Kimi AI Review (2026): Free Plan, K3 and Pricing Tested

This Kimi AI review covers a chatbot that many Americans found only this summer. In July 2026, its maker released K3, a very large open model. Since then, US searches for Kimi have stayed steady.

However, the product is more than a model. Kimi is also a chat app with research, slides and agent tools. It has a free plan and four paid plans.

Therefore the testing behind this Kimi AI review followed three questions. First, what does the free plan include? Second, what do the paid plans add? Finally, where does your data go?

So, what is Kimi AI? It is an AI assistant from Moonshot AI, a startup in Beijing. You type a request, and it answers, searches the web or completes a task. The site describes it as built for agentic coding and knowledge work.

The name covers two things. First, Kimi is a family of language models, such as K2 and K3. Second, it is the chat app at kimi.com that runs them. Some people also search for it as Kimi.ai, Kimi Moonshot AI or the Kimi AI chatbot.

The app launched in China in October 2023. Back then, its selling point was very long documents. Today, in contrast, the focus is on agents that plan and finish multi-step work.

Kimi suits researchers, students, analysts and developers. For example, it handles long reports, slide decks, spreadsheets and code well. Similarly, people who want many tools in one tab will like it.

On the other hand, it is a weaker fit for confidential work. It also suits casual users poorly if they only need a quick free chat. The sections below explain why.

Kimi works like most chat assistants. First, you open kimi.com or the app. Then you pick a model, choose a tool and type your request. Below are the four parts that shape daily use.

Chat is the core of the product. On 10 October 2026, the model menu listed three options. K3 is the flagship for chat and agent work. K2.8 is a preview model for everyday tasks. Finally, Instant gives quick replies.

In addition, a Thinking effort setting controls how long the model reasons. The plus button adds files, images, plugins, skills and web search. According to Moonshot, K3 also understands images, video and documents.

Deep Research is the strongest tool for knowledge work. According to the official page, each task runs dozens of searches. Then it writes a report with traceable sources. You can also feed it your own files.

Moreover, the output is flexible. Reports can arrive as an interactive web page, a Word file, a slide deck, a spreadsheet or a PDF. The feature is available on the free plan, but only with a monthly allowance. For a search-first rival, see our Perplexity AI review.

Kimi Slides builds a deck from a prompt, a document, an image or a template. You can upload an old PowerPoint, and Kimi copies its colors, fonts and layout. Afterward, you export the deck as PowerPoint or PNG, or you present it online.

Similarly, Docs and Sheets create written reports and spreadsheets. Build makes and deploys small websites. For a tool that only does presentations, our Gamma AI review covers the category leader.

Agents are where Kimi differs from a plain chatbot. An agent plans a task, uses tools and keeps working until it finishes. Swarm goes further, because it splits one job across several agents at once.

In addition, scheduled tasks run on a timer and push results to a dashboard. Kimi Code brings the model into the terminal, VS Code, Cursor, JetBrains and Zed. However, the heavy agent features sit behind paid plans. Our Manus AI review looks at a rival built only around agents.

Is Kimi AI good in daily use? For research and documents, yes. The tools are well organized, and the output formats are practical. In particular, a report that arrives as an editable deck saves real time.

However, three weak spots appeared. First, the free plan is thin for agent work. The comparison table gives it one agent task at a time, two projects and 500MB of storage. Second, the credit system is hard to read. Plans promise more agent credits, 2x, 5x or 10x, yet the page never states the base number.

Third, capacity has been a problem. One third-party pricing guide reports that Moonshot paused new sign-ups in July 2026, when K3 demand outran its hardware. We could not confirm that on an official page. Still, this Kimi AI review treats it as a fair warning.

Every price in this Kimi AI review comes from the official pricing page, read on 10 October 2026. The page showed US dollars. Prices can differ by region, so check your own checkout.

