One of Japan's largest directories x find the right AI in as little as a minute

  1. AI BEST SEARCH
  2. Kimi
Kimi

Kimi
: How to use it, features, and the business problems it solves

What is Kimi?

Kimi is an AI chat and agent service from Beijing-based Moonshot AI. Its flagship model Kimi K3.1 Pro (announced October 24, 2026; 1M-token context; weights due by November 4) handles slide creation, Deep Research, coding, and 3D creation. There is a free plan, paid plans start with Plus at $19/month, and the K3.1 Pro API costs $2 input / $10 output per 1M tokens. The web version has a Japanese display.

Business problems it solves

About "Kimi"

What is Kimi?

Kimi is an AI chat and agent service developed by Moonshot AI (Beijing Moonshot Technology Co., Ltd.), based in Beijing, China. It goes beyond Q&A chat: it can build slides, documents, and spreadsheets, run Deep Research investigations, and hand off coding work to the Kimi Code agent — an "agentic" product designed to complete work tasks end to end.

The flagship model is Kimi K3.1 Pro, announced on the official blog on October 24, 2026. It is a post-trained version of Kimi K3 (released in July; a Mixture-of-Experts model with 2.8 trillion total and 104B active parameters) with substantially more post-training compute, and like K3 it supports a context window of roughly 1 million tokens (1,048,576) and image input. Moonshot also released Kimi K3.1 Pro Highspeed, which runs at 2–3× the speed of K3.1 Pro. Both are available from launch day on Kimi.ai (web and mobile apps), Kimi Work, Kimi Code, and the API.

Moonshot continues its open-weights strategy: the official blog says the full K3.1 Pro weights will be released by November 4, 2026 (not yet on Hugging Face as of October 24). The previous K3 is released under the MIT-based "Kimi K3 License," which adds conditions for large-scale commercial use (such as a separate agreement for model-as-a-service businesses exceeding $20M in revenue over 12 months, and "Kimi K3" attribution requirements). The K3.1 Pro license had not been published as of October 24 and is expected with the weights.

There are two main ways to use it: the web/mobile app for individuals and teams (international version at www.kimi.ai; Japanese display at www.kimi.ai/ja) and the developer API platform (platform.kimi.ai; platform.moonshot.cn for China).

Provider
Moonshot AI (Beijing Moonshot Technology Co., Ltd., China / privately held). International service via MOONSHOT AI PTE. LTD. (Singapore)
Flagship model
Kimi K3.1 Pro (announced October 24, 2026; 1M-token context) and the faster K3.1 Pro Highspeed. Weights due by November 4
Free tier
Yes (free "Adagio" plan at $0; usage limits not stated on the pricing page)
Paid plans
International: Plus, Pro, Max, Ultra at $19–$199/month (discounted annual billing) / Kimi Business: $600 per seat per year
API pricing (kimi-k3.1-pro)
$2.00 input / $10.00 output per 1M tokens (USD, tax excluded)
UI languages
Japanese available (web at www.kimi.ai/ja and the iOS app). The help center also has a Japanese version
Data storage
China version: within China. International API: servers in Singapore

How to use

  1. Create an account

    On the international version you can sign up with a Google account or an email verification code (no credit card required). The China version (Chinese UI) uses phone number or WeChat login.

  2. Ask questions in chat

    Type into the main input box ("Ask anything, or task an agent..."). Chat is available on the free plan (Adagio), but the pricing page does not state how much free usage is included. On paid plans, every feature, including chat, draws from a shared credit pool.

  3. Delegate tasks to the agent

    From the same input box, request deliverables such as research (Deep Research), presentations (Slides), documents (Docs), spreadsheets (Sheets), and websites (Websites).

  4. Set up scheduled and parallel runs

    Use Scheduled Tasks for recurring runs, or Agent Swarm to execute multiple subtasks in parallel (paid plans).

  5. Developers: issue an API key

    Issue an API key at platform.kimi.ai and call kimi-k3.1-pro, kimi-k3.1-pro-highspeed, and other models via OpenAI- and Anthropic-compatible endpoints.

