Kimi Code the CLI is free; the model access behind it is sold three ways — a Kimi membership from $19 a month, an Extra Usage wallet that picks up when the membership quota runs out, and Kimi's metered Open Platform API at $3 per million input tokens and $15 per million output for K3. The membership's allowance is unpublished, so the choice between them comes down to volume, and the arithmetic below shows where the line sits. Prices read from kimi.com and platform.kimi.ai on October 1, 2026.
For what Kimi Code is and how it compares, see Kimi Code vs Claude Code; for replacing the client, the membership or the models, see Kimi Code alternatives, which also covers what does not move when you change provider.
The membership tiers
| Tier | Monthly | Billed annually | Kimi Code and K3 | K3 1M context, HighSpeed |
|---|---|---|---|---|
| Plus | $19 | $15 a month ($180 a year) | Yes | No |
| Pro | $39 | $31 a month ($372 a year) | Yes | Yes |
| Max | $99 | $79 a month ($948 a year) | Yes | Yes |
| Ultra | $199 | $159 a month ($1,908 a year) | Yes | Yes |
Prices from kimi.com's membership page, which renders them in the browser rather than in the page source; the two right-hand columns from Kimi's membership documentation. The free Kimi account and the Go tier include no Kimi Code quota. Kimi renamed its tiers recently "with pricing unchanged"; legacy plans keep their old names and rules until you switch.
How the quota behaves
- One pool for everything. Requests from the CLI, VS Code, the desktop app and third-party tools all count toward the same membership quota, and every logged-in device and API key shares it.
- A rolling five-hour window throttles bursts and recovers by itself. New members no longer have a weekly limit; legacy plans still refresh every seven days from the subscription date, with no rollover.
- A monthly total quota, shared with Kimi on the web, freezes Kimi Code when reached, until it resets or you upgrade.
- No number is published for any of the three — not tokens, not requests. Your remaining quota is visible in the Kimi Code Console and with
/usagein the CLI.
Extra Usage: the wallet past the quota
When the membership quota runs out, Extra Usage keeps requests going from a prepaid balance. The rules, from the same documentation:
- Only subscribed members can turn it on, and Kimi on the web and Kimi Code draw on one shared balance.
- It is billed by actual usage in RMB, at rates Kimi describes as close to the Open Platform's API prices — and Kimi itself says that for heavy use, upgrading the subscription is usually more cost-effective.
- It is deducted last, after the subscription quota, so tasks keep running instead of failing. Without a monthly spending cap set in its settings, there is no cap.
- Top-ups start at ¥25, up to 10 a day and ¥3,000 a day, with a ¥10,000 balance ceiling. The balance does not expire, survives a lapsed subscription (though you cannot top up while lapsed), and is generally non-refundable.
The metered API, per token
| Model id | Input (cache miss) | Input (cache hit) | Output | Context |
|---|---|---|---|---|
kimi-k3 — cache write $3.00 (5-minute TTL) or $6.00 (1-hour) | $3.00 | $0.30 | $15.00 | 1,048,576 tokens |
kimi-k2.7-code | $0.95 | $0.19 | $4.00 | 262,144 tokens |
kimi-k2.7-code-highspeed | $1.90 | $0.38 | $8.00 | 262,144 tokens |
Per million tokens, before tax, from platform.kimi.ai; the API base URL is https://api.moonshot.ai/v1 and it takes OpenAI- and Anthropic-format requests. This is a different product from the membership: its own account, its own key, and model ids (kimi-k3, kimi-k2.7-code) that do not match the membership's (kimi-for-coding, k3). A membership does not fund API calls, and an API balance does not add membership quota.
