Updated on
Kimi: reviews and analysis
Moonshot's assistant with a million-token context window.
Affiliate link · no extra cost to you
Kimi now plays in the same league as the big names and has stopped being the cheap option: its plan starts at $19, one dollar under ChatGPT or Claude. Where it does win is the free tier, the best in its class, and the million-token context for document work. What decides against it is a privacy policy that never says which country stores your data.
Best for: People who read and summarize long documents daily and want a free tier that survives real use.
What the internet says
Kimi has stopped being the cheap option and now competes at the top while charging like it: K3 holds up against frontier models and the entry plan costs $19, roughly what ChatGPT or Claude charge. The recurring praise is the very long context and a free plan none of the big three match. The recurring complaint is that Moonshot paused new sign-ups for lack of GPUs, and that its privacy policy never says which country your data ends up in.
What the web repeats in favour
- Frontier-level quality on long research and documents: K3 entered the Frontend Code Arena at number one, and research reports are the most praised feature in app store ratings
- A one-million-token context window, the house specialty since 2023: you can drop a whole dossier on it without splitting it up
- The free plan allows unlimited basic chat and K2.6 consumes no credits, something the big three do not offer on their free tiers
- K3 weights ship open, so anyone with the hardware can self-host and skip Moonshot's cloud entirely
What the web repeats against
- The entry plan costs $19 a month, essentially what the big three charge, so the bargain-price argument no longer applies to the assistant
- The privacy policy allows conversations to be used for training, describes no opt-out toggle, and never names the country where data is stored
- Moonshot paused new sign-ups on 19 July 2026 because demand outran its compute capacity, reopening spots in batches
- K3's long context only unlocks from the $99 plan up, leaving the $19 and $39 tiers with the short window
- Chaining tools reliably as conversations grow is still its weak spot, and the model burns a lot of reasoning even on simple tasks
Sweep sources: Official pricing · Official pricing · Docs · App stores · Communities · Press · Press · Press
Pros / Cons
Pros
- Free plan with uncapped basic chat
- One-million-token context for long documents
- Top-tier research reports
Cons
- Costs almost the same as ChatGPT, Claude or Gemini
- Its policy never states which country stores your data
- The long context only unlocks on the higher plans
TLDR: Kimi is the general-purpose assistant from Moonshot AI, the Chinese lab whose K3 model now competes with the best of OpenAI and Anthropic. The house specialty is long context, a full million tokens, paired with a free plan that allows basic chat with no message cap. The surprise is on the invoice: the subscription starts at $19 a month, roughly what the big three charge, so anyone arriving in search of the cheap Chinese alternative is in for a letdown.
What Kimi is and how it works
Kimi is the conversational assistant built by Moonshot AI, a Beijing company founded in 2023 that made its name exactly where everyone else was weak, swallowing huge documents without losing the thread. That first version handled 128,000 tokens when the rest of the field sat around 8,000, and that head start still shapes the whole product.
The current model is Kimi K3, announced on 16 July 2026. It is a mixture-of-experts architecture with 2.8 trillion parameters and a one-million-token window, and Moonshot did not oversell it at launch: the company admits K3 sits behind the frontier models from Anthropic and OpenAI on its own evaluation suite, while beating everything else it measured. On the public frontend code arena it did take the top spot on launch day.
For the user it is a chat box with the usual parts of a modern assistant: deep research, file reading, vision and an agent mode. Moonshot has added Kimi Work, a desktop app for Mac and Windows that operates on your local files, and it maintains Kimi CLI as the sibling product for people who code in the terminal.
What it’s like to use day to day
Kimi pulls away from the pack on long document work. Handing it a two-hundred-page contract or a full transcript and asking for a synthesis with internal references is the job it was designed for, and app store ratings point the same way: what recurs most among people who score it highly is the quality of its research reports.
The learning curve is the one every assistant has, which is to say none. You type, it answers. The friction shows up elsewhere. First, availability: on 19 July 2026, three days after launching K3, Moonshot suspended new sign-ups because demand outran its compute capacity, and it has been reopening spots in batches since. If you decide to pay, you may have to wait your turn.
The second sits between what the model promises and what your plan delivers. The million-token window, the product’s headline argument, only opens up from the $99 plans upward. On the entry tier you work with a considerably shorter one, and it pays to know that before subscribing on the strength of the headline.
The third is flagged by independent analysis and matches what you read in the forums: chaining tools as a conversation grows is still the weak spot, and the model spends a lot of reasoning even on simple questions. For chat and documents it delivers. For long unsupervised automation, you still have to watch it.
Pricing and plans
Kimi runs on a permanent free plan plus a subscription from $19 a month. The free tier is among the most generous available right now: basic chat with no message cap, on the previous model generation, which also consumes no credits. If your use is asking questions and summarizing text, you can stay there without paying. Agent credits are tightly capped, and the most repeated complaint in store ratings is that this free layer used to be wider.
From there the tiers climb in big steps up to three-figure plans, and what each jump buys is agent credits, parallel tasks and access to the long context. Annual billing lowers the monthly rate.
Here is the number that changes the decision. The big three price their consumer plan at around twenty dollars a month and Kimi starts at nineteen. The gap is one dollar, so Moonshot’s assistant is not the budget pick against ChatGPT, Claude or Gemini, it is another option at the same price. The real saving lives behind two different doors: the free tier, which does beat all three, and the per-use API price. People expecting rock-bottom rates were thinking of earlier Kimi generations, because K3 is the most expensive model any Chinese lab has released to date.
Who it’s for (and who it isn’t)
Kimi fits if your job means reading a lot and summarizing with judgment: document analysis, market research, review of long reports, case file preparation. It also fits if you want a capable assistant for nothing, because its free plan absorbs real usage that other tools cut off after a handful of daily messages.
It does not fit if you handle information subject to jurisdiction requirements. Its privacy policy allows conversation content to be used for training the models, describes no in-product toggle to opt out, and never states which country the data is stored in. For personal use that may not bother you. For client data, medical records or legal material, the gap is enough to rule it out, and the practical reason outweighs any background suspicion: you cannot document an answer to whoever asks for one. Teams able to self-host K3’s open weights on their own hardware solve the problem another way.
It also isn’t the pick if you need the assistant wired into your work suite, or if you plan to delegate long tool chains without checking the output.
Alternatives to Kimi
The same price gets you the usual three, and the comparison is fought hand to hand: ChatGPT on ecosystem and feature breadth, Claude on judgment in long text, and Gemini if you already live inside Google. The detail sits on Kimi alternatives. For assistants that bundle several models under one account there is YesChat, and if your real problem is finding information rather than writing it, the paid search engine Kagi. The full set is in our guide to the best AI productivity tools.
Frequently asked questions
Is Kimi free?
It has a permanent free plan with basic chat and no message cap, which is plenty for asking questions and summarizing. What runs out fast is agent credits. Paid subscriptions start at $19 a month.
Which model does it run right now?
Kimi K3, released on 16 July 2026, with a one-million-token context window. That full window is only on the higher plans rather than the entry tier, which is the small print that catches most people out.
Is it cheaper than ChatGPT or Claude?
Not on the subscription, which starts at $19 against roughly twenty for the big names. It wins on the free plan and on the per-use API price.
Can I give it confidential documents?
Better not to. Moonshot’s policy allows content to be used for training its models, offers no opt-out toggle, and does not specify where the servers sit. If you handle third-party data, that uncertainty weighs more than any technical advantage.
Is this the same as Kimi CLI?
No. Kimi CLI is the terminal agent for coding, aimed at developers. This review covers the general-purpose assistant for chat, documents and research.