Key features

01Chat and research

  • Chat — Available on the free plan. Extra-long chat of up to 1M tokens is a Max/Ultra feature
  • Deep Research — Researches across the web and produces cited reports
  • Multimodal input — K3 and later models accept images in addition to text (K3 also accepts video)

02Create

  • Slides — Generates presentation decks from instructions
  • Docs / Sheets / Websites / Design — Generates and edits documents, spreadsheets, websites, and designs
  • Kimi Code — A coding agent and CLI tool; select K3.1 Pro with the /model command. Delegates repository-level implementation and fixes (Plus and above)
  • 3D Make — Added alongside K3.1 Pro. Preview and edit 3D models in the browser, download them for 3D printing, or export them as videos or Live Photos

03Delegate and automate

  • Agent Swarm — Splits one request into multiple sub-agents running in parallel (subtask counts vary by plan)
  • Scheduled Tasks — Automated recurring runs such as report generation
  • Kimi Work / Kimi Claw — Agent products for business use; dedicated Kimi Claw deployment requires Pro or above
  • Plugins, Dream Memory, Goal Mode — Plugin integrations, cross-conversation memory, goal-driven execution (varies by plan)

What is Kimi K3.1 Pro? (new model, October 2026)

Kimi K3.1 Pro is the first model in the K3.1 series, announced by Moonshot AI on October 24, 2026 in the official blog post "Introducing Kimi K3.1 Pro." The company calls it its "most capable 3T-class open-weight model to date," with improvements in coding, 3D creation, financial research, and agentic work.

Model IDs (API)
kimi-k3.1-pro / kimi-k3.1-pro-highspeed
Context
1,048,576 tokens (no price tiers by context length)
Reasoning
Thinking mode always on. Set reasoning_effort to low / high / max (default max)
Where to use it
Kimi.ai (web, iOS, Android, HarmonyOS), Kimi Work (desktop app 3.1.0 or later), Kimi Code, Kimi API
Highspeed
2–3× the speed of K3.1 Pro. In the app, available on Pro, Max, and Ultra plans
Weights
To be released by November 4, 2026 (not yet published as of October 24)

Benchmarks published on the official blog (self-reported)

BenchmarkWhat it measuresKimi K3.1 ProComparison
Terminal-Bench 4.0Coding agents on terminal tasks65.7%Claude Opus 5.5 66.4% / GPT-6 Astra 58.2%
BenchCAD (with tools)3D reconstruction from multi-view renders via CAD code92.7%Claude Opus 5.5 96.2% / GPT-6 Astra 95.9%
Kimi Code Bench 3.0 Long-horizon (internal)Long-running coding work61.2%Kimi K3 55.95%
Kimi Hallu Bench Hard (internal)Claim-level hallucination rate (lower is better)8.3%Kimi K3 23.4%

All figures were published by Moonshot AI on its official blog; evaluation conditions (reasoning effort set to max, Claude Code harness, etc.) are in the blog's footnotes. Internal benchmark results have not been independently verified under the same conditions.

According to the official blog, K3.1 Pro can:

  • Games and 3D — Combine game engines such as Unity, Unreal Engine, and Godot with web search and 3D generation to build complete experiences. Write and run scripts for Blender and CAD software to build parametric models such as mechanical parts, architectural models, and furniture
  • Finance — Financial modeling, investment analysis, and source-grounded research. Scores 64.4 on OfficeQA Pro v2 (described by the company as state of the art)
  • Long-horizon coding — From quick fixes in real codebases to full-project builds

On offensive cyber capability, Moonshot says it identified and removed cyber-related data from training for the open-weight release.

On Kimi Enterprise, K3.1 Pro is off by default at launch; administrators must enable it for their organization. For the API, "Zero Data Retention" is available to customers.

Pricing

Plan names and prices differ between the international version (USD) and the China version (RMB). Prices below are official listings on the pricing page and help center as of October 24, 2026. From Japan you would use the international USD plans; there is no JPY pricing.

In October 2026, the international individual plans were renamed Plus, Pro, Max, and Ultra (official help "Membership plans overview," showing the pricing screen captured on October 19). Prices are the same as the previous plans.

