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

# Agent models and credits

> Choose an agent model, read multiplier bands and run estimates, and understand credit reservation and settlement.

Every agent run uses a selected model and consumes credits based on measured model usage. The model picker shows a multiplier band as a quick relative-cost guide; the run estimate and final transaction use measured history and actual usage.

## Model picker

The **Model** picker appears when you start or schedule a built-in agent and when you create a custom agent. It shows only models that are available to your account.

The model catalog assigns these bands:

| Band      | Models                                                               |
| --------- | -------------------------------------------------------------------- |
| **0.25x** | DeepSeek V4 Flash                                                    |
| **0.5x**  | GPT-5.6 Luna, Kimi K2.6, Gemini 3.7 Flash, DeepSeek V4 Pro           |
| **1x**    | Claude Sonnet 4.6, Claude Sonnet 5, GPT-5.6 Terra, Kimi K3, Grok 4.6 |
| **2x**    | Claude Opus 5, GPT-5.6 Sol                                           |
| **4x**    | Claude Fable 5                                                       |

A band compares the model's representative input/output cost with the 1x reference. It is not a fixed charge and does not mean every 2x run costs exactly twice a particular 1x run. Prompt size, output length, tool work, and cached tokens affect the final amount.

## Estimated per run

When the same agent and model have completed measured runs, DevTune shows **Estimated per run** from the 75th percentile of up to 20 recent positive-usage runs. The dialog states how many runs form the estimate's basis.

When there is no measured history, the dialog shows **Credit estimate unavailable**. DevTune reserves one credit to protect the shared balance, then replaces that pending amount with the final measured cost after completion.

Changing the model changes the estimate because history is kept per agent and model.

## Estimated monthly

The schedule dialog also shows **Estimated monthly** when a per-run estimate exists:

* Daily schedules use 30 runs per display month
* Weekly schedules use `30 ÷ 7` runs per display month
* Monthly schedules use one run per display month

This is planning guidance, not a spending cap. Actual scheduled-run usage settles separately for every run.

## Balance checks and final cost

Before a run starts, DevTune reserves the estimated amount. If no estimate exists, it uses the one-credit first-run reservation. The run is blocked when the available account balance cannot cover that reservation.

After the run finishes:

* The final credit cost is calculated from measured model usage
* Any unused reservation returns to the account balance
* Additional usage is debited from the remaining balance when available
* The **Transactions** ledger replaces the pending estimate with the final amount

If measured usage exceeds the reservation and the remaining balance cannot cover
the full overage, DevTune debits only the available balance. The transaction
shows the credits actually charged, while settlement metadata records the unpaid
measured usage as `credit_micros_shortfall`. The balance never becomes negative,
and the shortfall is not carried forward or charged later.

Runs already in progress continue if the account later reaches zero credits. New runs remain blocked until credits become available.

Credits are one account-wide pool shared by agent runs and content workflows. Monthly allowance credits expire at the next billing anniversary; purchased credit packs do not expire. Spending uses expiring allowance credits before purchased credits.

## Next steps

* [Run and schedule agents](/agents/overview)
* [Review credit transactions](/account-billing/credit-transactions)
* [Manage Account & Billing](/account-billing/overview)