What your membership credits cover

Slides, Docs, Sheets, Deep Research, Work and other membership features draw from one shared allowance. Consumption depends on the tokens processed, so a longer input, larger output or more complex task can use more of the pool. Kimi’s usage rules also state that K2.6 chat consumes membership credits.

• Refresh: monthly and annual subscribers receive credits on a monthly subscription cycle, tied to their subscription anniversary.
• Expiry: unused membership credits do not roll over to the next cycle.
• Rate limits: Kimi Code’s new plans retain a rolling five-hour rate window and a monthly total; they no longer have a weekly limit. Legacy plans retain the weekly limit too. Remaining monthly credits do not prevent a five-hour limit from stopping requests. Check the limits shown in your account. See Kimi’s Code quota rules or our Code access and limits guide.

The plan documentation presents some allowances as approximate task equivalents. They are not promises of a fixed number of messages, slide decks or research reports. We have not measured a repeatable credit cost for our example files. See Kimi’s pricing footnote and credit explanation before using those figures in a budget.

Hosting can keep using membership credits

A completed task is not always the end of its credit use. Kimi’s credit FAQ describes ongoing consumption for Claw cloud hosts and published Agent websites. Include services you keep online when deciding how much allowance you need.

Kimi Claw: the cloud host uses credits even while idle

The FAQ puts a running Claw cloud host at about 0.6% of membership credits per day, including when it is idle. This percentage describes credit consumption; it is not a percentage of your subscription’s dollar price.

To stop that hosting consumption, Kimi says to delete the cloud host. First save the memory, soul and workspace files you want to keep, as the provider recommends. Deleting a host removes it, so do not treat this as a routine pause button.

Published websites: keeping the site online also uses credits

Kimi says an Agent website continues to consume membership credits for cloud resources while it is published, and Unpublish stops that cloud-service charge. The FAQ does not give a clear billing interval for website credit use, so it does not provide enough information to calculate a reliable daily or monthly hosting budget. Use the account’s usage records to track the actual consumption. Our website publishing and unpublishing guide explains the workflow.

Check what is using your allowance

1. Review your deployed Claw hosts and published websites, including experiments you have finished using.
2. Open Settings → Subscription and inspect the feature-level usage records. Allow for the reporting delay described in Kimi’s FAQ.
3. If a service is no longer needed, use the appropriate stop action above. Turning off Extra Usage controls spending from that wallet; it is not the documented action for stopping hosting consumption.

These are provider-documented hosting rules, checked on 1 October 2026. We did not run a hosting-cost measurement.

Track spending, Extra Usage and renewal

On the web, open your avatar, then Settings → Subscription. Kimi’s account guide explains where to check your current plan and remaining credits. Its credit FAQ describes feature-by-feature usage records and warns that updates can be delayed. Compare usage for your own tasks; do not infer a stable price per file from one run.

Extra Usage is optional prepaid spending

Extra Usage supplements a subscriber’s allowance. When enabled and funded, it can pay for overflow after subscription credits are exhausted or rate-limited. The documented minimum recharge is $10. The monthly spending cap is disabled by default; if you enable it, the documentation says the cap must be greater than $10. You can turn Extra Usage off in web settings to pause spending from that balance.

This wallet is distinct from the developer API balance. Its funds do not expire under the current rules, unlike monthly membership credits. Kimi describes its rates as close to API pricing, not a guarantee that our API examples equal the charge for an app task.

Renewal and cancellation

Check whether your subscription renews automatically and which service bills you. Kimi’s cancellation instructions separate direct subscriptions from Apple and Google Play purchases. Cancelling renewal keeps current-period benefits until expiry; it is different from deleting the account. We did not change a subscription or activate Extra Usage for this review.

K3 cache writes and thinking tokens

For K3, distinguish a cache read from a new cache write. The retention setting, or TTL, changes the write rate. These rates are USD per million tokens:

Chat Completions uses five-minute writes by default when its cache options are omitted. A cache hit does not incur another write charge. Choose the longer TTL only after estimating whether your repeated requests will benefit. See Kimi’s cache pricing and TTL rules.

