Kimi Code Pricing: Every Plan, the Limits, and What Parallel Agents Cost

By CodeAgentSwarm · Updated September 1, 2026

As of August 25, 2026, Kimi's official membership table has four tiers: Andante at ¥49/month, Moderato at ¥99, Allegretto at ¥199 and Allegro at ¥699. Every tier includes Kimi Code. K3 starts at Moderato, while Allegretto and higher unlock up to a one-million-token context in Kimi Code. There is also a pay-per-token route through Kimi Open Platform. Its current public rates are ¥2 cached input, ¥20 uncached input and ¥100 output per million tokens for K3. K2.7 Code costs ¥1.30 cached input, ¥6.50 uncached input and ¥27 output per million tokens. This guide separates those two billing systems and links every changing number to Kimi's official documentation. Prices can move, so use the date above and confirm the checkout total before paying.

Kimi Code plans at a glance

Kimi's official membership page lists these four paid plans as of August 25, 2026:

PlanPublished monthly pricePositioningKimi Code
Andante¥49Everyday useIncluded; K2.7 Code
Moderato¥99ProductivityIncluded; K3 up to 256K
Allegretto¥199ProfessionalIncluded; K3 up to 1M
Allegro¥699All-in-one premiumIncluded; K3 up to 1M

The plan credits are consumed according to actual token use. Kimi Code, Kimi Work, Deep Research, Slides, websites and the other membership features draw from one shared monthly pool, so heavy use elsewhere can reduce what remains for coding.

The official Help Center publishes these prices in Chinese yuan. Kimi can localise checkout pricing by market, so confirm the currency, taxes and annual discount shown in your own account before paying.

How Kimi Code limits work

Two counters matter. Membership credits reset at the start of each monthly billing cycle. Kimi Code also has a separate weekly allowance and 5-hour rate limit. That Kimi Code limit does not consume the limits of other membership features, but the underlying membership credit pool is shared.

Kimi does not publish a fixed token count per agent credit because task consumption varies. A long repository session can cost more credits than a short edit, and several parallel terminals reach the 5-hour limit faster than one.

Type /usage inside the Kimi Code interface to inspect the limits attached to your account. Treat that live account view as the final authority for your remaining capacity.

Which plan you need when several Kimi Code sessions run at once

Kimi does not publish a session count per tier, so nobody can honestly tell you that Moderato equals a fixed number of agents. What Kimi does publish is the shape of the limits: a monthly credit pool shared with the other membership features, a weekly Kimi Code allowance and a 5 hour rate limit. That shape is enough to predict which ceiling you meet first.

Sessions at onceWhat to expectWhat to watch
1 sessionThe 5 hour rate limit rarely gets in the wayThe shared monthly credit pool, since Kimi Work, Deep Research and Slides draw from the same one
2 to 3 sessionsThe 5 hour limit starts to bite before the month ends, since parallel terminals reach it faster than oneRun /usage between tasks, and prefer the k3-256k option for routine work because it uses less quota
4 or more sessionsExpect the weekly Kimi Code allowance to matter as much as the 5 hour oneThe HighSpeed K2.7 Code variant consumes roughly three times the quota, so keep it off bulk work
Kimi publishes no fixed session count per tier. This is the shape of the limits as documented on August 25, 2026; the live /usage view on your account is the final authority.

Because consumption is measured in credits rather than requests, moving up a tier buys you a bigger pool, not a documented number of parallel agents. The practical setup is in run multiple Kimi Code sessions, and the wider workflow in the Kimi Code agent swarm guide.

Current Kimi Code models and context windows

Kimi Code's model documentation now lists K3 as the default model. Moderato includes K3 with up to 256K context; Allegretto and higher unlock up to 1M. The separate k3-256k option uses less quota for routine work.

K2.7 Code remains available to every paid member under kimi-for-coding, with a 262,144-token context. Its HighSpeed variant requires Allegretto or higher and consumes roughly three times the quota. Disabling reasoning routes K3 or K2.7 requests to K2.6.

K3 and K2.7 Code have different context windows, entitlements and token prices. Record the model id used by each agent before comparing cost or quota consumption.

Kimi K3 and K2.7 Code API pricing

If you prefer usage billing, Kimi Open Platform publishes these rates as of August 25, 2026:

  • K3: ¥2 cached input, ¥20 uncached input and ¥100 output per million tokens; up to 1M context
  • K2.7 Code: ¥1.30 cached input, ¥6.50 uncached input and ¥27 output per million tokens; 256K context

Automatic context caching makes repeated repository context cheaper. API billing has no flat monthly ceiling, so set platform budgets before leaving an autonomous agent unattended.

If you specifically want to route the separate K3 model through an Anthropic-compatible client, see Kimi K3 with Claude Code. Its model, limits and prices should not be mixed with the K2.7 Code figures above.