International (www.kimi.ai pricing page and help center, USD):

PlanMonthlyAnnual (per month)Annual totalHighlights
Adagio (free)$0$0$01 concurrent agent task / 2 scheduled tasks / 2 projects, 500MB
Plus$19$15$180More agent credits / 2 concurrent tasks / Kimi Code / Agent Swarm
Pro$39$31$3722× agent credits / K3.1 Pro Highspeed / Goal Mode / dedicated Kimi Claw
Max$99$79$9485× agent credits / 4 concurrent tasks / extra-long chat up to 1M tokens
Ultra$199$159$1,90810× agent credits / 25 scheduled tasks and 25 widget tasks

All membership features draw from a single credit pool metered by actual token consumption. Credits refresh monthly on your subscription date and do not roll over. Kimi Code has its own rolling 5-hour usage limit. When credits run out, subscribers can also buy pay-as-you-go Extra Usage.

Subscribers on the legacy plans (Moderato, Allegretto, Allegro, Vivace) keep their plan names, benefits, and auto-renewal; switching to a new plan is optional. Legacy plans keep a weekly Kimi Code limit in addition to the 5-hour limit.

The China version (kimi.com help center, RMB) has five tiers: Go ¥49, Plus ¥99, Pro ¥199, Max ¥699, and Ultra ¥1,399 per month (Go does not include Kimi Code).

Kimi Business for organizations costs $600 per seat per year (about $50 per seat per month) on an annual contract, from 2 seats ($1,200/year) up to 150 seats per purchase. Content is not used for model training by default, and it includes admin seat allocation and member management, Kimi Code, and dedicated technical support (official help "Kimi Business").

API (platform.kimi.ai, USD, tax excluded, per 1M tokens):

ModelInputOutputCached inputCache writes (5 min / 1 h)
kimi-k3.1-pro$2.00$10.00$0.20$2.00 / $4.00
kimi-k3.1-pro-highspeed$4.00$20.00$0.20$4.00 / $8.00
kimi-k3$3.00$15.00$0.30$3.00 / $6.00

All K3-series models support a 1,048,576-token context, and unit prices do not change with context length. K3.1 Pro costs two-thirds of K3 for both input and output. The API is pay-as-you-go only, with no subscription plan (it is a separate contract from Kimi membership).

For the K2 series (kimi-k2.7-code at $0.95 input / $4.00 output, kimi-k2.6 at $0.95 / $4.00, etc.), Moonshot says that from October 31, 2026 they will no longer be available to newly registered users, and all K2 models are planned to be retired by the end of December 2026. China uses a separate site (platform.moonshot.cn) with RMB pricing.

Payment is in USD; JPY payment is not supported. If you pay with a Japanese credit card, the billed amount depends on the exchange rate at settlement and your card issuer's fees. We could not find refund terms for mid-term cancellation in the official help (as of October 2026). Annual billing offers larger discounts, but starting with monthly billing is safer.

Free plan and trial

There is a free "Adagio" plan ($0), and you can use chat without registering a credit card. According to the comparison table on the pricing page, it allows 1 concurrent agent task, 2 Kimi Work scheduled tasks and 2 widget tasks, and 2 projects with 500MB of storage.

The pricing page does not state how much you can use for free per day or month (number of uses or credit amount) (as of October 2026). In addition, the August help text stating that "chatting with K2.6 consumes no credits" has changed: the October help says that "K2.6 in Chat also consumes membership credits." Kimi Code, Agent Swarm, and K3.1 Pro Highspeed are paid features (Plus and above; Highspeed requires Pro or above).

Japanese language support

  • UI — The web version now has a Japanese display (www.kimi.ai/ja; menus and the pricing page are in Japanese). The iOS app also lists Japanese among its supported languages on the App Store. When this page was first published in August, the web version had no Japanese option.
  • Input/output — Kimi can answer Japanese questions in Japanese. K3 and K3.1 Pro are multilingual models and can generate Japanese documents and slides, though there is no official quality benchmark for Japanese, and reviewers note it is weaker on technical terminology and honorifics than English or Chinese.
  • Support — The help center has a Japanese version (www.kimi.ai/ja/help). Inquiries go by email (membership@moonshot.ai for membership), and there is no official statement on whether support is provided in Japanese.

