Kimi Pricing (2026): Free Chat, $19 Plans, Split Ahead

Kimi pricing starts at $0 for Adagio, with Plus at $19, Pro at $39, Max at $99, and Ultra at $199 a month. For the separately billed API, K3.1 Pro is now the cheaper flagship: $2 input and $10 output per million tokens, one-third below K3 at the same token counts. Start free, buy Plus for a first paid month, and choose Pro when Kimi Claw, larger Code context, or K3.1 Pro Highspeed earns the upgrade.

Kimi pricing at a glance

Plus is the safest first purchase; Pro is the practical upgrade for a founder who codes and delegates recurring work. Max earns its price when four concurrent Agent tasks or K3's one-million-token Chat capacity affect paid work. Ultra is a capacity purchase after you exhaust Max.

Verified against the makers' live pages on October 24, 2026: Kimi's USD membership page lists Free (Adagio), Plus, Pro, Max, and Ultra at unchanged prices. Pro and higher now list K3.1 Pro Highspeed access. Matching prices do not establish identical benefits: Kimi's current Code membership guide says legacy subscriptions can continue, while new plans have different quota rules.

The annual monthly equivalents are comparison figures, not monthly installments. These are the USD prices displayed on the purchase page; use the final checkout amount for your account and payment channel.

All capacities above come from the same membership comparison. Widget-task limits match scheduled-task limits on each tier. Scheduled tasks are recurring jobs you can configure, not a monthly allowance of completed jobs. Swarm subtasks are parallel pieces within an Agent Swarm job; they are a different limit from concurrent top-level Agent tasks.

An Agent task is a multi-step job in Kimi's workspace, such as research, slides, or a website. All paid plans list a 4x priority queue. Their plan cards say “More agent credits” for Plus, 2x for Pro, 5x for Max, and 10x for Ultra. Those labels do not promise a fixed number of completed tasks: longer inputs, outputs, and heavier tools consume more of the shared pool.

Is Kimi free?

Yes, Adagio costs $0, but its allowance is limited. It includes one concurrent Agent task, two scheduled tasks, two projects, and 500MB of project storage. It does not include Kimi Code, Agent Swarm, Kimi Claw, or the K3 one-million-token Chat entitlement.

The billing distinction matters more than the model name. Kimi's current credit rules say K2.6 in Chat consumes membership credits, just as K2.6 Agent use in Kimi Work does. Do not budget ordinary chat as an unlimited, credit-free escape from the membership meter.

Stay free when Adagio completes your work without repeatedly blocking useful outcomes. A founder who mainly asks questions and summarizes occasional documents should establish that limit before subscribing. A developer who needs Kimi Code in a terminal, or an operator whose recurring jobs exceed the free allowance, has a specific reason to pay.

Pro is the practical paid Kimi upgrade for regular users

Pro earns its price when Claw, K3.1 Pro Highspeed, or larger flagship context in Code removes a recurring constraint. Agent credits are a shared usage allowance, so the subscription price cannot establish a fixed cost per completed task. Plus remains the safer first purchase; Max and Ultra solve specific capacity problems.

Plus is the one-month validation tier

Plus costs $19 monthly or $180 upfront for a year. It gives you two concurrent Agent tasks, ten scheduled tasks, twenty projects, 20GB of storage, and Swarm with two simultaneous subtasks. Kimi Code is included; Claw is not.

Choose Plus when you need coding assistance or more recurring Agent work than Adagio supports, but have no usage history. A solo founder can have a research job and a document job running together. Skip the upgrade while the free allowance still finishes the work.

Pro is the capability upgrade

Pro costs $39 monthly or $372 annually. The extra $20 a month adds Claw, Goal mode for autonomous work toward completion, fifteen scheduled tasks, and four simultaneous Swarm subtasks. Its card advertises 2x Agent credits.

For a developer, the sharper distinction is flagship access inside Kimi Code: Plus includes K3.1 Pro, while Pro unlocks its context up to one million tokens and K3.1 Pro Highspeed. Pro also retains K3's larger context and K2.7 Code Highspeed. These are Code benefits, separate from the K3 long-Chat entitlement in the membership table. Kimi Code's model configuration says K3.1 Pro Highspeed uses about twice the quota of the regular version. Its faster model output does not accelerate tool calls or script execution.

