Minute Verdict
Kimi is the chat and agent product of Moonshot AI, built on the K3.1 Pro model, and it is sold as a membership rather than a seat. Plus at $19 a month buys two agents running at once, ten scheduled tasks, 20 projects and 20GB of storage, which covers document work and recurring jobs without a per-token meter. The 1M context window is not in that price: it appears on Max at $99.
Best for: someone who works through long reports and wants agents that run on a schedule, at a flat monthly price with no token arithmetic.
Not for: a buyer who needs the 1M context window on the cheapest plan, or a developer who wants a coding subscription, since Kimi Code is sold separately.
From: $19/mo billed monthly, checked November 5, 2026
What Kimi does
Kimi is an assistant from Moonshot AI, a company based at 76 Zhichun Road in Beijing’s Haidian District, and it runs on models Moonshot trains itself. The current flagship is Kimi K3.1 Pro, published on October 24, 2026: a 3-trillion-parameter-class model with a 1-million-token context window that Moonshot ships as open weights. It answers questions, reads long documents, produces slides and spreadsheets, builds small websites, and runs agents that act on their own.
The agent layer is what separates Kimi from a plain chatbot. Scheduled tasks run on their own and push results back. Swarm splits one job across several agents. Goal mode keeps an agent working until it decides the task is finished. Kimi Claw is the always-on client, on web, Android and PC, and plugins connect professional financial and economic databases. On the coding side, Moonshot measured K3.1 Pro at 65.7 percent on Terminal-Bench 4.0, against 58.2 percent for GPT-6 Astra and 66.4 percent for Claude Opus 5.5, and at 92.7 percent on BenchCAD with tools.
Kimi runs in the browser at kimi.ai, in apps for iOS, Android and HarmonyOS, in Kimi Work on Windows and Apple silicon Macs, and as a Chrome extension. The API sits on a separate platform with its own per-token pricing. What the membership is not: a way to run the model on your own hardware. The open weights are published, but that path goes through the API or your own servers, not through the $19 plan.
On this page
What you get for the money
| Plan | Price | What is included | The limit that matters |
|---|---|---|---|
| Free | $0 | One agent task, two scheduled tasks, two projects, 500MB storage, 15+ plugin types | One agent at a time stops any parallel work |
| Plus | $19/month, $180/year | More agent credits, Deep Research, slides and sheets, website deployment, Swarm | Two agent tasks and two Swarm subtasks |
| Pro | $39/month, $372/year | 2x agent credits, K3.1 Pro Highspeed, Goal mode, Kimi Claw deployment | Still two agent tasks, so the extra credits do not buy parallelism |
| Max | $99/month, $948/year | 5x credits, 1M token chat capacity, four agent tasks, 25 scheduled tasks, 100 projects, 50GB | The first plan where the long context and real parallelism meet |
| Ultra | $199/month, $1,908/year | 10x credits, same caps as Max | Four agent tasks, so the top plan is a bigger credit pool, not more parallel work |
Annual billing takes the same seats down to $15, $31, $79 and $159 a month. On Plus that is $180 against twelve monthly payments of $19, which is $228, so paying yearly saves about 21 percent. Moonshot puts the largest annual saving at $480, on Ultra.
The shape of the ladder is worth reading before choosing. Between Plus and Pro the credits double and Highspeed and Goal mode unlock, but concurrency stays at two agents. The jump that changes the product is Max: four agents at once, 25 scheduled tasks, 100 projects and the 1M token chat capacity. For $19 you get a working agent subscription with real scheduling. For $99 you get the version where long documents stop being a constraint.
Where it falls short
The 1M context window, the number most articles lead with, is not in the $19 plan. On the membership page the 1M chat capacity line appears on Max and Ultra only, and the API reaches 1,048,576 tokens on the K3 models at $2.00 per million input tokens. A buyer who subscribes to Plus expecting to paste a 900,000-token codebase is buying the wrong tier.
Concurrency is capped low on purpose. Plus and Pro both run two agent tasks at once, with two Swarm subtasks on Plus and four on Pro. Credits scale by two, five and ten, the number of agents does not, so a heavy parallel workload queues even on a plan with a large credit balance.
Kimi Code is a separate purchase. The membership lists “Kimi Code available” on every paid plan, while the API documentation describes Kimi Code as a product distinct from the membership and billed on its own terms. Anyone reading the feature list as a coding plan included in the $19 will be disappointed at checkout.