Count each input token once

For Chat Completions, usage.prompt_tokens includes cache reads, cache writes and the uncached remainder. Read cached_tokens and cache_write_tokens inside prompt_tokens_details; subtract both from the input total to find the remainder. Apply each category’s rate once. Do not add a write charge to an uncached charge for the same tokens.

For Chat Completions, the JSON usage fields do not split cache writes by retention tier. Retain your request setting; Kimi also documents response headers Msh-Usage-Cache-Write-Tokens-5m and Msh-Usage-Cache-Write-Tokens-1h for the tier-specific counts. See the cache usage reference. Other API formats report usage differently; follow the Chat Completions reference when using this formula.

Output includes reasoning

Kimi bills generated thinking as output as well as the final answer. In the examples below, 500 output tokens means 500 total billed output tokens, not a 500-token visible answer plus uncounted thinking. There is no fixed conversion from messages to billed tokens. See thinking-token billing.

What would 1,000 requests cost?

Our illustrative workload is 1,000 requests, each with 2,000 input tokens, no cache hits and 500 total billed output tokens. That totals 2 million input and 0.5 million output tokens. For K3, assume any cache writes use the default five-minute TTL: writes and the uncached remainder both cost $3 per million, so their split does not change this example. These are calculations, not measured performance tests.

If 80% of that same K3 input were cache reads and the remaining 20% were uncached tokens or five-minute writes, the calculation would be (1.6 × $0.30) + (0.4 × $3) + (0.5 × $15) = $9.18. That assumes the write/read categories are counted across the complete workload, with no extra setup requests omitted. It is a hypothetical usage mix, not a promised hit rate.
If the remaining 0.4 million tokens were instead all one-hour writes, the same calculation would use $6 for that category and total $10.38. Do not apply both write rates to the same tokens.

Use the actual billed token counts from your requests. Conversation history, documents and additional model calls can make one user task much larger than one short prompt. These estimates exclude tools, tax and your own application costs. The K3 examples assume the stated cache settings; use your actual write/read breakdown and TTL for a different workload.

Charges and limits beyond the headline rate

API activation and rate limits

Kimi’s API documentation currently requires at least a $1 top-up to begin using the platform. Account rate limits depend on cumulative recharge. Available credit and requests-per-minute capacity are different limits. Check the official recharge and rate-limit rules.

Batch processing

The documented Batch API price is 60% of the standard model price for supported models, including Kimi K2.7 Code and Kimi K2.6. This means a 40% reduction, not a 60% reduction. Use it for eligible work that can wait; do not apply the discount to ordinary live chat calls. Check batch pricing and supported models.

Search and page fetching

The standalone REST tools list $0.002 per successful Web Search Basic call, $0.003 for Web Search Pro and $0.002 for URL Fetch, subject to the documented non-empty-result conditions. These are not a universal price for every built-in or legacy search route. For example, 1,000 billable Basic searches add $2 before any model charges. Verify tool charges.

Kimi pricing questions

Does free Kimi access mean unlimited messages?

The free Adagio listing does not promise unlimited messages. Check the model, features and usage limits shown in your account. We do not have a verified fixed messages-per-day allowance to promise, and free app access does not make API usage free.

Which plan should I start with?

Choose by the task and restriction you need to address, using the plan-selection guide. A low monthly equivalent on an annual plan is useful only if the upfront payment and commitment fit your needs.

Can I buy a plan from ChinaAI?

No. ChinaAI is an independent guide. Purchases, billing and account support belong to Kimi or your selected API provider.

Before estimating usage, compare Kimi models and context windows. Then follow the Kimi API setup and troubleshooting guide to match the key, endpoint and model.

Prices and sources were checked on 1 October 2026, including both international membership billing views, the purchase-page feature comparison and the API price tables. The hosting guidance comes from Kimi’s help documentation. Linked practical studies retain their original September dates and account conditions. Worked API costs are arithmetic examples, not paid usage measurements. See the research methodology.

Continue with the Kimi overview, compare Qwen pricing or DeepSeek pricing, and read how we check sources. Report a pricing correction.