Mixer AI
Kimi
Kimi K3 and K3.1 Pro chat models: a long context, with text and images as input.
When to choose it
Kimi in Canvas comes as two versions to choose from: Kimi K3 and Kimi K3.1 Pro. Both read text and images and hold a context of about a million tokens, so they suit long material: contracts, correspondence, big drafts. Compare the versions on one small task: the table shows which is cheaper, and the answer shows which is more useful.
- Text and images in one request.
- A context of about a million tokens: you can attach a long document whole.
- Two versions to choose from: Kimi K3 and Kimi K3.1 Pro.
Modes and settings
- Kimi K3 and Kimi K3.1 Pro are separate choices in Canvas.
- Up to 5 text blocks and 10 images per task, producing a text answer.
- For repeated context the rate card has a separate cached input price.
Price in coins
| Version | Input, coins / 1M tokens | Cached input, coins / 1M tokens | Output, coins / 1M tokens |
|---|---|---|---|
| Kimi K3 | — | — | — |
| Kimi K3.1 Pro | — | — | — |
Rates are in coins per million tokens, not per reply. A token is a piece of text, not necessarily a word. Input, cached input and output have separate rates. Canvas reserves an estimate and settles from actual usage; images also affect usage.
≈ marks a fractional rate rounded to two decimal places for display; billing uses full precision.
The dollar price follows the current coin rate.
What you can make
- Go through a long document or correspondence and list contradictions and open questions.
- Review a draft against attached criteria and screenshots.
How to direct the result
- Set the answer format at once: a list, a table, a short summary or edits point by point.
- For long material say what to look for and ask it to mark passages that the text does not support.
Example: checking a draft against a brief
Check the connected draft against the brief. List point by point where the text departs from the brief, and separately any claims the materials do not support. Do not rewrite the text. Answer under 2000 characters: the departures first, then questions for the author.
What to keep in mind
- Canvas saves up to 3400 answer characters: build a long result in parts.
- A confident answer can be wrong: check figures, dates, quotations and outside facts.
Sources
Guides and comparisons using this model
FAQ
How is Kimi K3 different from K3.1 Pro?
They are two versions to choose from. Compare them on the same small task: the price per million tokens is in the table, and answer quality is checked on your own material.
Does Kimi understand images?
Yes. You can connect up to 10 images to a task, but recheck small lettering and numbers on them.
How long can the material be?
The context is about a million tokens, but Canvas limits the number of blocks: up to 5 text blocks and 10 images per task.
What is cached input?
Repeated context that is billed at a reduced rate. The price is in a separate column of the table.
Does the model check facts online?
Built-in search is not promised. Supply reliable material and ask it to mark claims without support.
Why can the final price differ from the estimate?
The amount before the request is approximate; the total depends on the actual size of the input and the answer.
What changes the price?
It depends on the version, the text and images supplied, and the response length. Canvas reserves an estimate first, then bills actual usage. The rates in the table help compare versions; they are not a fixed price for every reply.
Do I need a separate model subscription?
No. In Mixer AI, you top up one coin balance and use it for the tasks you need. You do not need a separate subscription to this model.