Kimi is an AI assistant from Moonshot AI. Its newest model, Kimi K3.1 Pro, launched on 23 October 2026 and its weights are public on Hugging Face under a custom licence. Kimi has a free plan, paid plans start at $19 a month, and the API costs $2 per million input tokens. Independent scores place it below Claude Opus 5.5.
Kimi is the product name for AI assistants and models from Moonshot AI. This page reads Kimi's own pricing page, API documentation, model repository and privacy policies on 11 November 2026, and checks its benchmark claims against independent leaderboards. Where a number comes from Moonshot itself it is labelled as a vendor claim.
For a second opinion, Masai's blog has Kimi AI Review 2026: Features, Pricing & Comparison. For the wider field, see best AI tools in 2026, ranked.
What Kimi is and what the products are
Moonshot AI makes the models and the apps. Kimi runs on the web and in mobile apps, and the K3.1 Pro launch post lists these ways to use it: kimi.com and the Kimi apps, Kimi Work (a desktop app for Windows and Apple silicon Macs), Kimi Code (a coding agent you point at a model with the /model command) and the Kimi API. Kimi's pricing page also sells Kimi Claw, a cloud version of the open-source OpenClaw assistant, on its Pro plan and above.
Two companies appear in Kimi's policies. The consumer privacy policy is issued by Beijing Moonshot Technology Co., Ltd. The API platform's policy says its services are provided by Moonshot AI Pte. Ltd. in Singapore. The privacy section below explains what that means for your data.
Kimi models and release dates, as of 11 November 2026
| Model | Launched | Context | Status |
|---|---|---|---|
| Kimi K3.1 Pro, and K3.1 Pro Highspeed (2 to 3 times faster) | 23 Oct 2026 (some outlets say 24 Oct) | 1 million tokens | Current flagship. Open weights. API names kimi-k3.1-pro and kimi-k3.1-pro-highspeed |
| Kimi K3 | 16 Jul 2026 (some outlets say 17 Jul; Moonshot's post has no date) | 1 million tokens | Previous flagship, still on the API. Moonshot calls it the first open model at 2.8 trillion parameters |
| Kimi K2.7 Code | June 2026 (API changelog) | 262,144 tokens | Closed to new API users since 31 Oct 2026, retirement planned by end of December 2026 |
| Kimi K2.6 | April 2026 (API changelog) | 262,144 tokens | Same as K2.7 Code |
Kimi's changelog says kimi-k2.5 and the older moonshot-v1 models were retired in August 2026, so old tutorials that call them now return a model-not-found error. Moonshot's own blog posts for K3 and K3.1 Pro carry no date, so launch dates above come from the API changelog and press reports.
Is Kimi free? Plans and pricing
Yes, there is a free plan. Kimi's pricing page lists the tiers below in US dollars. We opened it from India on 11 November and it showed no rupee prices. It also carried a notice that K3.1 Pro requests may queue or fail at launch.
| Plan | Monthly | Billed annually | What the comparison table shows |
|---|---|---|---|
| Free | $0 | $0 | 1 agent task at a time, 2 scheduled tasks, 2 projects with 500 MB. No Kimi Code, Swarm or Goal mode |
| Plus | $19 | $15 a month ($180 a year) | 2 agent tasks at a time, Kimi Code, faster agent queue, 20 projects with 20 GB |
| Pro | $39 | $31 a month ($372 a year) | 2x agent credits, K3.1 Pro Highspeed, Goal mode, Kimi Claw |
| Max | $99 | $79 a month ($948 a year) | 5x agent credits, 4 agent tasks at a time, chats up to 1 million tokens |
| Ultra | $199 | $159 a month ($1,908 a year) | 10x agent credits |
Credits are shared across features, so heavy use of one feature uses up the others. The page does not say which model the Free plan uses or how many chats you get, so we cannot state those limits. A separate help page lists China prices in yuan, which do not apply to the USD plans above.
