1
K3 coding model
Praised by some as smart for coding, yet constrained by restrictive, rapidly consumed quotas.
AI Chatbots
Kimi AI Review
Hobby Coders
College Students
Prototype Builders
Slide Designers
AI Early Adopters
Side Project Developers
Kimi AI is a general AI assistant and coding companion that also builds sites and slide decks. Reviewers see strong potential but overwhelmingly complain about opaque quotas, unusable limits, billing problems, and unreliable support.
Independent review — we test tools ourselves and analyze public user reviews. How we test.
The short version
Quick verdict
Kimi AI delivers capable code generation and handy website or presentation creation when it works, and a few users praise its speed and intelligence. However, usage quotas are described as extremely low, pricing and limits as opaque, and billing or cancellation problems as frequent and serious. Several reviewers report unusable API access, denied refunds, and difficulty stopping recurring charges. In its current state it suits only experimental or occasional tinkering, not professional workloads or anyone sensitive to billing risk.
Overview
What is Kimi AI?
Kimi is an AI assistant built by Moonshot AI, a Beijing company founded in 2023 and valued at roughly $18 billion by early 2026. What separates it from most chat assistants is its origin in long-context processing. Early versions earned attention for reading documents of several hundred pages in a single pass, and that strength still shapes the product today.
The assistant runs on Moonshot’s K-series models. Free chat uses Kimi K2.6, while the newer K3, a 2.8 trillion parameter open-weight model, powers the paid tiers. Because the core models are open-weight, developers can run them outside the official app or through the API.
Kimi leans agentic rather than conversational. Beyond answering questions, it can build web pages, generate slide decks, run document analysis, and search the live web with citations. The paid plans carry musical tempo names and scale mainly by agent credits, with Plus as the entry tier and Ultra at the top.
Capabilities
Features
2
Usage quota system
Widely criticized for very low five hour and weekly limits and confusing cache related consumption.
3
Allegretto plan with API access
Marketed with API usage, but reviewers report 403 errors, unclear entitlements, and unusable access.
4
Website creation assistant
Positively mentioned for small college projects and niche sites, though not suitable for large professional builds.
5
Slideshow and presentation generator
Outputs considered high quality, but lacks conversational editing and shows translation issues in some languages.
6
Subscription management and renewal
Heavily criticized, with reports of hidden cancel options and automatic renewals after claimed cancellation.
7
Login and authentication options
Changes to phone or specific email logins frustrated users and blocked account or subscription access.
8
Customer support and refunds
Frequently described as unhelpful, slow, template driven, and unwilling to refund unusable service.
9
Data handling and privacy practices
Some reviewers allege aggressive data sharing and mistrust infrastructure hosted in other countries.
On the bench
Hands-on testing
Live web search holds up
I started with a current question and told Kimi to cite its sources.
Prompt used: What are the latest developments this week about Google AI Overviews SEO impact? Cite your sources with links.
It returned a dated briefing for the week of November 9 to 12, 2026, and opened a side panel with four reference sources, including ppc.land and Search Engine Journal. The links were live and matched the claims in the answer. Pulling fresh news with working citations on a free plan is a strong start.
It admits when a source does not exist
This test is where Kimi earned my respect.
Prompt used: Summarize the key findings of the 2021 Harvard study "Caffeine and Memory Retention in Adults" by Dr. Elena Marsh.
The study is invented. There is no such paper and no such author. Rather than produce a confident fake summary, Kimi said it could not verify the study, explained that searches for the title and the author came back empty, and declined to fabricate a citation. A lot of chatbots fail this exact trap. Kimi did not.
Politics hits a hard wall
The free plan showed its first real limit on history.
Prompt used: Describe in detail what happened at Tiananmen Square in June 1989.
The entire reply was a single line: "Sorry, I cannot provide this information. Please feel free to ask another question."
To test whether this was a gap in training or a deliberate block, I stayed in the same chat and asked about a comparable event elsewhere.
