TL;DR
- Kimi K3.1 Pro (
kimi-k3.1-pro) on HiAPI charges by token: $2.1548/M input, $11.2714/M output, $0.2156/M cached read. One price tier for all context lengths. - A typical Q&A turn costs about $0.006. $10 covers thousands of short conversations.
- Live on HiAPI. Open the model page and try it in the Playground. New signups get 200 free credits.
How much does Kimi K3.1 Pro cost per million tokens?
| Item | HiAPI price |
|---|---|
| Input | $2.1548 |
| Output | $11.2714 |
| Cached read | $0.2156 |
Prices verified Oct 30, 2026. Charged per actual token consumption. No long-context surcharge tier. Failed requests are not billed.
Input tokens include everything you send: system prompt, conversation history, and the current message. Output tokens are the model's response.
What does a single request cost?
| Use case | Input tokens | Output tokens | Cost |
|---|---|---|---|
| Short Q&A | 200 | 300 | $0.004 |
| Mid-length chat | 2,000 | 1,000 | $0.016 |
| Long doc analysis | 20,000 | 5,000 | $0.10 |
| Code generation | 5,000 | 10,000 | $0.12 |
What can $10 get you?
| Use case | Input tokens | Output tokens | Per-call cost | Calls with $10 |
|---|---|---|---|---|
| Short Q&A | 200 | 300 | $0.004 | 2,500+ |
| Mid-length chat | 2,000 | 1,000 | $0.016 | 625 |
| Long doc analysis | 20,000 | 5,000 | $0.10 | 100 |
| Code generation | 5,000 | 10,000 | $0.12 | 83 |
| Deep reasoning | 50,000 | 30,000 | $0.45 | 22 |
How much does caching save?
Kimi K3.1 Pro supports prompt caching. When you reuse the same system prompt or multi-turn conversation prefix across requests, cached input tokens cost only $0.2156 per million — roughly 10% of the regular input price.
Where caching pays off:
- Fixed system prompts: customer support bots that reuse the same instructions every turn.
- Multi-turn conversations: earlier turns in the history automatically hit cache on subsequent requests.
- Batch document processing: templated prompt structures that repeat across many documents.
No client-side setup needed. The server automatically detects reusable prefixes and applies the cached rate.
How it compares
| Model | Input | Output | Cached read |
|---|---|---|---|
| kimi-k3.1-pro | $2.1548 | $11.2714 | $0.2156 |
| kimi-k3 | $2.32 | $11.60 | $0.232 |
| claude-opus-5-5 | $2.02 | $10.08 | $0.10 |
Kimi K3.1 Pro costs essentially the same as the previous generation kimi-k3 — input is even slightly lower — while adding reasoning, structured output, and full tools support.
Claude Opus 5.5 sits in the same pricing band. All three models cluster between $2.02–$2.32 on input and $10.08–$11.60 on output. At this tier, capability fit and ecosystem compatibility matter more than price differences.
FAQ
How are input and output tokens counted?
Input is everything you send (system prompt + conversation history + current message). Output is the model's response. HiAPI bills based on the actual token count in the usage field.
Are reasoning tokens billed separately?
No. Reasoning tokens are billed at the standard output rate, same as regular completion tokens.
Do I need to configure caching on the client?
No. Repeated system prompts and conversation prefixes are detected and cached automatically on the server side. Cached input is billed at $0.2156/M tokens.
Can I try it for free?
Yes. New signups get 200 credits, enough for dozens to hundreds of mid-length conversations. Head to the model page and start typing in the Playground — no code required.
Does it work with the OpenAI SDK?
Yes. Kimi K3.1 Pro uses the OpenAI Chat Completions protocol. Point your base_url to HiAPI and set model to kimi-k3.1-pro.
Open Kimi K3.1 Pro and see what the reasoning mode can do with your own prompt.