Pro still allows only two concurrent top-level Agent tasks and keeps the same twenty projects and 20GB storage as Plus. Upgrade for Claw, larger Code context, Highspeed, or measured credit demand; it will not fix a top-level Agent queue bottleneck.

Max is the concurrency and long-Chat tier

Max costs $99 monthly or $948 annually. The extra $60 over Pro buys four concurrent Agent tasks, twenty scheduled tasks, eight simultaneous Swarm subtasks, one hundred projects, and 50GB of storage. Its card advertises 5x Agent credits and adds K3 Chat capacity up to one million tokens.

A context window is how much material a model can consider in a conversation. A large document-heavy research trail can justify the larger Chat window. Short conversations cannot. A founder whose only need is K3's larger context in Code should start by comparing Pro, because that Code entitlement begins below Max.

Ultra is only for measured heavy use

Ultra costs $199 monthly or $1,908 annually. Its 10x Agent-credit label doubles Max's advertised 5x allocation, and scheduled tasks increase from twenty to twenty-five. It keeps Max's four concurrent Agent tasks, eight simultaneous Swarm subtasks, one hundred projects, and 50GB storage.

Buy Ultra after Max's credit pool or scheduled-task allowance repeatedly interrupts valuable work. Paying another $100 a month does not increase those other ceilings. A quiet month still loses unused membership credits at the reset.

Annual billing saves 20.1% to 21.1%, but credits still reset monthly

Annual billing saves $48 on Plus, $96 on Pro, $240 on Max, and $480 on Ultra against twelve monthly payments. Calculated from the current USD prices, those discounts are approximately 21.1%, 20.5%, 20.2%, and 20.1%, respectively.

An annual purchase does not create a year-long reservoir of credits. Both monthly and annual subscriptions refresh their pool monthly, and unused credits expire. Kimi Code remains included in membership, with different usage-limit rules for new and legacy subscribers.

A current legacy subscriber should compare quota rules before switching. A new buyer should observe one monthly cycle before paying upfront. The discount helps only when the plan fits the work throughout the commitment.

Kimi's hidden costs live in the credit rules

Kimi's biggest hidden cost is capacity that disappears or drains while you are not watching. The $19 entry price is only the subscription line. Shared credits, deployed services, Extra Usage, and refund limits determine the operating cost.

Kimi's usage rules put Agent, Deep Research, Slides, Docs, Sheets, Kimi Code, Kimi Work, and Kimi Claw in one pool. Credits refresh on your monthly subscription cycle, including for annual buyers. Unused credits do not roll over or convert to cash.

The same nominal allowance can support very different workloads. A long research job consumes input, output, and tool-related usage differently from a short code question. Use the account's consumption history rather than a fixed “tasks per month” estimate to decide whether the next tier pays.

Idle services keep consuming credits

A deployed Kimi Claw cloud host consumes about 0.6% of membership credits per day, even when you are not actively running a task. At that published estimate, thirty days costs about 18% of the monthly pool before useful work. Save needed files and delete an unused host to stop those charges. Kimi's credit FAQ explains the charge.

The same FAQ lists an approximately 0.08% cloud-service charge for an Agent-published website, but does not specify its billing interval. Do not turn that figure into a monthly estimate. Unpublish dormant sites to stop the charge.

Extra Usage is a prepaid overage wallet

Kimi Extra Usage lets subscribers continue after included credits or subscription usage limits are exhausted. Kimi and Kimi Code share the wallet across supported clients. Time-limited credits are used first; Extra Usage supplies the fallback.

Enable the monthly spending cap before enabling open-ended work. The cap is off by default. Charges depend on the content read and generated and the complexity of the task, so a top-up is not a fixed bundle of completed outcomes.

The current help page quotes its top-up thresholds and examples in RMB. Those are not verified USD checkout amounts. Check the currency and minimum displayed in your own account before paying, rather than relying on a converted or older dollar figure.