So is Kimi AI free? Yes, partly. The free plan is called Adagio, and it includes chat and a monthly Deep Research allowance. However, it limits agents, projects and storage. Plus adds more agent credits, website deployment, Kimi Code and 20GB of storage. Pro doubles the credits and adds Goal mode. Meanwhile, Max and Ultra unlock conversations of up to 1 million tokens. How much does Kimi AI cost for most people? Probably 19 dollars a month, which is one dollar under ChatGPT Plus. One caveat remains. A third-party guide says the free plan does not include K3 at all. Moonshot does not state that on its pricing page, so test it before you rely on it.

Moonshot AI owns Kimi AI. The legal name in the privacy policy is Beijing Moonshot AI Technology Co., Ltd. The site footer also carries Chinese ICP license numbers, so the service operates under Chinese rules.

So who makes Kimi AI? Yang Zhilin, Zhou Xinyu and Wu Yuxin founded the company in March 2023. All three studied at Tsinghua University. The first Kimi chatbot followed in October 2023.

Is Kimi AI Chinese? Yes, and its backers are too. According to Wikipedia’s entry on Moonshot AI, Alibaba led a 1 billion dollar round in February 2024, and Tencent joined later. The same entry lists a July 2026 round at a 35 billion dollar valuation.

Yes, Kimi AI is legitimate. It is a real product from a heavily funded company. In addition, Moonshot publishes the K3 weights on Hugging Face under its own license.

Is Kimi AI safe, then? For casual questions, mostly. We read the privacy policy that took effect on 31 August 2026. It says Moonshot stores personal information inside mainland China.

However, one clause deserves attention. The policy says Kimi may use your inputs and outputs to optimize its models, after de-identification. You can opt out, but only by contacting customer service. The service also targets adults, and users under 18 need a guardian’s consent.

There is a reputational issue as well. The Wikipedia entry records that Anthropic accused Moonshot in 2026 of using Claude conversations to train its models. In short, keep passwords, health records and client files out of it.

ChatGPT is the obvious comparison. It has a larger set of integrations, and its paid plan costs about the same. In addition, OpenAI stores data outside China.

Is Kimi AI better than ChatGPT? For exporting research into slides and spreadsheets, it can be. However, choose ChatGPT for polished English writing and for work you must trust.

Claude is the rival that Kimi users mention most. Both products target coding and long documents. As a result, people often test the same task on each.

Claude offers clearer privacy terms for US users. Kimi, in contrast, gives you open weights and more built-in office tools. So pick Claude for sensitive code and Kimi for experiments.

DeepSeek is the closest Chinese rival. It is free, and it is strong at code and math. In US search interest, DeepSeek is far ahead of Kimi.

In contrast, Kimi offers more tools in one place, such as slides and agents. Alibaba’s free assistant is a third option, and our Qwen AI review covers it.

Pros

• Free plan with chat and a monthly Deep Research allowance
• Deep Research exports to web page, Word, slides, spreadsheet and PDF
• Kimi Slides exports editable PowerPoint files
• K3 is an open-weights model with a context window of up to 1 million tokens
• Apps for web, iOS, Android, HarmonyOS, macOS and Windows
• Kimi Code works in the terminal, VS Code, Cursor, JetBrains and Zed

Cons

• Personal information is stored in mainland China
• Policy allows model training on de-identified content, with opt-out only through support
• Agent credits are described as multipliers, with no published base number
• Free plan is limited to one agent task, two projects and 500MB
• The 1 million token chat needs the Max or Ultra plan
• A third-party guide reports paused sign-ups during peak demand in July 2026

Is Kimi AI worth it? For research and document work, yes. Few rivals turn one prompt into a cited report, a deck and a spreadsheet. For students and analysts, the free plan is an easy test.

However, here is the short answer from this Kimi AI review. Start on the free plan, and move to Plus only if you hit the agent limits. Keep sensitive material elsewhere, because the policy stores data in China and allows training on it.

Ultimately, the open model and the paid app are two different deals. The weights cost nothing, yet running 2.8 trillion parameters at home is out of reach for most people. So you rent the convenience instead. If that trade suits you, Kimi is one of the most capable assistants to come out of China.

For everyday chat and writing, we compare the leading assistants in the best AI writing tools collection. If you want another free Chinese assistant, the Qwen AI review tests Alibaba’s app. Finally, the Perplexity AI review looks at a search-first alternative.