Kimi
Moonshot AI's assistant, now on Kimi K3 — a 2.8-trillion-parameter open-weight multimodal reasoning model with a 1M-token context window, priced at Sonnet-class rates.
About Kimi
Kimi built its reputation on reading things nothing else could — entire books, whole codebases, hundreds of pages of filings — and Kimi K3 extends that with a 1M-token context and native vision. Kimi is free to use in the app and, unusually for a model this size, the weights are public.
What is Kimi?
Kimi is the assistant from Moonshot AI. Its flagship, Kimi K3, launched on 16 July 2026: a 2.8-trillion-parameter mixture-of-experts model with a 1M-token context window, native visual understanding and a new architecture pairing Kimi Delta Attention — a hybrid linear attention scheme — with Attention Residuals. Two variants shipped at launch: K3 Max for chat and agents, and K3 Swarm Max for large-scale parallel work. The full weights landed on Hugging Face on 27 July 2026, making K3 one of the largest openly released models in existence. It went live in Kimi's own apps and API on the same day it was announced.
Key Features
- 1M-token context: Load entire books, codebases or document sets into a single conversation.
- 2.8T-parameter MoE: Frontier-scale capacity with sparse activation for tractable inference cost.
- Open weights: Full K3 weights published on Hugging Face — rare at this scale.
- Native multimodality: Visual understanding built into the model rather than bolted on.
- K3 Swarm Max: A variant purpose-built for running many parallel agent tasks at once.
Who Should Use Kimi?
Kimi is for people whose inputs are enormous — legal discovery, financial filings, academic literature, large monorepos. The free app is a genuinely capable everyday assistant, and the 1M-token window is the headline reason to choose it over rivals. Developers running large parallel agent workloads should look at K3 Swarm Max specifically. Self-hosting is theoretically possible thanks to the open weights, but at 2.8 trillion parameters it is realistic only for organisations with serious GPU capacity. As with other Chinese-hosted services, data residency rules it out for some buyers.
Pricing
The Kimi chat app and web interface are free with daily usage limits. The official API rate is $3.00 per million input tokens and $15.00 per million output tokens, with a $0.30 per million cache-hit input rate — the same tier as Anthropic's Claude Sonnet series, which makes Kimi a quality play rather than a discount play on the API. The full K3 weights are available free on Hugging Face for anyone with the hardware to run them.
Pros and Cons
| Pros | Cons |
|---|---|
| 1M-token context handles book-length inputs comfortably | API priced at Sonnet rates rather than undercutting them |
| Open weights for a 2.8T-parameter frontier model | Self-hosting needs serious GPU capacity |
| Native multimodal visual understanding | Hosted service processes data in China |
| Free consumer app with real capability | English-language tooling and docs still maturing |
Bottom Line
Kimi K3 is one of the most technically interesting releases of 2026 — frontier scale, genuinely open weights, and a context window that changes what you can do in a single prompt. Use the free app for long-document work where ChatGPT and Claude run out of room. On the API it is priced as a peer to Claude Sonnet rather than a bargain, so choose it for the context length and multimodality rather than for the cost.
Pros & Cons
Pros
- 1M-token context handles book-length inputs
- Open weights for a 2.8T-parameter frontier model
- Native multimodal visual understanding
- Free consumer app with generous limits
Cons
- API pricing matches Claude Sonnet rather than undercutting it
- 2.8T parameters make self-hosting impractical for most
- Hosted service processes data in China
- English-language ecosystem still developing
Use Cases
Tags
Company Info
- Company
- Moonshot AI
- Founded
- 2023~
- HQ
- Beijing, China~
- Pricing
- freemium
- Last verified
- 2026-08-29
~ Approximate. Verify at the official website.
Promote Your AI Tool
Reach a targeted audience of developers, creators, and businesses actively searching for AI tools.
View Ad Packages →Frequently Asked Questions
Is Kimi free?▾
Kimi offers a free plan with limited features. Paid plans unlock additional capabilities. Free chat app and web interface with daily limits. API: $3.00 per million input tokens, $15.00 per million output, $0.30 per million on cache hits. Full K3 weights available free on Hugging Face for self-hosting.
What is Kimi used for?▾
Moonshot AI's assistant, now on Kimi K3 — a 2.8-trillion-parameter open-weight multimodal reasoning model with a 1M-token context window, priced at Sonnet-class rates. Key use cases include: Very long document analysis, Research across large source sets, Multimodal reasoning over images and text.
What are the pros and cons of Kimi?▾
Pros: 1M-token context handles book-length inputs; Open weights for a 2.8T-parameter frontier model; Native multimodal visual understanding. Cons: API pricing matches Claude Sonnet rather than undercutting it; 2.8T parameters make self-hosting impractical for most.
Who makes Kimi?▾
Kimi is developed by Moonshot AI, founded in 2023.
What are the best alternatives to Kimi?▾
Top alternatives to Kimi include DeepSeek, ChatGPT, Claude. You can compare them all on AIFans.
Similar Tools
View allChinese AI lab whose DeepSeek-V4 models deliver near-frontier reasoning at a fraction of Western prices. Free chat app, 1M-token context, open weights and off-peak API discounts.
OpenAI's AI assistant, now running the GPT-5.6 family — Sol for frontier reasoning, Terra for everyday work and Luna for cheap high-volume tasks. Writing, coding, research, vision, voice and agents in one place.
Anthropic's AI assistant built on the Claude 5 family — Fable 5, Opus 5 and Sonnet 5, plus Haiku 4.5. Best-in-class at long-form reasoning, document analysis and agentic coding, with Claude Code included on every paid plan.
Google's multimodal AI assistant. Gemini 3.1 Pro handles deep reasoning while the fast Gemini 3.6 and 3.7 Flash models power everyday tasks, wired into Search, Workspace, Android and Chrome.