How Much Does Kimi K3 Cost?

Published on July 17, 2026 | Written by Alec Pow
This article was researched using 13 sources. See our methodology and corrections policy.

Kimi K3 is Moonshot AI’s flagship model for coding, reasoning, visual work, and long-context tasks. U.S. users can pay through an API account, a Kimi membership, a third-party model platform, or their own hosted infrastructure.

The official Kimi K3 API rate is $0.30 per million cached input tokens, $3 per million uncached input tokens, and $15 per million output tokens as of July 2026. Consumer memberships range from a free plan to $199 per month. The correct cost unit depends on whether the buyer needs chat access, Kimi Code, agent credits, or metered API calls.

Kimi K3 spending is measured per month for memberships and per million tokens for API use. Output volume, prompt caching, annual billing, shared credits, taxes, and infrastructure choice can move the total.

Light API use can cost a few dollars per month, while paid Kimi memberships run from $19 to $199 per month.

How Much Does Kimi K3 Cost?

Jump to sections

• Cached API input: $0.30 per million tokens
• Uncached API input: $3 per million tokens
• API output: $15 (at $30 per hour, earning that amount would take about 30 minutes, before taxes) per million tokens
• Monthly memberships: $19 to $199
• Annual memberships: $180 to $1,908

Who this cost makes sense for

Kimi K3 can suit developers who can track prompt and response volume, especially when an application reuses long instructions or documents that qualify for cached-input billing. It can also fit programmers who want Kimi Code in a terminal or editor without operating a model server. Membership buyers should match the plan to agent credits and coding activity, since the paid tiers are separated by usage allowances rather than model access alone.

A hosted API removes the need to acquire GPUs, configure inference software, monitor hardware, and keep enough capacity available for peak traffic. That trade can support a small product team whose workload changes from week to week. A fixed membership may be easier for one person who uses Kimi through the consumer interface, yet it does not replace a production API account.

Makes sense if

• Your application can track input, cached input, and generated output separately.
• You reuse prompts or documents that may qualify for cache pricing.
• You use Kimi Code often enough to need paid membership credits.
• You prefer hosted inference over managing GPU servers.

Doesn’t make sense if

• You only need occasional chat sessions covered by the free plan.
• Your application generates very long responses with little cached context.
• You expect a membership to include unlimited public API traffic.
• You cannot monitor token volume or set an account spending limit.

How Kimi K3 billing works

Kimi K3 does not have one universal subscription. Moonshot AI separates consumer membership, Kimi Code access, and developer API usage. The API is pay-as-you-go. Each request creates input and output tokens, with eligible repeated context billed at the lower cached-input rate. Moonshot’s K3 developer documentation lists a one-million-token context window and identifies K3 as a model for long coding and knowledge-work tasks.

Kimi Code uses membership billing rather than the public API balance. Its official billing comparison states that the Code platform uses monthly or annual subscriptions with rate limits, while the standard Kimi Platform uses prepaid, metered API billing. A developer who buys a $39 (about 1.3 hours of work at $30 per hour) monthly membership should not assume that the payment covers requests sent from a production application.

The consumer service also has a free tier. Paid membership adds agent credits, higher task capacity, Kimi Code credits, database calls, and access to selected products. The right checkout path is tied to the work being purchased. Chat, coding assistance, API inference, and self-hosting create separate bills.

The membership tiers

Kimi lists five consumer membership levels as of July 2026. Adagio is free. Moderato costs $19 (about 38 minutes at $30 per hour) per month, Allegretto costs $39 per month, Allegro costs $99 per month, and Vivace costs $199 per month. The current membership price schedule also lists annual totals of $180, $372, $948, and $1,908.

Plan	Monthly billing	Annual payment	Effective monthly rate
Adagio	$0	$0	$0
Moderato	$19	$180	$15
Allegretto	$39	$372	$31
Allegro	$99	$948	$79
Vivace	$199	$1,908	$159

Annual billing saves $48 on Moderato, $96 on Allegretto, $240 on Allegro, and $480 on Vivace. The Moderato calculation is twelve monthly payments of $19, or $228, minus the $180 annual charge. The result is a $48 yearly difference. Annual billing lowers the effective rate but requires payment for the full year at checkout.

Kimi K3 API rates

The official K3 rate card charges $0.30 per million cached input tokens, $3 per million uncached input tokens, and $15 per million output tokens. These July 2026 rates are also listed by OpenRouter’s K3 listing, which identifies a one-million-token context window and the same standard input and output prices.

Output is the expensive part of a K3 request. One million output tokens cost five times as much as one million uncached input tokens. They cost 50 times as much as one million cached input tokens. A workload that sends large documents but asks for brief classifications may produce a smaller bill than a writing or coding workload that generates long answers, even when both process the same total number of tokens.

Caching can cut the input portion when the same instructions, repository context, policy manual, or document set appears across requests and meets Moonshot’s caching rules. It does not reduce the output charge. Long context also does not mean every request must use the full one-million-token limit. Developers pay for the tokens processed, not merely for having access to the maximum window.

A worked K3 API bill

Consider a monthly application workload with 20 million cached input tokens, 5 million uncached input tokens, and 2 million output tokens. Cached input costs 20 multiplied by $0.30, producing $6. Fresh input costs 5 multiplied by $3, producing $15. Output costs 2 multiplied by $15, producing $30. The itemized API bill is $6 + $15 + $30 = $51 before taxes or payment-provider charges.