Extra Usage balances do not expire and remain spendable after the subscription lapses, but topping up again requires an active subscription. Kimi's Code benefits guide describes top-ups as generally non-refundable, non-transferable, and non-redeemable.

Membership refunds are narrow

Kimi's Paid Services Agreement treats an activated membership as consumed and generally non-refundable and non-transferable except where law requires otherwise. App-store purchases follow the store's refund review. Cancelling auto-renewal stops future renewal while preserving current valid benefits.

Start monthly, turn on the Extra Usage cap, remove idle deployments, and cancel renewal when the plan stops paying for itself. A monthly commitment limits the next decision; it does not make the current payment automatically refundable.

Kimi Code is included; API usage is separate

Kimi Code starts with Plus at $19 a month; the Open Platform API is separately metered. Code is Kimi's developer service for working in a desktop app, terminal, editor, or compatible development tool. Its membership key and the Open Platform API service are independent and cannot be treated as interchangeable prepaid balances. Kimi's product comparison separates them.

Plus and higher can use K3.1 Pro and K3 in Code. Pro and higher unlock their context up to one million tokens, K3.1 Pro Highspeed, and K2.7 Code Highspeed. The standard kimi-for-coding model remains K2.8 Preview, also available from Plus. See the current model configuration for access requirements and the Kimi K3 review for the previous flagship's capability context beyond price.

The quota change is material: new plans have a rolling five-hour usage limit and no weekly limit. Legacy plans retain their weekly limit, which resets every seven days, as well as the five-hour window. Both still consume the monthly membership pool. Requests across Code clients and keys share the allowance. These are usage windows, not a promise of a fixed request count or permission to work for only five hours a week. The benefits guide documents the rules.

For a founder coding interactively, membership can cover both Code and workspace work. For model calls inside a customer-facing product, budget the API separately and monitor its tokens and rate limits.

Current Kimi API rates

K3.1 Pro is the cheaper flagship API to run at equal token counts. Since October 23, 2026, Moonshot has offered K3.1 Pro and K3.1 Pro Highspeed, as recorded in its release notes. Moving from K3 to K3.1 Pro takes uncached input from $3 to $2 and output from $15 to $10 per million tokens: one-third lower on both. Moonshot plans to release the full model weights by November 4, 2026; that is a planned release, not confirmation that the weights are available.

The Moonshot API pricing page lists these USD rates per one million tokens, excluding applicable taxes. Cache-hit input is a reused request prefix; cache-miss input is material the model must process without that reuse.

• Kimi K3.1 Pro (kimi-k3.1-pro): $0.20 cached input, $2.00 uncached input, and $10.00 output; 1,048,576-token context.
• Kimi K3.1 Pro Highspeed (kimi-k3.1-pro-highspeed): $0.20 cached input, $4.00 uncached input, and $20.00 output; 1,048,576-token context.
• Kimi K3 (kimi-k3): $0.30 cached input, $3.00 uncached input, and $15.00 output; 1,048,576-token context.
• Kimi K2.7 Code (kimi-k2.7-code): $0.19 cached input, $0.95 uncached input, and $4.00 output; 262,144-token context.
• Kimi K2.7 Code Highspeed (kimi-k2.7-code-highspeed): $0.38 cached input, $1.90 uncached input, and $8.00 output; 262,144-token context.
• Kimi K2.6 (kimi-k2.6): $0.16 cached input, $0.95 uncached input, and $4.00 output; 262,144-token context.

Highspeed doubles K3.1 Pro's cache-miss and output rates in exchange for speed. The cache-hit rate stays at $0.20. Moonshot describes approximately two to three times faster model output; choose it when waiting is costly enough to justify the premium. K3.1 Pro has lower token rates than K3, but different answer lengths, reasoning usage, or retries can change the cost of a finished task.

The pricing page also announces that K2 models will close to newly registered users on October 31, 2026, while existing users are unaffected by that access restriction. Moonshot plans to retire all K2 models by the end of December 2026. Their lower rates remain useful for existing workloads, but a new integration needs a migration plan.

