There are three ways to use Kimi K3.1 Pro: the free tier gives limited use on the web and in Kimi Work; subscriptions start at $15 a month when billed annually; and the API charges per use, at $2 input and $10 output per million tokens. The free tier is right for getting a first feel for it. Heavy users, people who write code and people who run agents should look at subscriptions, while developers get the best deal paying per use through the API.
| Tier | Monthly, billed annually | Monthly | K3.1 Pro access |
|---|---|---|---|
| Free | $0 | $0 | Limited allowance on the web and in Kimi Work |
| Plus | $15 | $19 | Web, Kimi Work, Kimi Code |
| Pro | $31 | $39 | Adds Highspeed and Goal mode |
| Max | $79 | $99 | 5x agent allowance, long chats up to 1M tokens |
| Ultra | $159 | $199 | 10x agent allowance, long chats up to 1M tokens |
How much can you try for free?
The free tier costs nothing and gives you K3.1 Pro on the web and in Kimi Work, but with a limited allowance — heavy use hits that ceiling first. Kimi Code and Highspeed are not in the free tier, so if you want to write code or work faster you need to start at Plus. If you only want to test how good the model is or occasionally ask a few questions about long documents, the free tier is enough, and there is no need to pay right away.
How is the API billed per use?
The API has its own billing, separate from subscriptions, and a subscription does not include API allowance. The standard kimi-k3.1-pro costs $2 input and $10 output per million tokens, with cached input at $0.20. Highspeed runs about 2 to 3 times faster, and input and output cost twice as much — $4 and $20 respectively. The API requires a $1 top-up first to activate. A quick calculation: 100,000 input tokens and 10,000 output tokens cost about $0.30 on the standard model and about $0.60 on Highspeed. The model's context window is 1 million tokens, so long documents or long codebases can go in at once with no surcharge for length. The open weights have been available on Hugging Face since November 4 — if you want to host it yourself, the local requirements are a separate question.
