Understanding Credit Consumption

A message does not have a fixed credit price. The number of credits consumed depends on which model you use and how much content the AI processes and creates for your task.

How Consumption Arises

For each response, the AI processes information and then creates new content:

The AI processes	The AI creates
Your current message	The response to your message
the previous conversation	longer texts, tables, or drafts when requested
relevant content from attachments, knowledge collections, or integrations	possibly several intermediate steps for complex tasks

The more content that is processed or created, the higher the consumption will usually be. Models also differ in how many credits they require for the same amount of content.

You do not need to count words or text volumes yourself. The usage tier in the model selector helps you compare models at a glance. However, it is not a fixed price per message.

Why Credits Instead of Prices per Million Tokens?

Tokens are a technical unit for measuring how much text an AI model processes or creates. However, a price per million tokens provides only limited insight into how many tasks you can complete with them.

Depending on the task, Ayunis Core processes not only your current message, but also information such as the conversation history, attachments, knowledge collections, and the AI’s response. Models also differ in their consumption.

Ayunis Core therefore provides more than a certain quantity of tokens: it is a complete service for using AI in everyday work. Not every processing step is charged separately. Supporting functions such as text recognition in documents (OCR) or preparing documents for knowledge collections are part of the service and are not billed separately based on tokens.

Credits represent the usage-based parts of the service in a common unit. Short, simple tasks generally consume less, while extensive tasks and more powerful models consume more. This keeps consumption tied to actual usage without requiring you to calculate technical token prices yourself.

Tokens therefore contribute to consumption, but they are neither the product itself nor a complete measure of the service provided.

Approximate Guidance

The following examples show the general range in which credit consumption may fall. They are intended as rough guidance and can vary significantly depending on the task.

Model	Approximate consumption per message	Approximately what 1 million credits cover
Mistral Large	200–350 credits	2,900–5,000 messages
Claude Haiku 4.5	600–850 credits	1,200–1,700 messages
Mistral Medium	800–1,300 credits	770–1,250 messages
GPT Luna	470–1,600 credits	630–2,100 messages
GPT Terra	1,100–1,700 credits	590–910 messages
Claude Sonnet 5	3,200–6,800 credits	150–310 messages
GPT Sol	3,300–8,600 credits	120–300 messages
Claude Opus 5	7,900–18,900 credits	50–130 messages

How to Save Credits

Choose a model with a lower usage tier when the task does not require particularly high performance.
Start a new chat for a new topic. This prevents the AI from considering unrelated conversation history.
Specify the desired length, for example, “Answer in five bullet points” or “Write a summary of no more than 200 words.”
Describe the task as completely as possible. A clear request can avoid unnecessary correction rounds.
Attach only relevant documents, especially when files are extensive.

Checking Actual Consumption

Administrators can see how many credits the organization has used during the current month under Admin Settings → Usage. Consumption can also be evaluated by model and user there.

Next Steps

Understand and Select Language Models – compare models and usage tiers
Prompting: Basics & Best Practices – formulate clear requests and avoid unnecessary correction rounds