Cache writes have their own rates per million tokens: K3.1 Pro costs $2 for five minutes or $4 for one hour; Highspeed costs $4 or $8; K3 remains $3 or $6. Five minutes is the default. A cache stores a reusable request prefix; subsequent hits are billed at the cached-input rate and renew its lifetime without another write charge. The cache documentation says cache reads, cache writes, and the uncached remainder are mutually exclusive billing categories: do not charge the same input token twice in your spreadsheet.

Assume one hundred calls, each with 50,000 input tokens and 5,000 output tokens. That is five million input tokens and half a million output tokens:

• K3.1 Pro costs $15.00 with uncached input or default five-minute cache writes, or $6.00 when all input hits an already populated cache.
• K3.1 Pro Highspeed costs $30.00 with uncached input or default five-minute cache writes, or $11.00 when all input hits an already populated cache.
• K3 remains $22.50 with uncached input or default five-minute cache writes, or $9.00 when all input hits an already populated cache.
• K2.7 Code remains $6.75 with uncached input or $2.95 when all input hits cache.
• If all five million input tokens instead incur one-hour cache-write rates, the example costs $25.00 for K3.1 Pro, $50.00 for Highspeed, or $37.50 for K3.

Those are calculated scenarios using the published rates, not measured task costs. The fully cached examples exclude earlier requests that populated the cache. Actual requests mix uncached, cached, and cache-write tokens. Keep stable instructions and reference material at the front of requests, and record the billing breakdown rather than assuming a perfect hit rate.

At that illustrative workload, $19 equals about 127 uncached/default-write or 317 cached K3.1 Pro calls, compared with 84 uncached/default-write or 211 cached K3 calls. This compares spending only. Membership also buys an interface and Agent features, while API spend buys programmatic calls; neither is a substitute for the other's quota.

In the cached K3.1 Pro example, $5 of the $6 total comes from output. Shorter accepted answers and fewer retries can matter more than another improvement in the input cache-hit rate.

API throughput is also limited. The linked recharge and rate-limit page lists a $1-recharge Tier0 with one concurrent request, three requests per minute, 500,000 tokens per minute, and 1.5 million tokens per day. At $10 cumulative recharge, Tier1 lists fifteen concurrent requests, one hundred requests per minute, two million tokens per minute, and no daily token cap. Its notice says limits may be temporarily adjusted when cluster capacity is reached; prepaid balance does not guarantee those ceilings at every moment.

For a builder who prefers paying through Vercel, its Token Pass announcement offers Pro teams early access at $30/month for $80 in shared monthly GLM 5.3 and Kimi K3 credit through AI Gateway, with no rollover and requests stopping at exhaustion without automatic overage charges.

Kimi Business starts at $1,200 a year

Kimi Business's USD purchase page lists $600 per seat per year, about $50 per seat-month, with a two-seat minimum. That makes the starting annual bill $1,200. Five seats still cost $3,000, the example shown in the earlier decision figure.

At five seats, annual personal memberships cost $900 on Plus, $1,860 on Pro, $4,740 on Max, or $9,540 on Ultra. Business is $2,100 above five Plus seats and $1,140 above five Pro seats, but $1,740 below five Max seats. Those price differences do not establish equal quotas.

Stay on personal Plus or Pro subscriptions when independent accounts fit the work. Choose Business for company controls, separate workspaces, dedicated support, and its stated default of not training on business data. It is a governance purchase before it is a discount.

Business includes up to one-million-token conversations, Kimi Work, Swarm, Dashboard, Plugins, Kimi Code, and K3.1 Pro Highspeed access. Moonshot's launch announcement says K3.1 Pro starts disabled for enterprise accounts and administrators can enable it. The Business help page documents member management and separate personal and enterprise workspaces. Web and desktop support Business; mobile currently supports personal membership only. A team working mainly on mobile should resolve that limitation before buying.

Kimi costs slightly less than ChatGPT Plus and Claude Pro

Kimi Plus is the cheapest annual entry among these plans: $180 a year, compared with $200 for Claude Pro and $240 for twelve monthly ChatGPT Plus payments. The maximum gap is $60 a year, too small to rescue the wrong workflow.