How Kimi Code pricing compares with the other agent CLIs

A price only means something next to the alternatives. Kimi is published in yuan and sits low at the entry level, with an open source CLI you can install before paying anything, and it is one of the few that never publishes a fixed request count. Here is the whole category, one line each, checked on the same day.

AgentFree tierCheapest paidTop individual tierHow you are billed
Claude CodeNo (Free plan has no Claude Code)Pro, $20/monthMax 20x, $200/monthSubscription with 5 hour and weekly windows; API tokens optional
Codex CLILimitedGo, $8/monthPro, $200/monthChatGPT subscription shared with web and IDE; API tokens optional
Kimi CodeCLI is free, model usage is notAndante, ¥49/monthAllegro, ¥699/monthMembership with weekly and 5 hour limits; API per token
OpenCodeCLI is free and open sourceNone, pay as you goNoneYour own provider key, or OpenCode Zen prepaid per token
AntigravityYes, with weekly rate limitsGoogle AI Plus, around $8/monthGoogle AI Ultra 20x, $200/monthGoogle AI plan with rate limits and credits
Grok BuildYes, limited usageSuperGrok, $30/monthSuperGrok Plus, $100/monthxAI plan rate limits; API per token
Cursor AgentHobby, freePro, $20/monthUltra, $200/monthCursor plan with usage pools per model
Entry and top tiers per agent CLI, all verified August 25, 2026. Prices change often; each linked guide carries its own verification date.

Each row has its own guide with the full ladder and the quota mechanics behind it: Claude Code, Codex CLI, OpenCode, Antigravity, Grok Build and Cursor Agent.

Is Kimi Code free?

The Kimi Code CLI is open source under the MIT license and free to install. Model usage is separate. The current official membership table starts with Andante at ¥49 per month, and the API charges per token.

Kimi may offer trials or promotions, but they can change and are not a reliable production budget. Confirm any free allowance inside your account rather than assuming an older Adagio offer still applies.

Watching Kimi usage while agents run

Rate limits become an operational problem when several agents share them: you usually discover the wall when work stops. CodeAgentSwarm shows provider usage alongside parallel terminals so you can rebalance work before a session stalls.

That pairs with the Kimi Code agent swarm guide: use Kimi where it fits and keep another provider for overflow. If Cursor is the alternative, compare its subscription in the Cursor CLI pricing guide.

FAQ

The Kimi Code CLI is free and open source. Model usage is billed separately through a Kimi membership or the Kimi API. As of August 25, 2026, the official membership table starts with Andante at ¥49 per month.

The official Help Center lists Andante at ¥49, Moderato at ¥99, Allegretto at ¥199 and Allegro at ¥699 per month. Check your account for local currency, tax and annual pricing.

All four paid tiers include Kimi Code. K3 starts at Moderato. Kimi Code documentation gives Allegretto and higher access to the up-to-1M K3 context.

Membership credits reset at the start of each monthly billing cycle. Kimi Code also has its own weekly allowance and 5-hour rate limit. Check /usage inside Kimi Code for the limits and remaining capacity on your account.

K3 is the current default in Kimi Code. K2.7 Code remains available as kimi-for-coding with a 262,144-token context. Plan level controls K3 access and its maximum context.

Kimi Open Platform lists K3 at ¥2 cached input, ¥20 input and ¥100 output per million tokens. K2.7 Code is ¥1.30 cached input, ¥6.50 input and ¥27 output.

There is no separate coding plan to buy. Kimi Code is included in all four membership tiers: Andante at ¥49/month, Moderato at ¥99, Allegretto at ¥199 and Allegro at ¥699. K3 starts at Moderato, and Allegretto or higher unlocks the up-to-1M context in Kimi Code.

That is the pay-per-token route through Kimi Open Platform, billed by usage instead of a membership. As of August 25, 2026 K3 costs ¥2 cached input, ¥20 uncached input and ¥100 output per million tokens, and K2.7 Code costs ¥1.30, ¥6.50 and ¥27. It has no flat monthly ceiling, so set a platform budget first.

Any of the four paid tiers includes it. Andante at ¥49 gives you K2.7 Code, Moderato at ¥99 adds K3 up to 256K context, and Allegretto at ¥199 or Allegro at ¥699 unlock K3 up to 1M. Kimi does not publish how many parallel sessions each tier supports, so check /usage on your own account.

Moderato costs ¥99 per month and includes Kimi Code with K3 up to 256K context. Like every tier it is metered by a monthly credit pool shared with the other membership features, plus a weekly Kimi Code allowance and a 5 hour rate limit. Kimi publishes no request count for it, so run /usage inside Kimi Code to see where your account stands.

Run multiple Kimi Code agents with searchable history and real-time change tracking in a single app.

Download free betamacOS & Windows