The K2 generation is closing. Moonshot stopped offering K2 series models to newly registered API users on October 31, 2026 and plans to retire all of them by the end of December 2026. Anything built against kimi-k2.6 or kimi-k2.7-code has a migration to K3 ahead of it, and the two K2 models still on the price list carry the same $0.95 input rate.
Data location is the question a US buyer asks first, and the answer is thin. The privacy policy says information may be transferred to and stored on servers outside your country of residence, with no region named. Zero Data Retention is offered to API customers, not to membership subscribers, and Kimi Business is the route for an organization that needs account separation.
Kimi alternatives
The three names below are what a Kimi buyer weighs instead. The yardstick is the $19 Plus seat, because that is where the agent features start.
- DeepSeek, the closest competitor and also a Chinese lab with open weights: chat is free, and the API starts at $0.15 per million input tokens on flash at off-peak rates, against $2.00 on kimi-k3.1-pro: our DeepSeek review.
- Perplexity, for buyers who want sourced answers more than agents: Pro is $17 a month when billed yearly, and the research flow is the product rather than an add-on: our Perplexity review.
- Google AI Studio, for teams already on Google: Gemini 4 Argon bills $2.00 per million input tokens and $10.00 per million output tokens, the same list price as kimi-k3.1-pro, with no free tier on that model: our Google AI Studio review.
None of the three sells a $19 plan that bundles scheduling, slides and Swarm at the same time, which is the specific combination Kimi is pricing here.
Who it is for: Kimi
Kimi suits a solo operator who works with long documents and wants jobs to run without being asked: a weekly research digest on a schedule, a folder of PDFs read and summarized, a slide deck built from a brief, a small site deployed from a prompt. Plus at $19 covers those cases, and the ten scheduled tasks are the part that earns the money back, since the alternative is remembering to run the same prompt every Monday.
It also suits a small team that wants to test agent workflows before committing to a platform. The free tier runs one agent and two scheduled tasks, which is enough to judge whether the output holds, and the membership removes the per-token anxiety that makes people stop using an API.
It does not suit a developer shopping for a coding subscription, because Kimi Code is sold separately and the membership only grants access to it. It does not suit a buyer whose work depends on a million-token context either, since that sits at $99. For plain question and answer at the lowest price, DeepSeek is free and the API undercuts Kimi by a wide margin.
How we checked
This is a research review. We have not used Kimi, so there is no hands-on score, no timing and no sample project. Every price here was read on November 5, 2026, on the membership page in both billing views, on the API pricing documentation, and in the K3.1 Pro release notes, which supplied the model figures and the benchmark scores. We could not verify how the agent credits translate into tasks, because no credit-to-task ratio is published, and we did not subscribe, so the quality of the scheduling is untested. Our how we test page covers what a hands-on pass will add, and our editorial policy covers how a page like this is funded.
What to do next
Start on Free and give it one real document plus one scheduled task. If the output holds after a week, pay the $19 for Plus and keep the schedule running, which is the feature that justifies the seat. Move to Max only when you need the 1M token window or four agents at once, and take the yearly rate on Plus only after a month, since the $180 commits you to twelve months for a saving of about 21 percent. If your work is coding rather than documents, price Kimi Code before the membership, and compare the AI chat tools we have tested before picking a stack.
FAQ
How much does Kimi cost? Free is $0, Plus $19 a month, Pro $39, Max $99 and Ultra $199, billed monthly. Paying yearly takes Plus to $180, Pro to $372, Max to $948 and Ultra to $1,908.
Is there a free tier? Yes. Free gives one agent task at a time, two scheduled tasks, two projects, 500MB of project storage and access to 15 or more plugin types.
Does Kimi have a 1M token context window? The API models in the K3 series support 1,048,576 tokens. On the membership, the 1M chat capacity is listed on Max and Ultra only.
Does the membership include Kimi Code? No. The API documentation lists Kimi Code as a product separate from the membership, billed on its own terms.
Can I use Kimi through an API instead? Yes. kimi-k3.1-pro costs $2.00 per million input tokens and $10.00 per million output tokens, with cached input at $0.20.
Sources
Sources
- https://www.kimi.com/membership/pricing (checked November 5, 2026)
- https://platform.moonshot.ai/docs/pricing (checked November 5, 2026)
- https://www.kimi.ai/blog/kimi-k3-1-pro (checked November 5, 2026)
- https://www.kimi.ai/user/agreement/userPrivacy?version=v2 (checked November 5, 2026)
Update log
November 5, 2026: first publication, research review. Prices read on the Kimi membership page in both billing views and on the API pricing documentation.