ChatGPT Plus costs $20 a month.

Kimi Plus saves $1 in a month or $60 across a year against ChatGPT Plus. In exchange for the annual discount, you commit upfront while credits still reset monthly. Pick the workflow first and the billing term second.

Claude Pro costs $20 monthly or $200 annually.

Kimi saves $20 a year against annual Claude Pro, less than two dollars a month. Choose on whether Kimi's Agent, Code, and long-context workflow replaces enough work to matter. For a research-oriented comparison, see Perplexity pricing. Name the job before comparing subscriptions: chat, coding, research, API integration, or governed team access.

How much is Kimi per month?

Adagio is $0. Paid personal memberships are Plus at $19, Pro at $39, Max at $99, and Ultra at $199 per month. Annual totals are $180, $372, $948, and $1,908, equivalent to $15, $31, $79, and $159 per month when paid upfront.

Is Kimi totally free?

Adagio is a free membership with limited capacity. It includes one concurrent Agent task, two scheduled tasks, two projects, and 500MB of storage. K2.6 Chat consumes membership credits; Kimi Code, Swarm, and Claw require paid access.

What is the price of Kimi Premium?

There is no personal plan named Premium on the current USD membership page; personal paid plans start with Plus at $19 a month. The Business tab does use Premium, priced at $600 per seat annually with a two-seat minimum, or $1,200 a year to start.

How much does Kimi Code cost?

Kimi Code is included from Plus at $19 a month or $180 annually. New plans retain a rolling five-hour usage limit and the monthly shared credit pool, with no weekly limit. The Open Platform API is separately billed, and its keys and credits are not interchangeable with Code membership.

Does Kimi offer a student discount?

The USD membership page does not list a standing student price at this verification. Adagio is free, and the published annual prices are available for comparison; check account-specific offers before assuming a separate student rate.

What is Kimi's refund policy?

Activated paid services are generally non-refundable and non-transferable except where law requires otherwise. Extra Usage top-ups are also generally non-refundable. App-store purchases follow the store's review, and cancelling auto-renewal preserves current valid benefits.

Did Kimi pricing change in 2026?

The current personal paid names are Plus, Pro, Max, and Ultra; USD monthly prices remain $19, $39, $99, and $199. The new K3.1 Pro API costs $2 input and $10 output per million tokens, one-third below K3's $3 and $15. Highspeed costs $4 and $20, while both new variants charge $0.20 for cache-hit input. New Code plans retain the rolling five-hour limit and shared monthly pool, with no weekly limit; legacy subscribers can keep their existing rules.

The Monday move: buy one reversible month

Do not start with an annual Kimi plan. Start with the meter that matches the work, then make the commitment from observed usage.

1. 1

   Start on Free

   Use Adagio while its allowance completes the job. Record when a missing feature or depleted credits block useful work; include Chat in the credit budget.

2. 2

   Buy one paid month

   Choose Plus for a first Agent or Kimi Code month. Move to Pro for Claw, larger flagship context in Code, K3.1 Pro Highspeed, or measured credit demand. Choose Max when top-level Agent concurrency or long K3 Chat is the constraint.

3. 3

   Put automation on a capped API budget

   Use the API for your product's model calls. Compare K3.1 Pro against K3 on the same work before switching; the token rates are one-third lower, but the accepted outcome decides the saving. Set a budget in your application and record input, output, cache writes, retries, and accepted outcomes. Check the account's rate limits before planning parallel work.

4. 4

   Use Business for governance

   Budget at least $1,200 a year for two seats. Buy it for company controls, workspace separation, support, and data handling, and compare actual entitlements before replacing personal subscriptions.

After one cycle, annual billing is easier to judge. If the plan consistently saved more than it cost, the discount is useful. If credits expired unused or an idle deployment consumed the pool, a larger annual plan would magnify the mistake.

Want the current AI tools mapped to the work and budget they fit? Get the AI tools map for business owners, free.

Get the AI tools map for business owners

A plain-English map of which AI tools fit which job, updated as prices and capabilities change. Free to subscribers.