Reputation and track record

Figures verifiable from primary sources (GitHub stars and funding as of August 19, 2026; benchmarks from the October 24 announcement):

  • GitHub stars — About 11,100 for MoonshotAI/Kimi-K2 and about 8,500 for MoonshotAI/Kimi-K3. K2 drew major attention at its July 2025 release as a 1-trillion-parameter open-weights LLM.
  • Funding — According to Bloomberg (July 29, 2026), Moonshot AI raised $3.5 billion at a $35 billion valuation. China's state AI industry investment fund was among the lead investors, and the company reportedly aims for a Hong Kong listing.
  • Benchmarks — For K3.1 Pro, the official blog reports 65.7% on Terminal-Bench 4.0 (Claude Opus 5.5: 66.4%) (self-reported; see "What is Kimi K3.1 Pro?" above).

No official list of enterprise customers or user counts is published. Named reviews in Japanese remain limited.

How it differs from other tools

As a China-born open-weights LLM, its closest comparisons are DeepSeek and Qwen. Kimi's distinguishing trait is that beyond releasing models, its app is built around an agent that finishes deliverables — slides, documents, and code.

AspectKimiDeepSeekQwenChatGPT / Claude
ProviderMoonshot AI (China, private)DeepSeek (China)Alibaba (China)OpenAI / Anthropic (US)
Model weightsOpen weights (K3 published; K3.1 Pro due by November 4)Open weightsMany models open sourceClosed
Context~1M tokens1M tokens (V4)Varies by modelUp to ~1M tokens
App focusAgents (docs, research, code)Chat and APIChat, images, videoChat and agents
Japanese UIYes (web and iOS app)YesYesYes
Paid plansFrom $19/monthMostly free chatMostly freeFrom ~$20/month

If you prioritize a Japanese UI or general-purpose chat polish, ChatGPT, Claude, and Gemini come first. For developers weighing API unit costs and self-hosting, comparing Kimi, DeepSeek, and Qwen is the practical exercise.

What to check before adopting (data handling)

Because the provider is a Chinese company, check the following against your own standards before business use.

01Data storage location differs by edition

The China version (kimi.com's Chinese terms) states that personal information is stored within the People's Republic of China. Meanwhile, the API platform (platform.kimi.ai) is provided by the Singapore entity MOONSHOT AI PTE. LTD., which states that collected information is stored on servers in Singapore. Confirm which terms apply to your contract. With the K3.1 Pro announcement, Moonshot also said that "Zero Data Retention" is available to API customers.

02Training use of inputs

The China version's privacy policy states that inputs and outputs are used to optimize models, and that you can opt out by contacting customer service. Kimi Business for organizations states that content is not used for model training by default. We could not confirm explicit opt-out procedures for the international individual plans, so on individual plans, operating on the assumption of not entering confidential or personal information is the safe default.

03The operator is identifiable

Moonshot AI is not an anonymous operator but a prominent AI company headquartered in Beijing. Even so, judge against your internal information-management rules and cross-border data-transfer policies.

FAQ

Q. How much can I do for free?

You can use chat on the free "Adagio" plan. According to the pricing page's comparison table, it allows 1 concurrent agent task and 2 projects with 500MB. The number of free uses or amount is not stated. Kimi Code and Agent Swarm require a paid plan (Plus, from $19/month), and K3.1 Pro Highspeed requires Pro ($39/month) or above.

Q. Does it work in Japanese?

Yes. Chat input and output work in Japanese, and the web version (www.kimi.ai/ja) and the iOS app have a Japanese display. The help center also has a Japanese version.

Q. What is Kimi K3.1 Pro?

It is Kimi's flagship model, announced on October 24, 2026. It is a post-trained version of K3 with a 1M-token context, and is said to improve coding, 3D creation, and financial research. The official blog reports 65.7% on Terminal-Bench 4.0 (Claude Opus 5.5: 66.4%). It is available on Kimi.ai, Kimi Work, Kimi Code, and the API.

