> ## Documentation Index
> Fetch the complete documentation index at: https://docs.youniverse.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Tokens and usage
> Input tokens, output tokens, reasoning, context windows and what they cost.
AI models don't read words — they read **tokens**, small chunks of text. Speed, cost and limits are all measured in tokens.
## What is a token?
* About **4 characters** or **¾ of a word** in English.
* 100 tokens ≈ 75 words. A full page ≈ 500–700 tokens.
* Code, numbers and non-English text often use **more** tokens per word.
* Images, audio and PDFs are converted into tokens too.
## The four kinds of tokens
| Type | What it is | Relative cost |
| - | - | - |
| **Input** | Everything the model reads: your message, history, instructions, files, search results | Base price |
| **Cached input** | Input the provider processed recently | 75–95% cheaper |
| **Output** | The words the model writes back | Usually **5×** input |
| **Reasoning** | Hidden thinking before answering | Billed as output |
Output is the expensive part. Asking for "a short answer" or "bullets only" saves more than trimming your question.
## Context window
The **context window** is the most tokens a model can consider at once — history, files and its own answer combined.
| Model | Context window | Max answer |
| - | - | - |
| Claude Haiku 5.5 / Opus 5.5 | 1,000,000 | 128,000 |
| GPT-6 Luna / 6.1 Sol / 6 Astra | 1,050,000 | 128,000 |
| Gemini 3.8 Flash | 1,048,576 | 65,536 |
| Grok 4.7 | 500,000 | 450,000 |
Aura keeps the **last 40 turns** of a conversation in view and trims older ones in blocks, keeping answers focused and caching effective. Very long prompts (over roughly 100k–272k tokens depending on the model) are billed at a higher tier by providers.
## Reasoning effort
* **Low** — quick factual answers.
* **Medium** — Aura's default for most chat.
* **High** — complex analysis, math and code (Frontier High, Thinking).
More effort means better answers on hard problems, but more output tokens and time.
## How this becomes credits
Aura converts real provider token costs into credits. That's why:
* A typical Orchestration message costs about **\~20 credits** (four models answer, then a comparison).
* Long documents, high reasoning and long answers cost more.
* Follow-ups in the same chat are often cheaper thanks to [prompt caching](/learn/prompt-caching).
## Spend fewer tokens
1. One topic per chat; start a new chat when the topic changes.
2. State the format and length you want.
3. Paste only the relevant part of long documents.
4. Use Unlimited for everyday questions, Orchestration for high-stakes ones.
5. Pick Frontier Low or Efficient unless you need High.
The model also reads the whole conversation, attached files and live search results — those are input tokens too.
Yes, providers bill reasoning as output. Aura keeps reasoning at medium by default to balance quality and cost.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.