What is Kimi?
Kimi is an AI assistant built by Moonshot AI, a Chinese lab, first known for reading very long documents in a single conversation and more recently for its open-weight Kimi K2 model, a large mixture-of-experts model tuned for agentic tasks and coding.
You use Kimi through its web and mobile apps, through an OpenAI-compatible API, or, for K2, by self-hosting the published open weights. It handles long files, web search, and tool use, and is strong in both Chinese and English.
Kimi is worth evaluating when you need to reason over long inputs, want capable open weights you can host, or are comparing low-cost frontier-class models, while accounting for the data-governance questions of a China-based hosted service.
What it's best for
- Reading and reasoning over very long documents, transcripts, and codebases in one pass.
- Agentic and coding work through the open-weight Kimi K2 model, which is tuned for multi-step tool use.
- Self-hosting: K2's weights are openly published, so you can run inference in your own environment.
- Bilingual Chinese and English tasks where strong coverage of both matters.
- Cost-sensitive evaluation of a capable model, via a low-cost API or your own compute.
Where it falls short
- Sensitive or regulated data on the hosted app and API, which run on China-based infrastructure. Self-hosting the open weights avoids this.
- Topics subject to Chinese content restrictions, which the hosted model may avoid or steer.
- Native image, audio, or video generation. Pair it with a dedicated generation model.
- Teams needing the mature enterprise support and SLAs of the large US providers.
Getting started
Sign up at kimi.com (or the mobile app) and you can chat for free. Upload long PDFs, spreadsheets, or code files and ask questions across the whole document set.
Turn on web search when you need current information; without it, Kimi answers from training data.
Kimi K2 and the API
For building, the Moonshot API is OpenAI-compatible, so most SDKs work by changing the base URL and key. Kimi K2 is the model to reach for on agentic and coding workloads.
K2's open weights are published on hubs like Hugging Face and run under common runtimes, so a data-sensitive team can self-host rather than call the hosted service.
Getting better answers
Give Kimi the full document and a precise question rather than a summary; its strength is keeping long context in view.
For agentic runs with K2, describe the tools and the goal clearly and let it plan the steps. Check the data-governance path before sending confidential inputs to the hosted API.
What Kimi costs
Approximate, in USD, as of January 2026. Prices change often. Confirm on the official site before you rely on them.
Web and mobile app
$0
Free assistant with web search and long-document upload, subject to limits.
Open weights (Kimi K2)
$0 (self-host)
Download and run K2 yourself; you pay only for your own compute.
API
Low, usage-based
OpenAI-compatible, priced per million tokens by model.
Example prompts
Copy these into Kimi as starting points, then adapt them to your task.
Here is a long report. Summarize it in one paragraph, list the five key findings with the page they come from, and flag any claim the document does not support.
Using Kimi K2, plan and implement this feature step by step. List the files you would change, make the edits, and explain each change. Flag anything you are unsure about.
Rewrite this OpenAI API call to use the Moonshot (Kimi) API instead, changing only the base URL and model name and keeping everything else the same.
We are considering Kimi for an internal tool that touches customer data. List the data-governance questions to answer before using the hosted API, and what changes if we self-host the K2 open weights.
Kimi
common questions.
Direct answers to the questions we get asked the most. If yours isn't covered, write to the team.