Minute Verdict
Kimi Code is Moonshot’s coding agent, sold in three surfaces: a desktop app, a terminal tool called kimi, and a VS Code extension. None of that is billed on its own. Kimi Code sits inside the Kimi membership, so the cheapest seat is Plus at $19 a month, and the seat that removes its two walls, the 1M context window and the Highspeed models, is Pro at $39.
Best for: one developer who wants a 3T-class coding model for a flat monthly fee, in a terminal, VS Code or the desktop app.
Not for: a team that buys coding tools per seat with SSO, audit logs and a spend cap per member. The membership is a consumer subscription, and its quota is shared with Kimi chat.
From: $19/mo billed monthly, checked November 13, 2026
What Kimi Code does
The CLI is written in TypeScript and runs on Node.js 22.19.0 or later. The official script, curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash, needs no Node runtime and puts a kimi binary on the path. On Windows the PowerShell equivalent is irm https://code.kimi.com/kimi-code/install.ps1 | iex, with Git for Windows required for its bundled Git Bash shell. The npm route, npm install -g @moonshot-ai/kimi-code, is for machines that run Node. kimi --version confirms the install, and kimi upgrade moves to the latest release.
The first launch asks for a credential. /login opens a selector with two doors: Kimi Code through an OAuth device code, or a Kimi Platform API key. kimi opens the interactive view, kimi -p "..." sends one instruction and prints the answer, and kimi -c resumes the last conversation. Inside a session, /model switches model, /effort changes the reasoning level, /compact compresses the context, /fork copies a session, /usage shows the remaining quota, and Shift-Tab toggles Plan mode. Read-only tools run without a prompt; file writes and shell commands wait for confirmation, and the rules can be made permanent in config.toml or a project’s .kimi-code/local.toml, deny beating ask beating allow.
Two features matter in the first week. /import-sessions turns Claude Code and Codex sessions from the last 15 or 30 days into Kimi Code sessions. And --environment binds a session to an SSH host, a Docker container or a custom command, so file and shell tools run remotely while credentials stay local. The rest is conventional: MCP servers, skills, plugins, sub-agents, hooks, a built-in browser, and Tower, a multi-agent mode.
Six model IDs are published: k3.1-pro, k3.1-pro-highspeed, k3, k3-256k, kimi-for-coding and kimi-for-coding-highspeed. K3.1 Pro is the 3T-class model released on October 23, 2026, with a 1,048,576-token context window and thinking always on: a request that disables thinking is rejected. Highspeed variants trade quota for speed, about 2 to 3 times the output speed at twice the quota, or 6 times for K2.7 Code at three times the quota.
On this page
What you get for the money
| Plan | Monthly | Billed annually | What it opens for Kimi Code | The wall |
|---|---|---|---|---|
| Free | $0 | $0 | Kimi chat only | No coding quota at all |
| Plus | $19 a month | $180 a year, $15 a month | k3.1-pro, k3, k3-256k, kimi-for-coding |
Context capped at 262,144 tokens, no Highspeed |
| Pro | $39 a month | $372 a year, $31 a month | Adds k3.1-pro-highspeed and kimi-for-coding-highspeed, 1,048,576-token context |
Still two concurrent agent tasks |
| Max | $99 a month | $948 a year, $79 a month | 5x agent credits, four concurrent tasks, 100 projects | No model access beyond Pro |
| Ultra | $199 a month | $1,908 a year, $159 a month | 10x agent credits, 25 scheduled tasks | Same models as Pro |
A Plus seat paid monthly for a year costs $228, and the annual view takes it to $180. Pro goes from $468 to $372, and Ultra saves up to $480. Those yearly totals are our arithmetic from the monthly rate, not a published line.
The number that matters is where the walls sit. Kimi Code is available on all four paid plans, and k3.1-pro from Plus. The 1M context window needs Pro, and so does Highspeed. On Plus, the same model answers inside a 262,144-token window, and a large repository meets that limit long before the monthly quota runs out.
Quota is the second trap. Kimi Code, Kimi chat, the VS Code extension, the desktop app and third-party tools all draw on one monthly pool. New members have no weekly window, only a rolling 5-hour rate window. When the monthly total is reached, Kimi Code freezes until the quota resets or the plan goes up. Devices and API keys share the pool, and a device idle for 30 days is unbound until /login runs again.
When the pool is empty, Extra Usage keeps work going. It is a prepaid balance charged near the Open Platform rates, topped up in RMB from ¥25 and capped at ¥10,000, shared with Kimi on the web. Two examples are given, about ¥0.03 for a simple request and ¥1.6 for a multi-step task. No dollar price is published for it, and no separate coding plan page exists.
The Open Platform is the other meter: kimi-k2.7-code at $0.95 per million input and $4.00 per million output, and no monthly floor.
Where it falls short
- There is no standalone coding plan. Kimi Code is one benefit of a consumer membership that also pays for chat, slides and agents, so a buyer who only wants the coding agent subsidizes features they never open.
- Plus caps the context at 262,144 tokens while the model publishes 1,048,576. The gap closes only with Pro at $39, which doubles the entry price.
- The monthly pool is shared between chat and code, and the freeze at the monthly limit stops both.