API pricing
The API is pay as you go with no subscription. Prices per million tokens on Kimi's pricing documentation:
| Model | Input | Output | Cached input |
|---|---|---|---|
| kimi-k3.1-pro | $2.00 | $10.00 | $0.20 |
| kimi-k3.1-pro-highspeed | $4.00 | $20.00 | $0.20 |
| kimi-k3 | $3.00 | $15.00 | $0.30 |
| kimi-k2.7-code | $0.95 | $4.00 | $0.19 |
Batch jobs on K3.1 Pro and K3 are billed at 60% of these prices. As arithmetic on list prices, a session with 1 million input tokens and 200,000 output tokens on K3.1 Pro costs $2 plus $2, so $4. The same session on DeepSeek's deepseek-flash costs about $0.27 at its off-peak rates of $0.15 and $0.60. Real bills depend on caching and reasoning tokens. See the free LLM APIs list for free options.
Open weights: what you can download and the licence
The full K3.1 Pro weights are on Hugging Face and are not gated. It is a mixture-of-experts model with 2.8 trillion parameters in total and 104 billion active per token, and the files add up to about 1.56 TB across 112 files. TechNode reported the release on 5 November 2026. A student laptop cannot run it, so you would use Kimi's API or a hosted provider. For small local models, see the guide to running a free local coding assistant.
The licence is custom, not MIT. In short, you may use, modify and sell it, with two conditions. If you run a model-as-a-service business and its revenue passes $20 million over 12 months, you need a separate agreement with Moonshot. If a product built on it has more than 100 million monthly users or more than $20 million in monthly revenue, you must show the name Kimi K3.1-Pro in the interface. Internal use is exempt. Moonshot also says it removed cyber-related data from training before releasing the weights.
What Kimi is good and bad at: vendor claims against independent evidence
| Question | Evidence | Who measured it |
|---|---|---|
| Is it strong at long coding tasks? | Moonshot reports 65.7% on Terminal-Bench 4.0, ahead of GPT-6 Astra (58.2%) and just behind Claude Opus 5.5 (66.4%). The public tbench.ai board had no Kimi row on 11 Nov | Moonshot, vendor claim |
| Is it the best open-weight model? | Artificial Analysis Intelligence Index: Kimi K3.1 Pro 46, then DeepSeek V4.1 Flash 1017 (Max) 33 and GLM-5.3 (Max) 32 | Independent |
| Is it near the top overall? | Same index: Claude Opus 5.5 (max) 59, GPT-6 Astra (max) 57, Gemini 4 Argon (high) 50, all above Kimi's 46 | Independent |
| How does it build web apps? | Arena WebDev: rank 11, score 1688 from 4,916 votes. Claude Fable 5.2 is first at 1836 | Independent, human votes |
| Is it cheap and fast? | Cost per task $3.68 against $10.04 for Claude Opus 5.5 (max). Time per task 37.2 minutes against 25.0 | Independent |
| Does it hallucinate less? | Moonshot's internal stress test shows claim-level hallucinations falling from 23.4% (K3) to 8.3% (K3.1 Pro). Moonshot says this is not a daily error rate | Moonshot, vendor claim |
Read this as: strong and cheap for an open model, slower, and behind the top closed models on independent boards. Votes measure preference, not truth, and every board changes within days.
Privacy and data location, from Kimi's own policies
- Consumer policy: Beijing Moonshot Technology's policy (updated 5 November 2026, effective 12 November) says it applies to users in mainland China and tells users outside the mainland to read the policy on kimi.ai. It says personal information is stored in the People's Republic of China and not sent abroad without separate consent.
- Training: the same policy says your input and output, once de-identified, can be used to improve its models, and that you can ask customer service to stop using your content for model training.
- API policy: Moonshot AI Pte. Ltd. in Singapore says it stores data on servers in Singapore and may use data to improve its services, including training. Zero Data Retention is offered to enterprise API customers.
- Gap: when we opened kimi.ai's privacy page on 11 November it showed a Chinese-language policy last updated in December 2024, so we could not find a current English policy for users outside China.
Practical rule: do not paste Aadhaar or PAN details, exam IDs, other people's data or your employer's code into a chatbot whose policy you have not read.
