> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.sparkian.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sparkian.com/_mcp/server.

# Sparks (credit system)

Every Sparkian account runs on **Sparks**. It is a credit balance that gets deducted with every message you send, instead of billing you by raw token count. This keeps usage predictable, since the cost of a request depends on the model and content you use rather than an opaque per-token rate.

The estimated Spark cost for each model is shown live on that model's detail panel inside Sparkian, so you have a good idea of the price before you send a message.

![Sparks](https://cdn.sparkian.com/docs/sparks-model-details.png)

## How Sparks are estimated

Each model's detail panel shows a starting cost, like **"1+ Sparks / Request"** or **"25+ Sparks / Request"**. The **+** means this is an estimate, not a fixed price. Sparkian calculates your actual Spark cost based on real usage for that request, not a flat rate per model.

**Examples from the model picker:**

* Qwen 3.5 Flash (a Lite, general-purpose model) → **1+ Sparks / Request**
* Claude Opus 5 (an Advanced, reasoning/coding model) → **25+ Sparks / Request**

## Why your actual cost can be higher than the estimate

The final Spark cost scales with how much your request actually processes like your message, the AI's response, any retained [chat history](/memory-and-context), and attachments, all combined. A short, simple message will land close to the model's listed estimate; a request with a long conversation history or large attachments will cost more, since there's more for the model to read and respond to.

## Knowledge Base fee

Chatting with files through a Data Source (Knowledge Base) adds a flat **+5 Sparks** per request, on top of the base model cost.

## Tips to reduce Spark usage

* Start a fresh chat when switching topics. Older messages in [chat history](/memory-and-context) count toward your token total.
* Pick the lightest model tier that can handle the task.
* Upload files you reuse often to a Knowledge Base instead of re-attaching them to every message.
* Combine related questions into a single message rather than sending several small ones.

## Related

* [Supported file types](/supported-file-types)
* [Memory and context](/memory-and-context)
* [Long context notice](/long-context-notice)