• Cached input: 20 million tokens at $0.30 equals $6
• Uncached input: 5 million tokens at $3 equals $15
• Output: 2 million tokens at $15 equals $30
• Monthly API total: $51

Changing the output from 2 million to 5 million tokens raises the output charge from $30 to $75. The revised total becomes $6 + $15 + $75 = $96. That is an $45 increase caused only by added response length. This arithmetic shows why a token budget should separate prompt volume from generated text instead of treating all tokens as equal.

The API meter is separate from a Kimi membership charge. A developer using the cost structure of coding assistants as a comparison should distinguish a fixed seat subscription from usage billed by processed tokens.

Kimi Code credits and agent usage

Paid plans share credits across agent features, Kimi Code, documents, slides, spreadsheets, research tasks, and other eligible tools. Moonshot’s credit usage rules state that credits are consumed according to actual token use. Longer inputs and outputs draw more from the monthly pool. Kimi Chat does not consume those shared credits under the current policy.

Moderato includes a base Kimi Code allocation. Allegretto lists five times that amount, Allegro lists 15 times, and Vivace lists 30 times. Moonshot does not present those multipliers as an unlimited number of coding requests. The company also applies time-window and concurrency limits, and individual tasks can consume different amounts based on context and generated code.

Unused credits expire at the end of each cycle and do not roll into the next month. Annual subscribers still receive credits on a monthly schedule rather than receiving the full year’s allowance at once. Once credits run out, existing work can finish, but new tasks may be blocked until the next refresh or a plan upgrade. The expense of operating AI infrastructure provides a different comparison for teams studying self-hosting instead of recurring hosted credits.

Renewal, cancellation, upgrades

Kimi memberships renew until the subscriber turns off automatic renewal. Moonshot’s cancellation instructions state that benefits remain active through the end of the paid billing period. A website purchase can be managed through Kimi settings. An Apple purchase must be managed through Apple subscriptions, and an Android purchase must be managed through Google Play.

Upgrades take effect after a prorated payment is calculated. Downgrades cannot be applied during the current cycle, according to the plan change policy. A user who wants a cheaper tier must cancel the current subscription, retain the existing benefits until expiration, and purchase the lower tier after that date.

Possible extra charges include sales tax, foreign-transaction fees, currency conversion, mobile-platform checkout differences, and unused annual-plan value after a buyer stops using the service.

Invoices may be issued through Airwallex, Stripe, Apple, or Google, depending on the payment route. The invoice policy says receipts are sent to the email registered with the payment provider. A U.S. card issuer may also add a foreign-transaction charge if the merchant and card terms trigger one.

What three Kimi K3 users may pay

Light chat user. A person who uses basic Kimi chat and a small number of agent tasks may stay on Adagio at $0 per month. The free plan includes six agent credits and one concurrent task under the July 2026 plan grid. Moving to Moderato raises the payment to $19 per month or $180 per year.

Coding subscriber. A programmer who needs a larger Kimi Code allowance chooses Allegretto at $39 per month. Twelve monthly payments total $468. Paying the listed annual total of $372 saves $96. The main driver is the coding and shared-credit allowance, not API token traffic.

API developer. A small application sends 10 million uncached input tokens and generates 1 million output tokens. Input costs 10 multiplied by $3, or $30. Output adds $15. The API total is $45 before tax. If half the input qualifies for the $0.30 cache rate, the input portion becomes $15 for fresh tokens plus $1.50 for cached tokens, lowering the full bill to $31.50.

What we verified

• Checked the three token rates in the official API price card.
• Confirmed the one-million-token context limit through the K3 technical release.
• Cross-referenced the launch and open-weight plan with the Reuters launch report.
• Verified the free and paid tiers through the membership plan overview.
• Checked independent reporting of the $3 input and $15 output rates in the Business Insider report.

Article Highlights

• Kimi K3 API use costs $0.30 per million cached input tokens, $3 per million fresh input tokens, and $15 per million output tokens.
• Kimi memberships run from $0 to $199 per month.
• Annual subscriptions cost $180 to $1,908 for the paid tiers.
• Membership payments do not include unlimited API calls.
• Long outputs can drive a larger bill than long cached prompts.
• Unused membership credits expire at the monthly refresh.

Answers to Common Questions

Can Kimi K3 be used for free?

Yes. Kimi has a free Adagio membership, but its agent credits and task limits are lower than those of paid plans. Developer API traffic is billed separately.

How much does the Kimi K3 API cost?

The July 2026 rates are $0.30 per million cached input tokens, $3 per million uncached input tokens, and $15 per million output tokens.

What is the cheapest paid Kimi plan?

Moderato costs $19 per month. Annual billing costs $180, equal to an effective rate of $15 per month.

Does a Kimi membership include API credits?

The consumer membership and public API use separate billing systems. Membership covers eligible Kimi products and shared credits, while API calls are charged through the Kimi Platform account.

Can a Kimi subscription be canceled?

Yes. Automatic renewal can be canceled through Kimi, Apple, or Google, based on the original payment route. Paid benefits remain active until the current term ends.