Kimi vs ChatGPT, Claude, Gemini and DeepSeek for a student
| Tool | Free plan, read 11 Nov 2026 | Cheapest paid plan | Note |
|---|---|---|---|
| Kimi | $0: 1 agent task at a time, no Kimi Code, limits not published | Plus $19 a month, or $15 billed annually | Prices in USD only |
| ChatGPT | ₹0: unlimited text chats on GPT-6, limited uploads, images, voice, deep research and Codex | Go ₹399 a month, may show ads | Prices in rupees |
| Claude | $0: Sonnet and Haiku, limited usage every 5 hours and every week, no Claude Code | Pro $17 a month billed annually, $20 monthly | Prices in USD |
| Gemini | ₹0: Flash-Lite model, image generation, Deep Research, Gemini Live, Canvas | Google AI Plus ₹399 a month | Prices in rupees |
| DeepSeek | Chat app limits not checked. API deepseek-flash from $0.15 input and $0.60 output per million tokens off-peak | Pay as you go on the API | Its privacy policy (10 Feb 2026) says data is stored in the PRC |
For daily study help, the free plans of ChatGPT, Gemini and Claude are the safer starting point, with prices in rupees for two of them. Try Kimi if you want its agent tasks or a cheap long-context API, and keep sensitive data out. More lists: free AI tools and AI tools for students. To see how agents like Kimi's differ from chatbots, read what an AI agent is.
Frequently asked questions
What is Kimi AI?
Kimi is an AI assistant and model family from Moonshot AI. You can use it on kimi.com, in mobile apps, in the Kimi Work desktop app, in the Kimi Code coding agent and through the Kimi API. Its newest model is Kimi K3.1 Pro, launched on 23 October 2026 with a 1 million token context window.
Is Kimi AI free or paid?
Both. Kimi has a free plan at $0 with 1 agent task at a time and no Kimi Code. Paid plans on its pricing page are Plus at $19 a month, Pro at $39, Max at $99 and Ultra at $199, with discounts for annual billing. The API is separate and pay as you go.
How much does Kimi AI cost?
On 11 November 2026 Kimi's pricing page showed Plus at $19 a month or $15 a month billed annually ($180 a year), up to Ultra at $199 a month. The API charges $2 per million input tokens and $10 per million output tokens for Kimi K3.1 Pro. No rupee prices were shown.
What is the latest Kimi model?
Kimi K3.1 Pro, launched on 23 October 2026, with a faster Highspeed variant. It has 2.8 trillion parameters in total, 104 billion active, and a 1 million token context window. Its predecessor Kimi K3 launched in mid-July 2026 and is still available on the API.
Is Kimi open source?
Its weights are public on Hugging Face under the Kimi K3.1 Pro License, which is a custom licence rather than MIT. You can use and modify it, but large model-as-a-service businesses need a separate agreement and very large products must display the Kimi name.
Is Kimi AI safe to use?
Check the policy for your situation. The consumer policy says data is stored in mainland China and that de-identified content can be used to improve models, with an opt-out through customer service. The API policy says Singapore servers. Avoid pasting personal identifiers, exam details or confidential code into any chatbot.
Is Kimi better than ChatGPT or DeepSeek?
On the Artificial Analysis index Kimi K3.1 Pro scores 46, above DeepSeek V4.1 Flash 1017 at 33 but below Claude Opus 5.5 at 59 and GPT-6 Astra at 57. For free daily study help, ChatGPT, Gemini and Claude show rupee or clear free limits. Kimi's strength is its cheap, open long-context model.
Sources checked
- Kimi plans in USD, Free plan limits, notice about queueing
- Kimi API prices, batch discount, K2 retirement
- Model release months and retirements
- Current and deprecated model list
- K3.1 Pro launch post, vendor benchmark claims, products
- Open weights, parameters, files
- Kimi K3.1 Pro License text
- Independent index, cost and time per task
- Open-weights ranking
- Arena WebDev rank
- Public Terminal-Bench board
- Consumer privacy policy
- API privacy policy, Singapore
- DeepSeek API prices
- DeepSeek privacy policy
Learn this with mentors
AI and machine learning programmes with admissions open: live classes, mentors and project work.