Membership or metered: where the line sits
These are token arithmetic, not measured task costs. One session that sends 200,000 uncached input tokens and receives 12,000 output tokens:
| Route | Session | Sessions that equal Plus ($19) | Billed by |
|---|---|---|---|
kimi-k3, metered | $0.78 | about 24 | Kimi Open Platform |
kimi-k3, 80% of input cached | $0.35 | about 55 | Kimi Open Platform |
kimi-k2.7-code, metered | $0.24 | about 80 | Kimi Open Platform |
| Claude Haiku 4.5 through Kunavo | $0.18 | about 104 | Kunavo |
| Claude Sonnet 5 through Kunavo | $0.36 | about 52 | Kunavo |
Read it as a threshold. If you run fewer K3 sessions of that shape than the third column says, the metered API costs less than Plus; if you run more, Plus is likely cheaper — until its five-hour window or monthly total stops you, which is the part Kimi does not quantify. The cached row ignores the cache-write charge on the first pass, which Kimi bills at $3 or $6 per million depending on the cache lifetime. Cheapest listed rate and cheapest way to finish the task are different questions: a model that needs three attempts can cost more than a dearer one that needs one.
The Kunavo rows are live catalog rates for Claude models, which Kimi Code can reach through its openai provider type; Kunavo does not sell Kimi models. The catalog amount is a billing floor rather than a cap — when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. Kunavo bills per token from a prepaid balance with a $10 minimum top-up and no subscription (billing); the provider block is on Kunavo's Kimi Code setup page. Nobody at Kunavo has run Kimi Code against its endpoint, so read the first request's charge in your usage log.
FAQ
How much does Kimi Code cost?
The Kimi Code CLI is free, MIT-licensed software (MoonshotAI/kimi-code, 2.1.1 on September 24, 2026). Model access is what you pay for, and it comes three ways. A Kimi membership from Plus upward includes Kimi Code: on October 1, 2026 kimi.com listed Plus at $19 a month ($15 a month billed annually, $180), Pro at $39 ($31, $372), Max at $99 ($79, $948) and Ultra at $199 ($159, $1,908). Extra Usage, a separate RMB wallet, carries on when the membership quota runs out. Or skip the membership and pay per token on the Kimi Open Platform, where kimi-k3 lists at $3 per million input tokens and $15 per million output.
Which Kimi plan do I need for Kimi Code and K3?
Plus is the entry point: Kimi's membership documentation says Kimi Code and K3 access are available on Plus and above, and that the K3 1M-token context and HighSpeed access need Pro or above. The Go tier has no coding quota. Higher tiers raise the agent quota on kimi.com — the pricing page shows 2x on Pro, 5x on Max and 10x on Ultra against Plus — but no tier publishes a numeric token or request allowance for Kimi Code.
What happens when my Kimi Code quota runs out?
Two limits apply to new members. A rolling five-hour rate window throttles bursts and recovers on its own; the weekly limit has been removed for new members, though legacy plans still refresh every seven days. And a monthly total quota, shared with Kimi on the web, freezes Kimi Code until it resets or you upgrade. Past that, Extra Usage keeps requests going from a prepaid RMB balance, billed at rates Kimi says are close to its Open Platform API prices. With Extra Usage switched on and no monthly spending cap set, there is no cap.
Is the Kimi membership cheaper than the Kimi API?
For heavy use, probably — Kimi's own documentation says upgrading the subscription is usually more cost-effective than heavy Extra Usage. The line is volume. At kimi-k3's Open Platform price, a session that sends 200,000 uncached input tokens and receives 12,000 output tokens costs about $0.78, so Plus's $19 equals about 24 such sessions a month; at kimi-k2.7-code's price it is about 80. Kimi does not publish how many sessions Plus actually allows, so the honest test is: if your month on the metered API would cost more than the plan, try the plan and read its usage in the Kimi Code Console.
Can Kimi Code use a cheaper or different API?
Yes. Kimi Code's providers documentation describes an openai provider type for any OpenAI-compatible service with base_url overridden, set in ~/.kimi-code/config.toml; the /login selector itself offers only Kimi Code OAuth or a Kimi Platform API key, so other providers are a config-file edit. The membership and the Open Platform use different model ids and endpoints, so a config written for one does not carry to the other. Kunavo does not sell Kimi models; its setup page for Kimi Code points the openai type at Claude and GPT ids instead.
Read on October 1, 2026: kimi.com/membership/pricing (rendered in a browser), Kimi Code's membership documentation, platform.kimi.ai's chat pricing and quickstart pages, and the GitHub API for MoonshotAI/kimi-code. No Kimi account was used and nothing was bought or run; every dollar figure is arithmetic on published rates, and membership allowances are unpublished, so no per-token figure for a plan is given.