> ## Documentation Index
> Fetch the complete documentation index at: https://docs.51ultron.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Credits

> How credits work, what they cost, and where to see your usage.

Credits are consumed when Ultron executes tasks. Every token processed — input and output — costs credits. The model used and the task type determine the rate.

## What drives credit consumption

| Factor                    | Impact                                                    |
| ------------------------- | --------------------------------------------------------- |
| Model: Kimi K2            | Higher credit cost — used for complex reasoning tasks     |
| Model: Fast (Haiku-class) | Lower credit cost — used for simple lookups and routing   |
| Sandbox session           | Higher base cost (50-min runtime, 50+ MCP servers loaded) |
| API mode                  | Lower base cost (3-min timeout, fewer servers)            |
| Tool calls                | Each tool call adds input/output tokens                   |
| Canvas generation         | Pre-research (3 searches) + generation = medium cost      |

## Approximate costs by task

| Task                           | Model   | Approx. cost |
| ------------------------------ | ------- | ------------ |
| Quick question or lookup       | Fast    | Low          |
| Lead scoring (per lead)        | Fast    | Very low     |
| LinkedIn post (Pulse)          | Kimi K2 | Medium       |
| Cold email (Striker)           | Kimi K2 | Medium       |
| Competitive analysis           | Kimi K2 | Medium-high  |
| Company deep-dive              | Kimi K2 | Medium-high  |
| Cold email sequence (10 leads) | Kimi K2 | High         |
| Sandbox session (complex task) | Kimi K2 | High         |

## Credits don't expire

Unused credits roll over month to month. You can also purchase additional credits at any time from Settings → Credits without changing your plan.

## Where to see usage

**Settings → Usage** shows:

* Credits consumed per session
* Model breakdown (Kimi K2 vs fast model)
* Top sessions by credit consumption
* Daily and monthly trend

<Tip>
  Check usage after your first few complex sessions (competitive analysis, multi-lead enrichment batches) to calibrate which task types consume the most — then decide whether to run them in sandbox mode or API mode.
</Tip>