Q. How much does the Kimi K3.1 Pro API cost?

kimi-k3.1-pro costs $2.00 input and $10.00 output per 1M tokens, and the faster kimi-k3.1-pro-highspeed costs $4.00 input and $20.00 output (USD, tax excluded). That is cheaper than the previous K3 ($3.00 input / $15.00 output).

Q. When will the K3.1 Pro open weights be released?

The official blog says "by November 4, 2026." As of October 24, the K3.1 Pro weights are not yet on Moonshot AI's Hugging Face page, and the license terms have not been published.

Q. Can I pay in Japanese yen?

No. Prices are in USD (RMB for the China version); there is no JPY display or settlement. Payments with a Japanese card are converted at the exchange rate at the time of settlement.

Q. Are my inputs used for training?

The China version's terms state that inputs/outputs are used for model optimization and provide an opt-out channel. Detailed procedures for the international version could not be confirmed in writing, so we recommend not entering confidential information.

Q. Can I cancel or get a refund?

Subscriptions auto-renew, and you can cancel auto-renewal at any time from "Settings → Subscription" (or from the App Store / Google Play if you subscribed there). After canceling, your benefits stay active until the end of the paid period. We could not find refund terms for mid-term cancellation in the official help (as of October 2026). Consider trying monthly billing before committing to an annual plan.

Q. What is the difference between Kimi K2, K3, and K3.1 Pro?

K2 (July 2025) is a text LLM with 1 trillion total / 32B active parameters, released under a Modified MIT License. K3 (July 2026) has 2.8 trillion total / 104B active parameters and supports image/video input and a 1M-token context. In between came K2.5, K2.6, and K2.7-Code (coding-focused). K3.1 Pro (October 2026) is a post-trained version of K3 with the same 1M-token context. The K2 series on the API is planned to be retired by the end of December.

Provider

Moonshot AI (Beijing Moonshot Technology Co., Ltd.) is an AI company founded in Beijing in March 2023 by Yang Zhilin and others. It is privately held; Bloomberg reported (July 2026) that a $3.5 billion raise brought its valuation to $35 billion, with a Hong Kong listing as the goal. The international service and API are provided by the Singapore entity MOONSHOT AI PTE. LTD.

Its main products are the Kimi AI assistant (web/iOS/Android), the Kimi Code coding agent, Kimi Work for business, Kimi Business for organizations, the developer API platform (platform.kimi.ai / platform.moonshot.cn), and the open-weights Kimi model series published on Hugging Face.

The information on this page is based on content published on the official site as of October 2026. Features and pricing may change; please check the official site for the latest information.

Try Kimi

Frequently compared services

ChatGPT

ChatGPT

ChatGPT is an AI assistant provided by OpenAI, the conversational service at the center of the generative AI boom. The desktop app has moved to a new app that combines Chat, Work, and Codex, and the previous version, "ChatGPT Classic," will be discontinued on January 30, 2027 (chat history carries over). Since October 2026, Chat runs on GPT-6 (GPT-6 Sol on paid plans, GPT-6 Luna on Free and Go), and the new Intelligent UI builds charts, buttons, forms, and small tools into answers. With the Chrome extension, Chat can read your open tabs and Work can operate them to fill out forms, send messages, and download files (all plans). In ChatGPT Work and Codex, you can also choose GPT-6.1 Sol (September 2026) and GPT-6.1 Terra (October 26), which excels at reading images and computer use and costs half as much as Sol on the API (input $1 / output $5 per 1M tokens) (Plus and above). ChatGPT Slides lets you create and co-edit decks in real time (Pro, Business, Enterprise). On November 10, 2026, Business Hub launched in beta, letting companies manage their information and check metrics such as mentions. It also offers Deep Research, image generation, data analysis, voice conversation, and a coding agent (Codex), plus projects, scheduled tasks, custom GPTs, and workspace agents. Pricing is free / Go ¥1,400 / Plus ¥3,000 / Pro ¥16,800– (individual, monthly), with Business and Enterprise for organizations.

# Generative AI (LLM)# AI Chatbot# Summarization & Translation# Text Generation# AI Agent# Japanese Supported

Articles featuring Kimi