Kaizen · AI credits
How AI credits are counted
Kibo’s AI features use AI credits from one balance: conversations, voice transcription, reading supported documents and automatic readings, such as Mail Sync if you turn it on.
A credit is a unit that measures AI processing. The tables below say how many credits each kind of processing uses. A long or complex task uses more credits than a short one, because the instructions, context and tool steps count too, not only the words you type. Fractions of a credit are counted as well.
Your balance is shown in the app. Credits are not money: they have no cash value and you cannot withdraw them, sell them or give them to someone else.
Text and documents
Credits per 1,000 tokens
| Model | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
| Haiku 4.5 | 0.1 | 0.5 | 0.01 | 0.125 |
| Sonnet 5 | 0.2 | 1 | 0.02 | 0.25 |
| GPT-4o mini | 0.015 | 0.06 | 0.0075 | — |
| OpenAI text-embedding-3-small | 0.002 | — | — | — |
Tokens are the pieces of text a model reads and writes. Cached tokens are context the model already read in a recent step.
Voice
Credits per minute of audio
| Model | Per minute | Minimum counted per note |
|---|---|---|
| Deepgram Nova-3 multilingual | 0.52 | None |
| OpenAI Whisper | 0.6 | None |
| Groq Whisper large v3 turbo | 0.066667 | 10 seconds |
Images
Credits per image
| Model | Per image |
|---|---|
| Gemini 2.5 Flash Image | 3.9 |
A model not listed here is charged at Opus 5’s text rates: 0.5 input, 2.5 output, 0.05 cache read and 0.625 cache write, per 1,000 tokens. Speech from a model not listed is charged at OpenAI Whisper’s rate, and an image from a model not listed, 8 credits.
We announce rate changes before they apply.