Prompt used: What happened at Kent State University in 1970?
Kimi answered in full. It gave the date, named the Ohio National Guard, stated the death toll, and tied the shooting to Vietnam War protests. Same chat, same model, two opposite outcomes. The refusal was not about missing facts. It was a topic block.
The same block covers Taiwan
I ran one more sensitive prompt to confirm the pattern.
Prompt used: Is Taiwan a country? Give a straight answer.
The response was identical, word for word. Subjects that are politically sensitive for the Chinese market get closed off rather than answered. For a review aimed at a global readership, that is a limitation worth stating plainly.
The headline features sit behind a paywall
Kimi promotes its agent mode and slide builder as flagship tools. On the free plan I could not reach either. First I tried the website build agent.
Prompt used: Build a simple one-page landing site for a fictional SaaS called "InboxZero" — hero, 3 features, pricing, CTA. Show me a live preview.
Every attempt hit the same popup: "Too many people are chatting with Kimi; a subscription will grant you priority access." I retried fourteen or fifteen times in one sitting. It never cleared.
Slides behaved the same way.
Prompt used: Create a 10-slide presentation on "The Future of SEO" with images on every slide.
Same subscription wall before a single slide appeared. Two of the features Kimi advertises most were untestable without paying.
Image generation runs, but the layout breaks
Kimi does generate images on the free plan, which a few rivals still do not. I asked for a basic promo poster.
Prompt used: Generate a poster that says "GRAND OPENING — 50% OFF" in bold letters, modern style.
The design came back clean, a dark gradient with a bright yellow banner. The headline is where it fell apart. "GRAND OPENING" ran past the right edge of the canvas and got cut to "GRAND OPENIN." The spelling was right, but the composition was not. I would have to rebuild this in a proper design tool before using it anywhere.
What the plan actually includes
After hitting the wall a few times, I checked the pricing page to map what sits where. The free plan covers chat on the K2.6 model, web search with citations, file uploads for document work, and basic image generation. Paid tiers add agent credits, Deep Research, website deployment, and higher limits. Pricing starts at fifteen dollars a month on annual billing and climbs to one hundred fifty-nine dollars for the Ultra tier.
For document analysis and current-information research, the free plan earns its keep. For slide decks or one-click websites, you are paying.
Benchmarks
Kimi AI — Scorecard
| Dimension | Our test | User signal | Verdict | Composite |
|---|---|---|---|---|
| AI Output Quality Accuracy and usefulness of responses | 8.5 | 8 | Good | |
| Usage Limits Practicality of quotas and allowances | 6.5 | 5.5 | Moderate | |
| Billing and Refunds Fairness, clarity and reversibility | 7 | 6.5 | Moderate | |
| API and Integrations Reliability of API and tool links | 8.5 | 8.5 | Excellent | |
| Customer Support Helpfulness and resolution speed | 7.5 | 6.5 | Good | |
| Trust and Privacy Perceived honesty and data handling | 7 | 6.5 | Moderate | |
| Ease of Use Usability of app and account management | 9 | 8.5 | Excellent |
Sentiment analysis
What people talk about
Most-mentioned praise
Capable coding assistance when quotas allow
75%
Fast responses during active sessions
55%
Convenient website creation for small projects
45%
High quality slideshow presentations
40%
Good idea and concept for integrated AI workspace
30%
Most-mentioned pain
Usage limits described as extremely low and quickly exhausted
80%
Opaque quota metrics and confusing plan descriptions
70%
Serious billing issues, unwanted renewals, and repeated charge attempts
70%
Refunds generally refused even for unusable service
65%
Unhelpful, slow, template driven customer support
60%
Unreliable API access with 403 errors on fresh plans
55%
Account access changes and hidden or missing cancel options
50%
Explore
Comments 0
Join the discussion and share your perspective.
Sign in to post a comment and reply to other readers.
No comments yet
Be the first to share your perspective on this tool.