    <tr>
      <td>**Flagship**</td>
      <td>The most demanding tasks that need the most capable model available</td>
    </tr>
  </tbody>
</table>

A task's **mix** is how its spend is split across these tiers. For example: Light 29%, Medium 12%, Frontier 59%.

### Model savings

Shows your total **potential** **annualized** **savings** and what share of your annualized AI spend that represents.

All figures on this page are annualized: Nebuly takes the interactions in the last 30 days window and projects them to a full year. This gives you one consistent yearly figure to compare with your AI budget, whichever team or task you look at.

Click **View details** to open the full breakdown:

* **Savings opportunity** and **Actual AI spend**, both annualized
* A plain-language summary of where spend goes today and how much of it could move to lighter models
* **Recommended model tier mix**: the share of spend each tier should have
* **Actual vs optimal catalog cost mix**: two bars comparing today's tier split with the recommended one
* **Annualized savings opportunity calculation**: a model-by-model table showing annualized actual spend, annualized optimal spend, and the change for each model

In the **Change** column, a green negative value means you'd spend less on that model each year. A red positive value means you'd spend more on it, because work moves there from a more expensive model.

<img src="https://mintcdn.com/nebulyai/73dNoiWDlX9XFjz2/images/Screenshot-2026-09-23-at-16.00.22.png?fit=max&auto=format&n=73dNoiWDlX9XFjz2&q=85&s=301d31c0d5f03035708b7aff90f35a50" alt="Screenshot 2026 09 23 At 16 00 22" width="1254" height="820" data-path="images/Screenshot-2026-09-23-at-16.00.22.png" />

### Top savings opportunities

Lists the 3 tasks where switching models saves the most. Each shows the recommended mix, the annual savings, and the share of that task's spend you'd save.

The figure at the top is how much of your total AI spend those three tasks alone could save.

Click **All opportunities** to see the full ranked list, with the total savings from the top 10.

### Savings opportunity by task

Lists every task with:

<table>
  <colgroup>
    <col width="124" />

    <col width="692" />
  </colgroup>

  <thead>
    <tr>
      <th>Column</th>
      <th>What it shows</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>**Saving / yr**</td>
      <td>Annual savings from switching to the optimal mix, with the percentage of the task's spend</td>
    </tr>

    <tr>
      <td>**Spend / yr**</td>
      <td>Current annual spend on the task</td>
    </tr>

    <tr>
      <td>**Current mix**</td>
      <td>Today's tier split. The label shows the largest tier. Hover to see the full split</td>
    </tr>

    <tr>
      <td>**Optimal mix**</td>
      <td>The recommended tier split. The label shows the largest tier</td>
    </tr>
  </tbody>
</table>

Click **View details** on any row to open that task's breakdown. It uses the same layout as the Model savings panel, filtered to that task. Use the arrows at the top to move between tasks.

### Filtering the view

Use **All**, **Team**, **Market**, or **Tenure** at the top to see savings for the whole organization or for a specific group.

## Build your own reports

Model optimization metrics are also available in Nebuly's reporting engine.

You can combine **model usage**, **estimated savings**, **task type**, and user attributes with your existing filters and groupings to analyze where more efficient model choices could reduce cost.

For example, you can build reports to understand:

* Which tasks have the largest optimization potential
* Which teams or business units could save the most
* How model usage differs across teams
* Where expensive models are being used for simpler tasks


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