Moonshot AI doesn't say how many requests or tokens Kimi Code gives you on any plan. What its documentation does describe is how the allowance works. On the current plans, a rolling five-hour window stops bursts of heavy use, and a monthly total, shared with the rest of your Kimi membership, sets the overall ceiling. The weekly limit is gone for new subscribers but still applies to older plans. Everything below comes from Kimi's pricing pages, the Kimi Code documentation and the Kimi API Platform pricing page, checked as of 10/22, Beijing time.
One naming note: none of Kimi's official pages we checked sell a separately named "Kimi Coding Plan." Kimi Code comes as part of a Kimi membership, and the membership tier decides what you can use.
Plans and prices
Kimi Code is available on the Plus plan and above. The free Adagio tier doesn't include it. Kimi lists four paid plans:
| Plan | Billed annually | Per month on annual billing |
|---|---|---|
| Plus | $180 | $15 |
| Pro | $372 | $31 |
| Max | $948 | $79 |
| Ultra | $1,908 | $159 |
The membership page describes the tiers in terms of agent credits: "More agent credits" for Plus, then "2x," "5x" and "10x agent credits" for Pro, Max and Ultra. Kimi doesn't say whether those multiples apply to Kimi Code usage, so don't read them as Kimi Code quotas. K3 is available from Plus. The 1M-token context for K3 and the faster coding model need Pro or above.
How the limits work
Kimi Code shares one allowance with your Kimi membership. According to the docs, requests from the CLI, the VS Code extension, the desktop app and third-party tools all count toward it. Every signed-in device and every API key draws on the same quota, and devices inactive for more than 30 days are signed out automatically.
The rules depend on when you subscribed. Kimi introduced new plans at the same prices and kept older subscriptions on their original terms:
| Rule | Older plans | New plans |
|---|---|---|
| Minimum tier for Kimi Code | Andante | Plus |
| Five-hour rolling window | Yes | Yes |
| Weekly window | Yes | No |
| Monthly total | Yes | Yes |
| Minimum tier for K3 | Moderato | Plus |
| Minimum tier for K3 with 1M context | Allegretto | Pro |
| Minimum tier for the HighSpeed model | Allegretto | Pro |
On the new plans, the docs say the weekly limit is removed and only the five-hour rolling window remains. Sending too many requests in a short time triggers rate limiting, and access comes back as the window rolls forward. On older plans, the allowance refreshes every seven days counting from your subscription date, and unused allowance doesn't carry over. Both groups share the membership's monthly total. If that total runs out, the docs say Kimi Code usage freezes until the monthly allowance resets or you upgrade. Kimi's Code landing page sums up the change as "no weekly cap in Code, total allowances adjusted."
Kimi publishes no numbers for any of this. The size of the five-hour window, the size of the monthly total and the per-plan request or token counts are not published. The date the new plans took effect isn't given on the pages we checked either. You can see your remaining allowance and rate-limit status in the Kimi Code console, and the CLI's /usage command shows usage too.
What happens when you hit a limit
Kimi's error reference lists the messages you'll see. These return HTTP 403:
- "You've reached your 5-hour usage limit"
- "You've reached your weekly (7-day) usage limit" (older plans only)
- "You've reached your monthly usage limit for this billing cycle"
- "You've reached your concurrent request limit"
Two other errors come back as HTTP 429: "We're receiving too many requests" and "The engine is currently overloaded." If you ask for something your plan doesn't include, the server returns 401 with a message such as "Your current subscription does not have access to k3." Third-party clients such as Claude Code or OpenCode may rewrap these codes, so Kimi suggests matching on the message text.
To keep working without waiting, subscribers can turn on Extra Usage. Once the subscription allowance runs out, requests draw on a prepaid balance instead, and the balance is shared between Kimi on the web and Kimi Code. Kimi says Extra Usage pricing is close to the official Kimi API Platform rates, and that upgrading is usually better value for heavy use. The balance is shown in RMB, never expires and can carry a monthly spending cap. Top-ups start at 25 RMB, with up to 10 top-ups and 3,000 RMB a day and a balance cap of 10,000 RMB. As examples of scale, Kimi's docs put a simple request at about 0.03 RMB and a complex multi-step task at about 1.6 RMB.
Models and how fast they use the allowance
Kimi Code currently offers four model IDs:
| Model ID | Model | Who can use it (new plans) | Notes from the docs |
|---|---|---|---|
| k3 | K3 | Plus and above; 1M context from Pro | Uses about twice as much as k3-256k |
| k3-256k | K3 | Plus and above | Context fixed at 256K; no video input |
| kimi-for-coding | K2.8 Preview | Plus and above | Up to 1M context |
| kimi-for-coding-highspeed | K2.7 Code HighSpeed | Pro and above | Six times the speed at three times the consumption |
Model choice is the biggest lever you have. Kimi's docs say k3 at 1M context consumes roughly twice what k3-256k does, and the HighSpeed model costs three times as much as the standard version. Switching models or changing the thinking effort in the middle of a session invalidates the context cache, so the next requests cost more. Kimi recommends starting a new session when you switch.
According to Kimi's what's-new page, K2.8 Preview has now fully replaced the model behind kimi-for-coding, with no change to the ID. The page doesn't give a date for that switch. A newer model has also shown up in Kimi Code; we covered that in our report on Kimi K3.1 appearing in Kimi Code. K3.1 isn't listed in Kimi's model documentation and hasn't been released.
API keys and pay-as-you-go
Members can create up to five API keys in the Kimi Code console and use them in third-party tools. The Kimi Code API supports both OpenAI- and Anthropic-compatible protocols. These keys draw on your membership allowance, not a separate budget. Kimi also warns that tampering with a client's User-Agent counts as a violation and may suspend membership benefits.
If you'd rather pay per token, the Kimi API Platform is separate from the membership. Its list prices per 1M tokens, as of 10/22:
| Model | Input | Cached input | Output | Context window |
|---|---|---|---|---|
| kimi-k3 | $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 |
For kimi-k3, cache writes are billed separately: $3.00 per 1M tokens for the five-minute tier and $6.00 for the one-hour tier.
Kimi Code vs Claude Code and Codex usage limits
None of the three publishes hard numbers. Anthropic meters Claude Code with a five-hour session limit and a weekly limit, describes plans as multiples, and charges $20 a month for Pro and $100 or $200 for the two Max tiers, as our guide to Claude Code usage limits explains. OpenAI also uses five-hour and weekly windows for Codex, with Plus at $20 and Pro at $100, and doesn't publish weekly caps; see our Codex usage limits guide.
The difference is the weekly meter. On Kimi's new plans, the weekly window is gone and a monthly total takes its place. Kimi Code's entry tier, Plus, is $180 a year on annual billing. All three let you keep going past the limit: Kimi with Extra Usage, Anthropic with usage credits, and OpenAI with credits or an API key.