- K3.1 Pro cannot run without thinking, so unless the effort is lowered every request pays reasoning tokens, and a request that disables thinking returns an error.
- Extra Usage is denominated in RMB with a ¥25 minimum top-up, so a US buyer estimates the overflow in a second currency.
- The Claude Code route is a configuration exercise, not a supported integration. The guide has you patch
~/.claude.json, clear leftoverANTHROPIC_*variables and store an API key in plaintext insettings.json. - Kimi Claw appears in the plan grid with a one-click deployment from Pro and 10 group chats, and the Kimi Code documentation does not describe it beyond that. Treat it as an unmapped feature, not a reason to upgrade.
Kimi Code alternatives
- Grok Build, $8 a month on the X Subscription Plus seat, puts xAI’s coding agent in a subscription bundle at less than half the Kimi entry price. Grok Build review
- Cursor, $20 a month on Pro, is the full editor with its own agent, model picker and cloud runs, for a developer who wants an app rather than a TUI. Cursor review
- Windsurf, $20 a month on Pro, now ships as Devin Desktop with a separate agent and quota system, and puts seats under one admin. Windsurf review
Who it is for: Kimi Code
It suits a solo developer or a very small team that wants a top-tier coding model without a per-seat contract. The install is one line, the login is a device code, and the price does not change with the number of machines. It also suits a developer already paying for Kimi chat: the membership is there, and Kimi Code costs nothing extra from Plus up.
It suits less a team that needs admin controls. There is no seat price, no SSO, no per-member spend cap and no audit trail, and the quota is one pool for everyone. And a developer who wants the 1M context pays $39, where the comparison is no longer with a $20 editor but with a $40 bundle.
The build and data tools hub lists the other terminal agents, and our full Kimi review covers the chat membership this product rides on.
How we checked
We read the Kimi membership page on November 13, 2026, in English and in dollars, in both billing views, and told them apart by the mention printed under the amounts: the monthly view shows “Billed monthly” beside $19, $39, $99 and $199, the annual view shows the struck prices with “Billed annually” and $180, $372, $948 and $1,908. We read the Kimi Code documentation the same day: the CLI getting started guide, the migration page, the model configuration page, the membership benefits page, the FAQ, the three third-party routes and the Open Platform price list. We installed nothing and ran no session, so this is a research review: no timing, no output of our own and no quota measurement. A hands-on test would add the behaviour of the 262,144-token cap on a large repository.
What to do next
Start with a Free account and the install script on a repository you can throw away, to see whether the interface and the approval flow fit how you work. If they do, Plus at $19 is the cheapest seat that includes Kimi Code, and it is the right one while your repository fits in 262,144 tokens. The day the context runs out, or you want the Highspeed models, Pro at $39 opens them, and the annual view takes it to $31 a month. Before paying for a year, check whether the shared pool survives your use of Kimi chat, because the coding allowance is not a separate bucket.
FAQ
Is Kimi Code free?
No. Kimi Code is included from the Plus plan up, and Plus costs $19 a month billed monthly. The Free plan has no coding quota, and no free trial is published for the membership.
How much does Kimi Code cost?
It is not sold on its own. The cheapest seat that includes it is Plus at $19 a month, or $15 a month billed yearly at $180. Pro costs $39 a month, or $31 a month billed yearly at $372, and it is the first plan with the 1M context window and the Highspeed models.
Can I use Kimi Code in Claude Code?
Yes, through an Anthropic-compatible endpoint. The guide sets ANTHROPIC_BASE_URL to https://api.kimi.ai/coding/, puts a Kimi Code API key in ANTHROPIC_API_KEY and points the model variables at k3.1-pro, or k3.1-pro[1m] for the 1M window. The key sits in plaintext in settings.json.
Does Kimi Code work without Node.js?
Yes. The install script needs no Node runtime and places the kimi binary on the path. The npm route needs Node.js 22.19.0 or later. On Windows, Git for Windows is required either way.
Is the Kimi quota shared with Kimi chat?
Yes. The CLI, the VS Code extension, the desktop app, third-party tools and Kimi on the web draw on the same monthly total. When that total is reached, Kimi Code freezes until the quota resets or the plan is upgraded.
Which plan unlocks the 1M context window?
Pro and above. On Plus, k3.1-pro answers inside a 262,144-token window. Pro raises it to 1,048,576 tokens and adds k3.1-pro-highspeed and kimi-for-coding-highspeed.
Sources
Sources
- https://www.kimi.com/membership/pricing (checked November 13, 2026, monthly and annual views)
- https://www.kimi.com/code/docs/en/kimi-code-cli/guides/getting-started.html (checked November 13, 2026)
- https://www.kimi.com/code/docs/en/kimi-code/models.html (checked November 13, 2026)
- https://www.kimi.com/code/docs/en/kimi-code/membership.html (checked November 13, 2026)
- https://www.kimi.com/code/docs/en/third-party-tools/claude-code.html (checked November 13, 2026)
- https://platform.moonshot.ai/docs/pricing (checked November 13, 2026)
Update log
Published November 13, 2026. Plans, prices and quota rules checked November 13, 2026, in both billing views.
