Kimi K3.1 Pro: API Prices, Highspeed and Access Guide

Official API prices · API quickstart · Use Kimi · Membership details

Official information checked October 25, 2026. Varwo reviewed documentation and verified the arithmetic below; we did not run paid model requests, install weights or reproduce provider benchmarks.

What is Kimi K3.1 Pro?

Moonshot AI introduces it as a 3T-class model for coding, 3D creation and other agent tasks, with a context window of 1,048,576 tokens. It is available through Kimi products and the API. The launch announcement says full weights will be released by November 4. On the review date, that is a release plan, not a verified public download or permission to assume an unrestricted licence.

The spelling “Kimi K 3.1 Pro” refers to the model covered here. When selecting an API model, use its exact identifier rather than a search phrase.

API pricing: standard, Highspeed and K3

The official API price table lists these USD rates per 1,000,000 tokens, excluding applicable taxes:

Standard K3.1 Pro has lower listed rates than K3. That does not establish a lower cost for every completed task: generated output, retries and tool calls can change the total. Highspeed doubles regular-input, output and cache-write prices compared with standard K3.1 Pro, while its cache-hit rate stays the same.

Work out a request budget

The official billing guide separates input into mutually exclusive categories. A token charged as a cache write should not also be counted as ordinary uncached input.

Suppose a request reports 100,000 input tokens: 80,000 cache hits, 10,000 ordinary uncached tokens and 10,000 new 5-minute cache-write tokens. It also reports 5,000 output tokens.

This is a worked estimate, not an observed bill. It excludes taxes, other service charges and later requests. Use the provider’s reported usage categories; a final answer’s visible length alone is not the output usage. A cache-heavy Highspeed request is not necessarily exactly twice the standard bill.

Membership and API are different purchase routes

Choose a Kimi membership when you want the benefits available inside its products; choose API access when integrating the model into your own application. The provider’s membership and pricing explanation distinguishes these routes and says API usage is billed separately.

Its comparison table lists annual-billing effective monthly prices of $15 for Plus, $31 for Pro, $79 for Max and $159 for Ultra. Those figures are annual-plan comparisons, not a promise of monthly billing at that amount. Plus includes Kimi Code, while Pro and higher include Highspeed in the listed Kimi products. Confirm the actual plan, billing period, region and usage allowance at checkout.

For broader membership and document-workflow context, use the existing Kimi tool guide. This page focuses on the new model and its API cost decisions.

Start with one bounded task

1. Choose your access route. Open the official Kimi product or API platform and confirm that your account can select the intended model. A product membership does not supply an API credit balance.
2. Use the exact model ID. Start with kimi-k3.1-pro unless you have a reason to pay for faster output. Configure any credentials only in your own secure environment, following the official quickstart.
3. Limit the first task. For coding, ask for a focused patch with an observable acceptance check. For a document, provide a short source and require citations to its actual sections.
4. Review reasoning and usage. The quickstart says thinking is always enabled and reasoning_effort accepts low, high and max, with max the default. Compare settings on your own task before choosing a routine budget.
5. Judge the deliverable. Build and test a code change; verify document claims against sources. Save token usage, cache categories and the settings used before scaling up.

These are suggested evaluation steps, not tasks Varwo has executed against the model. A large context window does not guarantee correct retrieval, and provider benchmark scores do not establish performance on your project.

Common questions

Can I download the open weights now?

The reviewed announcement sets a November 4 release deadline. Follow the actual published release and its licence before planning self-hosting; this guide does not verify an available weight download.

Is Highspeed always twice as expensive?

Its regular-input, output and cache-write rates are doubled, but cache hits have the same $0.20-per-million rate. Apply the categories in your actual usage rather than doubling the whole bill.

Does “1M context” mean a million output tokens?

No. Context capacity and the allowed output for a request are different limits. Check the current model documentation and request settings for output constraints.

Sources

• Kimi: K3.1 Pro launch and planned weight release
• Kimi API: official inference prices and context size
• Kimi: membership, cache accounting and pricing explanation
• Kimi API: quickstart and reasoning settings
• Dessa AI Voice: another model-access guide

Prices and access checked October 25, 2026. Recheck the linked provider pages before purchasing or deploying a model.